AbhishekGhosh1998 commited on
Commit
686e979
·
verified ·
1 Parent(s): 026f6fd

granularity results tiera-rerun-fixed-moonep @ 2026-08-30T07:10:44Z from sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0

Browse files
Files changed (32) hide show
  1. tiera-rerun-fixed-moonep/hf_watcher.log +3 -0
  2. tiera-rerun-fixed-moonep/study.log +36 -0
  3. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1/mem_samples.csv +224 -0
  4. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1/run.env +70 -0
  5. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1/train.log +1057 -0
  6. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2/mem_samples.csv +232 -0
  7. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2/run.env +70 -0
  8. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2/train.log +1057 -0
  9. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3/mem_samples.csv +232 -0
  10. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3/run.env +70 -0
  11. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3/train.log +1057 -0
  12. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1/mem_samples.csv +224 -0
  13. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1/run.env +70 -0
  14. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1/train.log +1170 -0
  15. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2/mem_samples.csv +224 -0
  16. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2/run.env +70 -0
  17. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2/train.log +1171 -0
  18. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3/mem_samples.csv +224 -0
  19. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3/run.env +70 -0
  20. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3/train.log +1170 -0
  21. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1/mem_samples.csv +232 -0
  22. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1/run.env +70 -0
  23. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1/train.log +1308 -0
  24. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv +224 -0
  25. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2/run.env +70 -0
  26. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2/train.log +1310 -0
  27. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv +40 -0
  28. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep3/run.env +70 -0
  29. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep3/train.log +576 -0
  30. tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/results.tsv +9 -0
  31. tiera-rerun-fixed-moonep/ultraep/a1/manifest.txt +19 -0
  32. tiera-rerun-fixed-moonep/ultraep/a1/pip_freeze.txt +317 -0
tiera-rerun-fixed-moonep/hf_watcher.log CHANGED
@@ -35,3 +35,6 @@ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
35
  uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
36
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
37
  [2026-08-30T07:00:41+00:00] backup pass
 
 
 
 
35
  uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
36
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
37
  [2026-08-30T07:00:41+00:00] backup pass
38
+ uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
39
+ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
40
+ [2026-08-30T07:10:43+00:00] backup pass
tiera-rerun-fixed-moonep/study.log CHANGED
@@ -623,3 +623,39 @@ Already logged in to Hugging Face as: AbhishekGhosh1998
623
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
624
 
625
  === STEP ultraep__a1__gbs16
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
623
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
624
 
625
  === STEP ultraep__a1__gbs16
626
+ ==============================================================
627
+ CAMPAIGN: ultraep backend=ultraep phase=a1
628
+ Megatron: /workspace/Megatron-UltraEP @ ultraep-bracing-regions-through-kernels
629
+ GBS=16 seq=4096 repeats=3 nsys=0
630
+ results -> /workspace/tiera-rerun-fixed-moonep/ultraep/a1
631
+ ==============================================================
632
+ --------------------------------------------------------------------------
633
+ [07:00:42] CELL G1__ultraep__one__seq4096__gbs16__cap2p40625 (slots=1)
634
+ [guard] iter 5/12 (60s)
635
+ [guard] iter 12/12 (70s)
636
+ [guard] tearing down process group (outcome=DONE)
637
+ [guard] iter 5/12 (60s)
638
+ [guard] iter 12/12 (70s)
639
+ [guard] tearing down process group (outcome=DONE)
640
+ [guard] iter 6/12 (60s)
641
+ [guard] iter 12/12 (70s)
642
+ [guard] tearing down process group (outcome=DONE)
643
+ --------------------------------------------------------------------------
644
+ [07:04:25] CELL G2__ultraep__one__seq4096__gbs16__cap1p08203125 (slots=1)
645
+ [guard] iter 8/12 (60s)
646
+ [guard] iter 12/12 (70s)
647
+ [guard] tearing down process group (outcome=DONE)
648
+ [guard] iter 8/12 (60s)
649
+ [guard] iter 12/12 (70s)
650
+ [guard] tearing down process group (outcome=DONE)
651
+ [guard] iter 8/12 (60s)
652
+ [guard] iter 12/12 (70s)
653
+ [guard] tearing down process group (outcome=DONE)
654
+ --------------------------------------------------------------------------
655
+ [07:08:01] CELL G3__ultraep__one__seq4096__gbs16__cap1p072265625 (slots=1)
656
+ [guard] iter 6/12 (60s)
657
+ [guard] iter 12/12 (70s)
658
+ [guard] tearing down process group (outcome=DONE)
659
+ [guard] iter 7/12 (60s)
660
+ [guard] iter 12/12 (70s)
661
+ [guard] tearing down process group (outcome=DONE)
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1/mem_samples.csv ADDED
@@ -0,0 +1,224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 1
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 2
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 4
42
+ 1, 629
43
+ 2, 47
44
+ 3, 629
45
+ 4, 45
46
+ 5, 629
47
+ 6, 629
48
+ 7, 629
49
+ 0, 919
50
+ 1, 907
51
+ 2, 923
52
+ 3, 903
53
+ 4, 923
54
+ 5, 911
55
+ 6, 923
56
+ 7, 917
57
+ 0, 1895
58
+ 1, 1895
59
+ 2, 1895
60
+ 3, 1895
61
+ 4, 1895
62
+ 5, 1895
63
+ 6, 1895
64
+ 7, 1895
65
+ 0, 2419
66
+ 1, 2405
67
+ 2, 2423
68
+ 3, 2429
69
+ 4, 2429
70
+ 5, 2413
71
+ 6, 2421
72
+ 7, 2449
73
+ 0, 42469
74
+ 1, 42467
75
+ 2, 42467
76
+ 3, 42471
77
+ 4, 42467
78
+ 5, 42467
79
+ 6, 42467
80
+ 7, 42467
81
+ 0, 43427
82
+ 1, 43409
83
+ 2, 43441
84
+ 3, 43419
85
+ 4, 43433
86
+ 5, 43429
87
+ 6, 43455
88
+ 7, 43453
89
+ 0, 44167
90
+ 1, 44165
91
+ 2, 44165
92
+ 3, 44165
93
+ 4, 44165
94
+ 5, 44165
95
+ 6, 44165
96
+ 7, 44165
97
+ 0, 44167
98
+ 1, 44165
99
+ 2, 44165
100
+ 3, 44165
101
+ 4, 44165
102
+ 5, 44165
103
+ 6, 44165
104
+ 7, 44165
105
+ 0, 45939
106
+ 1, 45937
107
+ 2, 45937
108
+ 3, 45937
109
+ 4, 45937
110
+ 5, 45937
111
+ 6, 45937
112
+ 7, 45937
113
+ 0, 80375
114
+ 1, 80373
115
+ 2, 80373
116
+ 3, 80373
117
+ 4, 80373
118
+ 5, 80373
119
+ 6, 80373
120
+ 7, 80373
121
+ 0, 81289
122
+ 1, 81287
123
+ 2, 81287
124
+ 3, 81287
125
+ 4, 81287
126
+ 5, 81287
127
+ 6, 81287
128
+ 7, 81287
129
+ 0, 82367
130
+ 1, 82365
131
+ 2, 82365
132
+ 3, 82365
133
+ 4, 82365
134
+ 5, 82365
135
+ 6, 82365
136
+ 7, 82365
137
+ 0, 93565
138
+ 1, 93499
139
+ 2, 93507
140
+ 3, 93529
141
+ 4, 93585
142
+ 5, 93523
143
+ 6, 93547
144
+ 7, 93543
145
+ 0, 95707
146
+ 1, 95639
147
+ 2, 95669
148
+ 3, 95643
149
+ 4, 95711
150
+ 5, 95637
151
+ 6, 95667
152
+ 7, 95677
153
+ 0, 96821
154
+ 1, 96757
155
+ 2, 96829
156
+ 3, 96795
157
+ 4, 96933
158
+ 5, 96813
159
+ 6, 96849
160
+ 7, 96889
161
+ 0, 102823
162
+ 1, 120031
163
+ 2, 119303
164
+ 3, 120087
165
+ 4, 83117
166
+ 5, 119527
167
+ 6, 120087
168
+ 7, 120031
169
+ 0, 129221
170
+ 1, 129109
171
+ 2, 129109
172
+ 3, 129109
173
+ 4, 129475
174
+ 5, 129109
175
+ 6, 129109
176
+ 7, 129109
177
+ 0, 129221
178
+ 1, 129109
179
+ 2, 129109
180
+ 3, 129109
181
+ 4, 129475
182
+ 5, 129109
183
+ 6, 129109
184
+ 7, 129109
185
+ 0, 129221
186
+ 1, 129109
187
+ 2, 129109
188
+ 3, 129109
189
+ 4, 129475
190
+ 5, 129109
191
+ 6, 129109
192
+ 7, 129109
193
+ 0, 129221
194
+ 1, 129109
195
+ 2, 129109
196
+ 3, 129109
197
+ 4, 129475
198
+ 5, 129109
199
+ 6, 129109
200
+ 7, 129109
201
+ 0, 128313
202
+ 1, 128201
203
+ 2, 128201
204
+ 3, 128201
205
+ 4, 128567
206
+ 5, 128201
207
+ 6, 128201
208
+ 7, 128201
209
+ 0, 124065
210
+ 1, 123953
211
+ 2, 123953
212
+ 3, 123953
213
+ 4, 124319
214
+ 5, 123953
215
+ 6, 123953
216
+ 7, 123953
217
+ 0, 641
218
+ 1, 4
219
+ 2, 4
220
+ 3, 4
221
+ 4, 4
222
+ 5, 4
223
+ 6, 132
224
+ 7, 123953
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G1
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=14336
11
+ num_experts=8
12
+ topk=2
13
+ experts_per_rank=1
14
+ expert_params=176160768
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=8192
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=2.40625
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1/train.log ADDED
@@ -0,0 +1,1057 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G1 / ultraep (slots/rank=1 policy=one)
3
+ E=8 K=2 H'=14336 m=1
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=2.40625 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:00:44.146000 693438 torch/distributed/run.py:766]
10
+ W0830 07:00:44.146000 693438 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:00:44.146000 693438 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:00:44.146000 693438 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
18
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
19
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
20
+ Number of virtual stages per pipeline stage: None
21
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
22
+ using torch.bfloat16 for parameters ...
23
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
24
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
25
+ ------------------------ arguments ------------------------
26
+ account_for_embedding_in_pipeline_split ......... False
27
+ account_for_loss_in_pipeline_split .............. False
28
+ accumulate_allreduce_grads_in_fp32 .............. True
29
+ adam_beta1 ...................................... 0.9
30
+ adam_beta2 ...................................... 0.999
31
+ adam_eps ........................................ 1e-08
32
+ add_bias_linear ................................. False
33
+ add_position_embedding .......................... False
34
+ add_qkv_bias .................................... False
35
+ adlr_autoresume ................................. False
36
+ adlr_autoresume_interval ........................ 1000
37
+ align_grad_reduce ............................... True
38
+ align_param_gather .............................. False
39
+ app_tag_run_name ................................ None
40
+ app_tag_run_version ............................. 0.0.0
41
+ apply_layernorm_1p .............................. False
42
+ apply_query_key_layer_scaling ................... False
43
+ apply_residual_connection_post_layernorm ........ False
44
+ apply_rope_fusion ............................... True
45
+ async_save ...................................... None
46
+ async_tensor_model_parallel_allreduce ........... True
47
+ attention_backend ............................... AttnBackend.auto
48
+ attention_dropout ............................... 0.0
49
+ attention_softmax_in_fp32 ....................... False
50
+ auto_detect_ckpt_format ......................... False
51
+ barrier_with_L1_time ............................ True
52
+ bert_binary_head ................................ True
53
+ bert_embedder_type .............................. megatron
54
+ bert_load ....................................... None
55
+ bf16 ............................................ True
56
+ bias_dropout_fusion ............................. True
57
+ bias_gelu_fusion ................................ False
58
+ bias_swiglu_fusion .............................. True
59
+ biencoder_projection_dim ........................ 0
60
+ biencoder_shared_query_context_model ............ False
61
+ block_data_path ................................. None
62
+ calc_ft_timeouts ................................ False
63
+ calculate_per_token_loss ........................ False
64
+ check_for_large_grads ........................... False
65
+ check_for_nan_in_loss_and_grad .................. False
66
+ check_for_spiky_loss ............................ False
67
+ check_weight_hash_across_dp_replicas_interval ... None
68
+ ckpt_assume_constant_structure .................. False
69
+ ckpt_convert_format ............................. None
70
+ ckpt_convert_save ............................... None
71
+ ckpt_convert_update_legacy_dist_opt_format ...... False
72
+ ckpt_format ..................................... torch_dist
73
+ ckpt_fully_parallel_load ........................ False
74
+ ckpt_fully_parallel_save ........................ True
75
+ ckpt_fully_parallel_save_deprecated ............. False
76
+ ckpt_step ....................................... None
77
+ classes_fraction ................................ 1.0
78
+ clip_grad ....................................... 1.0
79
+ clone_scatter_output_in_embedding ............... True
80
+ config_logger_dir ...............................
81
+ consumed_train_samples .......................... 0
82
+ consumed_valid_samples .......................... 0
83
+ context_parallel_size ........................... 1
84
+ cp_comm_type .................................... ['p2p']
85
+ create_attention_mask_in_dataloader ............. True
86
+ cross_entropy_fusion_impl ....................... native
87
+ cross_entropy_loss_fusion ....................... False
88
+ cuda_graph_scope ................................ full_iteration
89
+ cuda_graph_warmup_steps ......................... 3
90
+ data_args_path .................................. None
91
+ data_cache_path ................................. None
92
+ data_parallel_random_init ....................... False
93
+ data_parallel_sharding_strategy ................. no_shard
94
+ data_parallel_size .............................. 8
95
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
96
+ data_per_class_fraction ......................... 1.0
97
+ data_sharding ................................... True
98
+ dataloader_type ................................. single
99
+ ddp_average_in_collective ....................... False
100
+ ddp_bucket_size ................................. None
101
+ ddp_num_buckets ................................. None
102
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
103
+ decoder_first_pipeline_num_layers ............... None
104
+ decoder_last_pipeline_num_layers ................ None
105
+ decoder_num_layers .............................. None
106
+ decoder_seq_length .............................. None
107
+ decoupled_lr .................................... None
108
+ decoupled_min_lr ................................ None
109
+ decrease_batch_size_if_needed ................... False
110
+ defer_embedding_wgrad_compute ................... False
111
+ delay_wgrad_compute ............................. False
112
+ deprecated_use_mcore_models ..................... True
113
+ deterministic_mode .............................. False
114
+ dino_bottleneck_size ............................ 256
115
+ dino_freeze_last_layer .......................... 1
116
+ dino_head_hidden_size ........................... 2048
117
+ dino_local_crops_number ......................... 10
118
+ dino_local_img_size ............................. 96
119
+ dino_norm_last_layer ............................ False
120
+ dino_teacher_temp ............................... 0.07
121
+ dino_warmup_teacher_temp ........................ 0.04
122
+ dino_warmup_teacher_temp_epochs ................. 30
123
+ disable_bf16_reduced_precision_matmul ........... False
124
+ disable_mamba_mem_eff_path ...................... False
125
+ disable_straggler_on_startup .................... False
126
+ dist_ckpt_format_deprecated ..................... None
127
+ dist_ckpt_strictness ............................ assume_ok_unexpected
128
+ distribute_saved_activations .................... False
129
+ distributed_backend ............................. nccl
130
+ distributed_timeout_minutes ..................... 10
131
+ embedding_init_method_std ....................... None
132
+ embedding_path .................................. None
133
+ empty_unused_memory_level ....................... 0
134
+ enable_cuda_graph ............................... True
135
+ enable_experimental ............................. False
136
+ enable_ft_package ............................... False
137
+ enable_gloo_process_groups ...................... True
138
+ enable_msc ...................................... True
139
+ enable_one_logger ............................... True
140
+ encoder_num_layers .............................. 16
141
+ encoder_seq_length .............................. 4096
142
+ end_weight_decay ................................ 0.1
143
+ eod_mask_loss ................................... False
144
+ error_injection_rate ............................ 0
145
+ error_injection_type ............................ transient_error
146
+ eval_interval ................................... 1000000
147
+ eval_iters ...................................... 0
148
+ evidence_data_path .............................. None
149
+ exit_duration_in_mins ........................... None
150
+ exit_interval ................................... None
151
+ exit_on_missing_checkpoint ...................... False
152
+ exit_signal_handler ............................. False
153
+ exp_avg_dtype ................................... torch.float32
154
+ exp_avg_sq_dtype ................................ torch.float32
155
+ expert_model_parallel_size ...................... 8
156
+ expert_tensor_parallel_size ..................... 1
157
+ export_force_local_attention .................... False
158
+ export_kd_cfg ................................... None
159
+ export_kd_teacher_ckpt_format ................... None
160
+ export_kd_teacher_load .......................... None
161
+ export_kv_cache_quant ........................... False
162
+ export_legacy_megatron .......................... False
163
+ export_model_type ............................... GPTModel
164
+ export_moe_apply_probs_on_input ................. False
165
+ export_qk_l2_norm ............................... False
166
+ export_quant_cfg ................................ None
167
+ export_real_quant_cfg ........................... None
168
+ export_te_mcore_model ........................... False
169
+ external_cuda_graph ............................. False
170
+ ffn_hidden_size ................................. 14336
171
+ finetune ........................................ False
172
+ finetune_data_split ............................. train
173
+ finetune_hf_dataset ............................. None
174
+ first_last_layers_bf16 .......................... False
175
+ flash_decode .................................... False
176
+ fp16 ............................................ False
177
+ fp16_lm_cross_entropy ........................... False
178
+ fp32_residual_connection ........................ False
179
+ fp8 ............................................. e4m3
180
+ fp8_amax_compute_algo ........................... most_recent
181
+ fp8_amax_history_len ............................ 1
182
+ fp8_interval .................................... 1
183
+ fp8_margin ...................................... 0
184
+ fp8_param_gather ................................ False
185
+ fp8_recipe ...................................... mxfp8
186
+ fp8_wgrad ....................................... True
187
+ fsdp_double_buffer .............................. False
188
+ global_batch_size ............................... 16
189
+ grad_reduce_in_bf16 ............................. False
190
+ gradient_accumulation_fusion .................... True
191
+ gradient_reduce_div_fusion ...................... True
192
+ group_query_attention ........................... True
193
+ head_lr_mult .................................... 1.0
194
+ heterogeneous_layers_config_encoded_json ........ None
195
+ heterogeneous_layers_config_path ................ None
196
+ hidden_dropout .................................. 0.0
197
+ hidden_size ..................................... 4096
198
+ hierarchical_context_parallel_sizes ............. None
199
+ high_priority_stream_groups ..................... []
200
+ hybrid_attention_ratio .......................... 0.0
201
+ hybrid_mlp_ratio ................................ 0.0
202
+ hybrid_override_pattern ......................... None
203
+ hysteresis ...................................... 2
204
+ ict_head_size ................................... None
205
+ ict_load ........................................ None
206
+ img_h ........................................... 224
207
+ img_w ........................................... 224
208
+ indexer_batch_size .............................. 128
209
+ indexer_log_interval ............................ 1000
210
+ inference_batch_times_seqlen_threshold .......... -1
211
+ inference_dynamic_batching ...................... False
212
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
213
+ inference_dynamic_batching_buffer_overflow_factor None
214
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
215
+ inference_dynamic_batching_chunk_size ........... 256
216
+ inference_dynamic_batching_max_requests_override None
217
+ inference_dynamic_batching_max_tokens_override .. None
218
+ inference_dynamic_batching_num_cuda_graphs ...... 16
219
+ inference_max_batch_size ........................ 8
220
+ inference_max_seq_length ........................ 2560
221
+ inference_rng_tracker ........................... False
222
+ init_method_std ................................. 0.01
223
+ init_method_xavier_uniform ...................... False
224
+ init_model_with_meta_device ..................... False
225
+ initial_loss_scale .............................. 4294967296
226
+ inprocess_active_world_size ..................... 8
227
+ inprocess_barrier_timeout ....................... 120
228
+ inprocess_completion_timeout .................... 120
229
+ inprocess_empty_cuda_cache ...................... False
230
+ inprocess_granularity ........................... node
231
+ inprocess_hard_timeout .......................... 90
232
+ inprocess_heartbeat_interval .................... 30
233
+ inprocess_heartbeat_timeout ..................... 60
234
+ inprocess_last_call_wait ........................ 1
235
+ inprocess_max_iterations ........................ None
236
+ inprocess_monitor_process_interval .............. 1.0
237
+ inprocess_monitor_thread_interval ............... 1.0
238
+ inprocess_progress_watchdog_interval ............ 1.0
239
+ inprocess_restart ............................... False
240
+ inprocess_soft_timeout .......................... 60
241
+ inprocess_termination_grace_time ................ 1
242
+ is_hybrid_model ................................. False
243
+ iter_per_epoch .................................. 1250
244
+ iterations_to_skip .............................. []
245
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
246
+ kitchen_config_file ............................. None
247
+ kitchen_recipe_number ........................... None
248
+ kv_channels ..................................... 128
249
+ kv_lora_rank .................................... 32
250
+ lazy_mpu_init ................................... None
251
+ load ............................................ None
252
+ load_main_params_from_ckpt ...................... None
253
+ load_model_opt_format ........................... False
254
+ local_rank ...................................... 0
255
+ log_energy ...................................... False
256
+ log_interval .................................... 1
257
+ log_loss_scale_to_tensorboard ................... True
258
+ log_memory_to_tensorboard ....................... False
259
+ log_num_zeros_in_grad ........................... False
260
+ log_params_norm ................................. False
261
+ log_progress .................................... False
262
+ log_straggler ................................... False
263
+ log_throughput .................................. True
264
+ log_timers_to_tensorboard ....................... False
265
+ log_validation_ppl_to_tensorboard ............... False
266
+ log_world_size_to_tensorboard ................... False
267
+ logging_level ................................... None
268
+ loss_scale ...................................... None
269
+ loss_scale_window ............................... 1000
270
+ lr .............................................. 0.0001
271
+ lr_decay_iters .................................. 8
272
+ lr_decay_samples ................................ None
273
+ lr_decay_style .................................. cosine
274
+ lr_warmup_fraction .............................. None
275
+ lr_warmup_init .................................. 0.0
276
+ lr_warmup_iters ................................. 5
277
+ lr_warmup_samples ............................... 0
278
+ lr_wsd_decay_iters .............................. None
279
+ lr_wsd_decay_samples ............................ None
280
+ lr_wsd_decay_style .............................. exponential
281
+ main_grads_dtype ................................ torch.float32
282
+ main_params_dtype ............................... torch.float32
283
+ make_vocab_size_divisible_by .................... 128
284
+ mamba_head_dim .................................. 64
285
+ mamba_num_groups ................................ 8
286
+ mamba_num_heads ................................. None
287
+ mamba_state_dim ................................. 128
288
+ manual_gc ....................................... False
289
+ manual_gc_eval .................................. True
290
+ manual_gc_interval .............................. 0
291
+ mask_factor ..................................... 1.0
292
+ mask_prob ....................................... 0.15
293
+ mask_type ....................................... random
294
+ masked_softmax_fusion ........................... False
295
+ max_position_embeddings ......................... 32768
296
+ max_tokens_to_oom ............................... 12000
297
+ memory_snapshot_path ............................ snapshot.pickle
298
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
299
+ micro_batch_size ................................ 1
300
+ microbatch_group_size_per_vp_stage .............. None
301
+ mid_level_dataset_surplus ....................... 0.005
302
+ min_loss_scale .................................. 1.0
303
+ min_lr .......................................... 1e-05
304
+ mlp_chunks_for_prefill .......................... 1
305
+ mmap_bin_files .................................. True
306
+ mock_data ....................................... False
307
+ moe_apply_probs_on_input ........................ False
308
+ moe_aux_loss_coeff .............................. 0.01
309
+ moe_capacity_monitor ............................ True
310
+ moe_deepep_num_sms .............................. 20
311
+ moe_echo_enable_random_offloading ............... False
312
+ moe_echo_expert_dispatch_overlap ................ False
313
+ moe_echo_expert_dispatcher_type ................. hybridep
314
+ moe_echo_recompute_expert_dispatch .............. False
315
+ moe_enable_deepep ............................... False
316
+ moe_enable_echo ................................. False
317
+ moe_enable_hybridep ............................. True
318
+ moe_enable_ultraep .............................. True
319
+ moe_expert_capacity_factor ...................... None
320
+ moe_extended_tp ................................. False
321
+ moe_ffn_hidden_size ............................. 14336
322
+ moe_grouped_gemm ................................ True
323
+ moe_hybridep_num_sms ............................ 16
324
+ moe_input_jitter_eps ............................ None
325
+ moe_layer_freq .................................. 1
326
+ moe_layer_recompute ............................. False
327
+ moe_num_echo_experts ............................ None
328
+ moe_num_redundant_experts_per_rank .............. 1
329
+ moe_pad_expert_input_to_capacity ................ False
330
+ moe_per_layer_logging ........................... False
331
+ moe_permute_fusion .............................. False
332
+ moe_received_token_capacity ..................... 2.40625
333
+ moe_router_bias_update_rate ..................... 0.001
334
+ moe_router_dtype ................................ fp32
335
+ moe_router_enable_expert_bias ................... False
336
+ moe_router_force_load_balancing ................. False
337
+ moe_router_group_topk ........................... None
338
+ moe_router_load_balancing_type .................. aux_loss
339
+ moe_router_num_groups ........................... None
340
+ moe_router_padding_for_fp8 ...................... False
341
+ moe_router_pre_softmax .......................... False
342
+ moe_router_score_function ....................... softmax
343
+ moe_router_topk ................................. 2
344
+ moe_router_topk_scaling_factor .................. None
345
+ moe_shared_expert_intermediate_size ............. None
346
+ moe_shared_expert_overlap ....................... False
347
+ moe_token_dispatcher_type ....................... flex
348
+ moe_token_drop_policy ........................... probs
349
+ moe_ultraep_max_microbatches .................... None
350
+ moe_ultraep_overlap_weight_sync ................. True
351
+ moe_ultraep_weight_sync_join .................... before_dispatch
352
+ moe_upcycling_granularity ....................... 1
353
+ moe_use_device_initiated_grouped_gemm ........... True
354
+ moe_use_legacy_grouped_gemm ..................... False
355
+ moe_use_upcycling ............................... False
356
+ moe_z_loss_coeff ................................ None
357
+ mrope_section ................................... None
358
+ mscale .......................................... 1.0
359
+ mscale_all_dim .................................. 0.0
360
+ mtp_loss_scaling_factor ......................... 0.1
361
+ mtp_num_layers .................................. None
362
+ multi_latent_attention .......................... False
363
+ nccl_all_reduce_for_prefill ..................... False
364
+ nccl_communicator_config_path ................... None
365
+ nccl_ub ......................................... False
366
+ no_load_optim ................................... None
367
+ no_load_rng ..................................... None
368
+ no_persist_layer_norm ........................... False
369
+ no_rope_freq .................................... None
370
+ no_save_optim ................................... None
371
+ no_save_rng ..................................... None
372
+ non_persistent_ckpt_type ........................ None
373
+ non_persistent_global_ckpt_dir .................. None
374
+ non_persistent_local_ckpt_algo .................. fully_parallel
375
+ non_persistent_local_ckpt_dir ................... None
376
+ non_persistent_save_interval .................... None
377
+ norm_epsilon .................................... 1e-05
378
+ normalization ................................... RMSNorm
379
+ num_attention_heads ............................. 32
380
+ num_channels .................................... 3
381
+ num_classes ..................................... 1000
382
+ num_dataset_builder_threads ..................... 1
383
+ num_distributed_optimizer_instances ............. 1
384
+ num_experts ..................................... 8
385
+ num_layers ...................................... 16
386
+ num_layers_at_end_in_bf16 ....................... 1
387
+ num_layers_at_start_in_bf16 ..................... 1
388
+ num_layers_per_virtual_pipeline_stage ........... None
389
+ num_query_groups ................................ 8
390
+ num_virtual_stages_per_pipeline_rank ............ None
391
+ num_workers ..................................... 2
392
+ object_storage_cache_path ....................... None
393
+ one_logger_async ................................ False
394
+ one_logger_project .............................. megatron-lm
395
+ one_logger_run_name ............................. None
396
+ onnx_safe ....................................... None
397
+ openai_gelu ..................................... False
398
+ optimizer ....................................... adam
399
+ optimizer_cpu_offload ........................... False
400
+ optimizer_offload_fraction ...................... 1.0
401
+ output_bert_embeddings .......................... False
402
+ overlap_cpu_optimizer_d2h_h2d ................... False
403
+ overlap_grad_reduce ............................. True
404
+ overlap_moe_expert_parallel_comm ................ False
405
+ overlap_p2p_comm ................................ False
406
+ overlap_p2p_comm_warmup_flush ................... False
407
+ overlap_param_gather ............................ True
408
+ overlap_param_gather_with_optimizer_step ........ False
409
+ override_opt_param_scheduler .................... False
410
+ params_dtype .................................... torch.bfloat16
411
+ patch_dim ....................................... 16
412
+ per_split_data_args_path ........................ None
413
+ perform_initialization .......................... True
414
+ pin_cpu_grads ................................... True
415
+ pin_cpu_params .................................. True
416
+ pipeline_model_parallel_comm_backend ............ None
417
+ pipeline_model_parallel_layout .................. None
418
+ pipeline_model_parallel_size .................... 1
419
+ position_embedding_type ......................... rope
420
+ pretrained_checkpoint ........................... None
421
+ profile ......................................... False
422
+ profile_ranks ................................... [0]
423
+ profile_step_end ................................ 12
424
+ profile_step_start .............................. 10
425
+ q_lora_rank ..................................... None
426
+ qk_head_dim ..................................... 128
427
+ qk_l2_norm ...................................... False
428
+ qk_layernorm .................................... False
429
+ qk_pos_emb_head_dim ............................. 64
430
+ query_in_block_prob ............................. 0.1
431
+ rampup_batch_size ............................... None
432
+ rank ............................................ 0
433
+ recompute_granularity ........................... None
434
+ recompute_method ................................ None
435
+ recompute_modules ............................... None
436
+ recompute_num_layers ............................ None
437
+ record_memory_history ........................... False
438
+ relative_attention_max_distance ................. 128
439
+ relative_attention_num_buckets .................. 32
440
+ replication ..................................... False
441
+ replication_factor .............................. 2
442
+ replication_jump ................................ None
443
+ rerun_mode ...................................... validate_results
444
+ reset_attention_mask ............................ False
445
+ reset_position_ids .............................. False
446
+ result_rejected_tracker_filename ................ None
447
+ retriever_report_topk_accuracies ................ []
448
+ retriever_score_scaling ......................... False
449
+ retriever_seq_length ............................ 256
450
+ retro_add_retriever ............................. False
451
+ retro_attention_gate ............................ 1
452
+ retro_cyclic_train_iters ........................ None
453
+ retro_encoder_attention_dropout ................. 0.1
454
+ retro_encoder_hidden_dropout .................... 0.1
455
+ retro_encoder_layers ............................ 2
456
+ retro_num_neighbors ............................. 2
457
+ retro_num_retrieved_chunks ...................... 2
458
+ retro_project_dir ............................... None
459
+ retro_verify_neighbor_count ..................... True
460
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
461
+ rope_scaling_factor ............................. 8.0
462
+ rope_type ....................................... None
463
+ rotary_base ..................................... 1000000
464
+ rotary_interleaved .............................. False
465
+ rotary_percent .................................. 1.0
466
+ rotary_scaling_factor ........................... 1.0
467
+ rotary_seq_len_interpolation_factor ............. None
468
+ run_workload_inspector_server ................... False
469
+ sample_rate ..................................... 1.0
470
+ save ............................................ None
471
+ save_interval ................................... None
472
+ save_retain_interval ............................ None
473
+ scatter_gather_tensors_in_pipeline .............. True
474
+ seed ............................................ 1234
475
+ seq_length ...................................... 4096
476
+ sequence_parallel ............................... False
477
+ sft ............................................. False
478
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
479
+ sgd_momentum .................................... 0.9
480
+ short_seq_prob .................................. 0.1
481
+ skip_train ...................................... False
482
+ skipped_train_samples ........................... 0
483
+ spec ............................................ None
484
+ split ........................................... 100,0,0
485
+ squared_relu .................................... False
486
+ start_weight_decay .............................. 0.1
487
+ straggler_ctrlr_port ............................ 65535
488
+ straggler_minmax_count .......................... 1
489
+ suggested_communication_unit_size ............... None
490
+ swiglu .......................................... True
491
+ swin_backbone_type .............................. tiny
492
+ symmetric_ar_type ............................... None
493
+ te_rng_tracker .................................. True
494
+ tensor_model_parallel_size ...................... 1
495
+ tensorboard_dir ................................. None
496
+ tensorboard_log_interval ........................ 1
497
+ tensorboard_queue_size .......................... 1000
498
+ test_data_path .................................. None
499
+ test_mode ....................................... False
500
+ tiktoken_num_special_tokens ..................... 1000
501
+ tiktoken_pattern ................................ None
502
+ tiktoken_special_tokens ......................... None
503
+ timing_log_level ................................ 0
504
+ timing_log_option ............................... minmax
505
+ titles_data_path ................................ None
506
+ tokenizer_model ................................. None
507
+ tokenizer_type .................................. GPT2BPETokenizer
508
+ torch_fsdp2_reshard_after_forward ............... True
509
+ tp_comm_bootstrap_backend ....................... nccl
510
+ tp_comm_bulk_dgrad .............................. True
511
+ tp_comm_bulk_wgrad .............................. True
512
+ tp_comm_overlap ................................. False
513
+ tp_comm_overlap_ag .............................. True
514
+ tp_comm_overlap_cfg ............................. None
515
+ tp_comm_overlap_rs .............................. True
516
+ tp_comm_overlap_rs_dgrad ........................ False
517
+ tp_comm_split_ag ................................ True
518
+ tp_comm_split_rs ................................ True
519
+ train_data_path ................................. None
520
+ train_iters ..................................... 12
521
+ train_samples ................................... None
522
+ train_sync_interval ............................. None
523
+ transformer_impl ................................ transformer_engine
524
+ transformer_pipeline_model_parallel_size ........ 1
525
+ untie_embeddings_and_output_weights ............. True
526
+ use_checkpoint_args ............................. False
527
+ use_checkpoint_opt_param_scheduler .............. False
528
+ use_cpu_initialization .......................... None
529
+ use_custom_fsdp ................................. False
530
+ use_dist_ckpt ................................... True
531
+ use_dist_ckpt_deprecated ........................ False
532
+ use_distributed_optimizer ....................... True
533
+ use_flash_attn .................................. False
534
+ use_legacy_models ............................... False
535
+ use_mp_args_from_checkpoint_args ................ False
536
+ use_one_sent_docs ............................... False
537
+ use_persistent_ckpt_worker ...................... False
538
+ use_precision_aware_optimizer ................... False
539
+ use_pytorch_profiler ............................ False
540
+ use_ring_exchange_p2p ........................... False
541
+ use_rope_scaling ................................ False
542
+ use_rotary_position_embeddings .................. False
543
+ use_sharp ....................................... False
544
+ use_tokenizer_model_from_checkpoint_args ........ True
545
+ use_torch_fsdp2 ................................. False
546
+ use_torch_optimizer_for_cpu_offload ............. False
547
+ use_tp_pp_dp_mapping ............................ False
548
+ v_head_dim ...................................... 128
549
+ valid_data_path ................................. None
550
+ variable_seq_lengths ............................ False
551
+ virtual_pipeline_model_parallel_size ............ None
552
+ vision_backbone_type ............................ vit
553
+ vision_pretraining .............................. False
554
+ vision_pretraining_type ......................... classify
555
+ vocab_extra_ids ................................. 0
556
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
557
+ vocab_size ...................................... None
558
+ wandb_exp_name ..................................
559
+ wandb_project ...................................
560
+ wandb_save_dir ..................................
561
+ weight_decay .................................... 0.1
562
+ weight_decay_incr_style ......................... constant
563
+ wgrad_deferral_limit ............................ 0
564
+ world_size ...................................... 8
565
+ yaml_cfg ........................................ None
566
+ -------------------- end of arguments ---------------------
567
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
568
+ > building GPT2BPETokenizer tokenizer ...
569
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
570
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
571
+ > initializing torch distributed ...
572
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
573
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
574
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
575
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
576
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
577
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
578
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
579
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
580
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
581
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
582
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank 524 is connected to is connected to is connected to 777 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77
586
+ 7
587
+
588
+ 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank
592
+ 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
597
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [Gloo] Rank [Gloo] Rank 00 is connected to is connected to 00 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 00
604
+
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : [Gloo] Rank 00
607
+ is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank5]:[W830 07:01:02.329387677 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank2]:[W830 07:01:02.334618071 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank7]:[W830 07:01:02.334719399 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank3]:[W830 07:01:02.335063415 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank6]:[W830 07:01:02.335200542 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank4]:[W830 07:01:02.336913144 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank1]:[W830 07:01:02.336971424 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.034 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:01:02.497875239 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.752 seconds
625
+ time to initialize megatron (seconds): 7.203
626
+ [after megatron is initialized] datetime: 2026-08-30 07:01:07
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6720983040
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (41988096 elements, 41988096 padded size):
670
+ module.decoder.final_layernorm.weight
671
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
672
+ module.decoder.layers.15.mlp.router.weight
673
+ module.decoder.layers.15.self_attention.linear_proj.weight
674
+ module.decoder.layers.15.pre_mlp_layernorm.weight
675
+ module.decoder.layers.15.self_attention.linear_qkv.weight
676
+ Params for bucket 3 (41984000 elements, 41984000 padded size):
677
+ module.decoder.layers.14.mlp.router.weight
678
+ module.decoder.layers.14.self_attention.linear_proj.weight
679
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
680
+ module.decoder.layers.14.self_attention.linear_qkv.weight
681
+ module.decoder.layers.14.pre_mlp_layernorm.weight
682
+ Params for bucket 4 (41984000 elements, 41984000 padded size):
683
+ module.decoder.layers.13.mlp.router.weight
684
+ module.decoder.layers.13.self_attention.linear_proj.weight
685
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
686
+ module.decoder.layers.13.pre_mlp_layernorm.weight
687
+ module.decoder.layers.13.self_attention.linear_qkv.weight
688
+ Params for bucket 5 (41984000 elements, 41984000 padded size):
689
+ module.decoder.layers.12.mlp.router.weight
690
+ module.decoder.layers.12.self_attention.linear_proj.weight
691
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
692
+ module.decoder.layers.12.self_attention.linear_qkv.weight
693
+ module.decoder.layers.12.pre_mlp_layernorm.weight
694
+ Params for bucket 6 (41984000 elements, 41984000 padded size):
695
+ module.decoder.layers.11.mlp.router.weight
696
+ module.decoder.layers.11.self_attention.linear_proj.weight
697
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.11.pre_mlp_layernorm.weight
699
+ module.decoder.layers.11.self_attention.linear_qkv.weight
700
+ Params for bucket 7 (41984000 elements, 41984000 padded size):
701
+ module.decoder.layers.10.mlp.router.weight
702
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
703
+ module.decoder.layers.10.self_attention.linear_proj.weight
704
+ module.decoder.layers.10.self_attention.linear_qkv.weight
705
+ module.decoder.layers.10.pre_mlp_layernorm.weight
706
+ Params for bucket 8 (41984000 elements, 41984000 padded size):
707
+ module.decoder.layers.9.mlp.router.weight
708
+ module.decoder.layers.9.self_attention.linear_proj.weight
709
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
710
+ module.decoder.layers.9.self_attention.linear_qkv.weight
711
+ module.decoder.layers.9.pre_mlp_layernorm.weight
712
+ Params for bucket 9 (41984000 elements, 41984000 padded size):
713
+ module.decoder.layers.8.mlp.router.weight
714
+ module.decoder.layers.8.self_attention.linear_proj.weight
715
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
716
+ module.decoder.layers.8.pre_mlp_layernorm.weight
717
+ module.decoder.layers.8.self_attention.linear_qkv.weight
718
+ Params for bucket 10 (41984000 elements, 41984000 padded size):
719
+ module.decoder.layers.7.mlp.router.weight
720
+ module.decoder.layers.7.self_attention.linear_proj.weight
721
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
722
+ module.decoder.layers.7.self_attention.linear_qkv.weight
723
+ module.decoder.layers.7.pre_mlp_layernorm.weight
724
+ Params for bucket 11 (41984000 elements, 41984000 padded size):
725
+ module.decoder.layers.6.mlp.router.weight
726
+ module.decoder.layers.6.self_attention.linear_proj.weight
727
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
728
+ module.decoder.layers.6.pre_mlp_layernorm.weight
729
+ module.decoder.layers.6.self_attention.linear_qkv.weight
730
+ Params for bucket 12 (41984000 elements, 41984000 padded size):
731
+ module.decoder.layers.5.mlp.router.weight
732
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
733
+ module.decoder.layers.5.self_attention.linear_qkv.weight
734
+ module.decoder.layers.5.self_attention.linear_proj.weight
735
+ module.decoder.layers.5.pre_mlp_layernorm.weight
736
+ Params for bucket 13 (41984000 elements, 41984000 padded size):
737
+ module.decoder.layers.4.pre_mlp_layernorm.weight
738
+ module.decoder.layers.4.self_attention.linear_qkv.weight
739
+ module.decoder.layers.4.mlp.router.weight
740
+ module.decoder.layers.4.self_attention.linear_proj.weight
741
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
742
+ Params for bucket 14 (41984000 elements, 41984000 padded size):
743
+ module.decoder.layers.3.self_attention.linear_qkv.weight
744
+ module.decoder.layers.3.pre_mlp_layernorm.weight
745
+ module.decoder.layers.3.mlp.router.weight
746
+ module.decoder.layers.3.self_attention.linear_proj.weight
747
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
748
+ Params for bucket 15 (41984000 elements, 41984000 padded size):
749
+ module.decoder.layers.2.pre_mlp_layernorm.weight
750
+ module.decoder.layers.2.self_attention.linear_qkv.weight
751
+ module.decoder.layers.2.mlp.router.weight
752
+ module.decoder.layers.2.self_attention.linear_proj.weight
753
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
754
+ Params for bucket 16 (41984000 elements, 41984000 padded size):
755
+ module.decoder.layers.1.pre_mlp_layernorm.weight
756
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
757
+ module.decoder.layers.1.mlp.router.weight
758
+ module.decoder.layers.1.self_attention.linear_qkv.weight
759
+ module.decoder.layers.1.self_attention.linear_proj.weight
760
+ Params for bucket 17 (41984000 elements, 41984000 padded size):
761
+ module.decoder.layers.0.mlp.router.weight
762
+ module.decoder.layers.0.self_attention.linear_qkv.weight
763
+ module.decoder.layers.0.pre_mlp_layernorm.weight
764
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
765
+ module.decoder.layers.0.self_attention.linear_proj.weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 32
769
+ Params for bucket 1 (58720256 elements, 58720256 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
771
+ Params for bucket 2 (117440512 elements, 117440512 padded size):
772
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
773
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
774
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
775
+ Params for bucket 4 (117440512 elements, 117440512 padded size):
776
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
777
+ Params for bucket 5 (58720256 elements, 58720256 padded size):
778
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
779
+ Params for bucket 6 (117440512 elements, 117440512 padded size):
780
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
781
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
782
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
783
+ Params for bucket 8 (117440512 elements, 117440512 padded size):
784
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
785
+ Params for bucket 9 (58720256 elements, 58720256 padded size):
786
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
787
+ Params for bucket 10 (117440512 elements, 117440512 padded size):
788
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 11 (58720256 elements, 58720256 padded size):
790
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
791
+ Params for bucket 12 (117440512 elements, 117440512 padded size):
792
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
793
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
794
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
795
+ Params for bucket 14 (117440512 elements, 117440512 padded size):
796
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
797
+ Params for bucket 15 (58720256 elements, 58720256 padded size):
798
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
799
+ Params for bucket 16 (117440512 elements, 117440512 padded size):
800
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
801
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
802
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
803
+ Params for bucket 18 (117440512 elements, 117440512 padded size):
804
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
805
+ Params for bucket 19 (58720256 elements, 58720256 padded size):
806
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
807
+ Params for bucket 20 (117440512 elements, 117440512 padded size):
808
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
809
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
811
+ Params for bucket 22 (117440512 elements, 117440512 padded size):
812
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
813
+ Params for bucket 23 (58720256 elements, 58720256 padded size):
814
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
815
+ Params for bucket 24 (117440512 elements, 117440512 padded size):
816
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
817
+ Params for bucket 25 (58720256 elements, 58720256 padded size):
818
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
819
+ Params for bucket 26 (117440512 elements, 117440512 padded size):
820
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
821
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
822
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
823
+ Params for bucket 28 (117440512 elements, 117440512 padded size):
824
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
825
+ Params for bucket 29 (58720256 elements, 58720256 padded size):
826
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
827
+ Params for bucket 30 (117440512 elements, 117440512 padded size):
828
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
829
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
830
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
831
+ Params for bucket 32 (117440512 elements, 117440512 padded size):
832
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
833
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x789a6c9f4e00>, config_logger_dir='')
834
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
835
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:01:11
836
+ > building train, validation, and test datasets ...
837
+ > datasets target sizes (minimum size):
838
+ train: 192
839
+ validation: 0
840
+ test: 0
841
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
842
+ > building train, validation, and test datasets for GPT ...
843
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x789a6d1f99a0>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
844
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
845
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
846
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
847
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
848
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
849
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
850
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
851
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
852
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
853
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
854
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
855
+ > finished creating GPT datasets ...
856
+ [after dataloaders are built] datetime: 2026-08-30 07:01:11
857
+ done with setup ...
858
+ (min, max) time across ranks (ms):
859
+ model-and-optimizer-setup ......................: (3279.64, 3282.88)
860
+ train/valid/test-data-iterators-setup ..........: (77.29, 92.32)
861
+ training ...
862
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
863
+ [before the start of training step] datetime: 2026-08-30 07:01:11
864
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
865
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
866
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
867
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
868
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
869
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
870
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
871
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
872
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
873
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
874
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
875
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
876
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
877
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
878
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
879
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
880
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
881
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
882
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
883
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
884
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
885
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
886
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
887
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
888
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
889
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
890
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
891
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
892
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
893
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
894
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
895
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
896
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
897
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
898
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
899
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
900
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
901
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
902
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
903
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
904
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
905
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
906
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
907
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
908
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
909
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
910
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
911
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
912
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
913
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
914
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
915
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
916
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
917
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
918
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
919
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
920
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
921
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
922
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
923
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
924
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
925
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
926
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
927
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
928
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
929
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
930
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
931
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
932
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
933
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
934
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
935
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
936
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
937
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
938
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
939
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
940
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
941
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
942
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
943
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
944
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
945
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
946
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
947
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
948
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
949
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
950
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
951
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
952
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
953
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
954
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
955
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
956
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
957
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
958
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
959
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
960
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
961
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
962
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
963
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
964
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
965
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
966
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
967
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
968
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
969
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
970
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
971
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
972
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
973
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
974
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
975
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
976
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
977
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
978
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
979
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
980
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
981
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
982
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
983
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
985
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
989
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
991
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
992
+ [2026-08-30 07:01:37] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26075.9 | throughput per GPU (TFLOP/s/GPU): 12.8 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.104376E+01 | loss scale: 1.0 | grad norm: 35.707 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
993
+ Number of parameters in transformer block in billions: 23.22
994
+ Number of parameters in embedding layers in billions: 0.41
995
+ Total number of parameters in billions: 23.63
996
+ Number of parameters in most loaded shard in billions: 23.6320
997
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
998
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58364.48876953125 | max allocated: 85653.79150390625 | reserved: 87692.0 | max reserved: 87692.0
999
+ [2026-08-30 07:01:38] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 827.5 | throughput per GPU (TFLOP/s/GPU): 402.9 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.104935E+01 | loss scale: 1.0 | grad norm: 37.623 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1000
+ [2026-08-30 07:01:39] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 867.8 | throughput per GPU (TFLOP/s/GPU): 384.2 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.680098E+00 | loss scale: 1.0 | grad norm: 42.136 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1001
+ Capture CUDA graph for training!!!
1002
+ CUDA graph capture done!!!
1003
+ [2026-08-30 07:01:41] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1921.7 | throughput per GPU (TFLOP/s/GPU): 173.5 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.120050E+01 | loss scale: 1.0 | grad norm: 88.922 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1004
+ [2026-08-30 07:01:42] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 875.8 | throughput per GPU (TFLOP/s/GPU): 380.7 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.017120E+01 | loss scale: 1.0 | grad norm: 15.488 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1005
+ [2026-08-30 07:01:42] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 907.1 | throughput per GPU (TFLOP/s/GPU): 367.5 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.014204E+01 | loss scale: 1.0 | grad norm: 42.271 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1006
+ [2026-08-30 07:01:43] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 898.0 | throughput per GPU (TFLOP/s/GPU): 371.3 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.865644E+00 | loss scale: 1.0 | grad norm: 8.146 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1007
+ [2026-08-30 07:01:44] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 895.6 | throughput per GPU (TFLOP/s/GPU): 372.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.028494E+00 | loss scale: 1.0 | grad norm: 6.679 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1008
+ [2026-08-30 07:01:45] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 884.2 | throughput per GPU (TFLOP/s/GPU): 377.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.672704E+00 | loss scale: 1.0 | grad norm: 6.730 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1009
+ [2026-08-30 07:01:46] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 869.8 | throughput per GPU (TFLOP/s/GPU): 383.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.460926E+00 | loss scale: 1.0 | grad norm: 5.894 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1010
+ [2026-08-30 07:01:47] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 882.1 | throughput per GPU (TFLOP/s/GPU): 377.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.341253E+00 | loss scale: 1.0 | grad norm: 4.649 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1011
+ [2026-08-30 07:01:48] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 883.4 | throughput per GPU (TFLOP/s/GPU): 377.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.275044E+00 | loss scale: 1.0 | grad norm: 4.839 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18688 | min safe capacity: 2.2812 |
1012
+ [after training is done] datetime: 2026-08-30 07:01:48
1013
+ [rank0]:[W830 07:01:49.451171155 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1014
+ [rank4]:[W830 07:01:50.810736529 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1015
+ [rank3]:[W830 07:01:50.913330553 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1016
+ [rank7]:[W830 07:01:50.931163487 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1017
+ [rank2]:[W830 07:01:50.931884165 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1018
+ [rank1]:[W830 07:01:50.932593088 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1019
+ [rank6]:[W830 07:01:50.933320394 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1020
+ [rank5]:[W830 07:01:51.480296105 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1021
+ W0830 07:01:52.852000 693438 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1022
+ W0830 07:01:52.852000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693527 closing signal SIGTERM
1023
+ W0830 07:01:52.853000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693528 closing signal SIGTERM
1024
+ W0830 07:01:52.854000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693529 closing signal SIGTERM
1025
+ W0830 07:01:52.855000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693530 closing signal SIGTERM
1026
+ W0830 07:01:52.855000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693531 closing signal SIGTERM
1027
+ W0830 07:01:52.856000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693532 closing signal SIGTERM
1028
+ W0830 07:01:52.856000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693533 closing signal SIGTERM
1029
+ W0830 07:01:52.857000 693438 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 693534 closing signal SIGTERM
1030
+ Traceback (most recent call last):
1031
+ File "/usr/local/bin/torchrun", line 8, in <module>
1032
+ sys.exit(main())
1033
+ ^^^^^^
1034
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1035
+ return f(*args, **kwargs)
1036
+ ^^^^^^^^^^^^^^^^^^
1037
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1038
+ run(args)
1039
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1040
+ elastic_launch(
1041
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1042
+ return launch_agent(self._config, self._entrypoint, list(args))
1043
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1044
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1045
+ result = agent.run()
1046
+ ^^^^^^^^^^^
1047
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1048
+ result = f(*args, **kwargs)
1049
+ ^^^^^^^^^^^^^^^^^^
1050
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1051
+ result = self._invoke_run(role)
1052
+ ^^^^^^^^^^^^^^^^^^^^^^
1053
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1054
+ time.sleep(monitor_interval)
1055
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1056
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1057
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 693438 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2/mem_samples.csv ADDED
@@ -0,0 +1,232 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 1
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 1
41
+ 0, 4
42
+ 1, 4
43
+ 2, 4
44
+ 3, 4
45
+ 4, 4
46
+ 5, 4
47
+ 6, 4
48
+ 7, 4
49
+ 0, 725
50
+ 1, 725
51
+ 2, 725
52
+ 3, 725
53
+ 4, 725
54
+ 5, 725
55
+ 6, 725
56
+ 7, 725
57
+ 0, 937
58
+ 1, 939
59
+ 2, 959
60
+ 3, 953
61
+ 4, 939
62
+ 5, 947
63
+ 6, 957
64
+ 7, 949
65
+ 0, 1895
66
+ 1, 1895
67
+ 2, 1895
68
+ 3, 1895
69
+ 4, 1895
70
+ 5, 1895
71
+ 6, 1895
72
+ 7, 1895
73
+ 0, 2297
74
+ 1, 2311
75
+ 2, 2301
76
+ 3, 2311
77
+ 4, 2313
78
+ 5, 2311
79
+ 6, 2319
80
+ 7, 2303
81
+ 0, 42469
82
+ 1, 42449
83
+ 2, 42467
84
+ 3, 42467
85
+ 4, 42467
86
+ 5, 42467
87
+ 6, 42433
88
+ 7, 42465
89
+ 0, 43259
90
+ 1, 43257
91
+ 2, 43257
92
+ 3, 43257
93
+ 4, 43257
94
+ 5, 43257
95
+ 6, 43257
96
+ 7, 43257
97
+ 0, 44167
98
+ 1, 44165
99
+ 2, 44165
100
+ 3, 44165
101
+ 4, 44165
102
+ 5, 44165
103
+ 6, 44165
104
+ 7, 44165
105
+ 0, 44167
106
+ 1, 44165
107
+ 2, 44165
108
+ 3, 44165
109
+ 4, 44165
110
+ 5, 44165
111
+ 6, 44165
112
+ 7, 44165
113
+ 0, 45939
114
+ 1, 45937
115
+ 2, 45937
116
+ 3, 45937
117
+ 4, 45937
118
+ 5, 45937
119
+ 6, 45937
120
+ 7, 45937
121
+ 0, 75579
122
+ 1, 75575
123
+ 2, 75575
124
+ 3, 75575
125
+ 4, 75631
126
+ 5, 76655
127
+ 6, 75577
128
+ 7, 75575
129
+ 0, 81289
130
+ 1, 81287
131
+ 2, 81287
132
+ 3, 81287
133
+ 4, 81287
134
+ 5, 81287
135
+ 6, 81287
136
+ 7, 81287
137
+ 0, 82367
138
+ 1, 82365
139
+ 2, 82365
140
+ 3, 82365
141
+ 4, 82365
142
+ 5, 82365
143
+ 6, 82365
144
+ 7, 82365
145
+ 0, 93455
146
+ 1, 93395
147
+ 2, 93395
148
+ 3, 93395
149
+ 4, 93453
150
+ 5, 93395
151
+ 6, 93395
152
+ 7, 93395
153
+ 0, 95571
154
+ 1, 95511
155
+ 2, 95511
156
+ 3, 95511
157
+ 4, 95571
158
+ 5, 95511
159
+ 6, 95511
160
+ 7, 95511
161
+ 0, 96811
162
+ 1, 96743
163
+ 2, 96739
164
+ 3, 96783
165
+ 4, 96829
166
+ 5, 96789
167
+ 6, 96815
168
+ 7, 96773
169
+ 0, 111681
170
+ 1, 109673
171
+ 2, 85457
172
+ 3, 77911
173
+ 4, 81511
174
+ 5, 78989
175
+ 6, 119751
176
+ 7, 119303
177
+ 0, 129221
178
+ 1, 129109
179
+ 2, 129109
180
+ 3, 129109
181
+ 4, 129475
182
+ 5, 129109
183
+ 6, 129109
184
+ 7, 129109
185
+ 0, 129221
186
+ 1, 129109
187
+ 2, 129109
188
+ 3, 129109
189
+ 4, 129475
190
+ 5, 129109
191
+ 6, 129109
192
+ 7, 129109
193
+ 0, 129221
194
+ 1, 129109
195
+ 2, 129109
196
+ 3, 129109
197
+ 4, 129475
198
+ 5, 129109
199
+ 6, 129109
200
+ 7, 129109
201
+ 0, 129221
202
+ 1, 129109
203
+ 2, 129109
204
+ 3, 129109
205
+ 4, 129475
206
+ 5, 129109
207
+ 6, 129109
208
+ 7, 129109
209
+ 0, 128313
210
+ 1, 128201
211
+ 2, 128201
212
+ 3, 128201
213
+ 4, 128567
214
+ 5, 128201
215
+ 6, 128201
216
+ 7, 128201
217
+ 0, 124065
218
+ 1, 123953
219
+ 2, 123953
220
+ 3, 123953
221
+ 4, 124319
222
+ 5, 124017
223
+ 6, 124599
224
+ 7, 124277
225
+ 0, 1
226
+ 1, 1
227
+ 2, 1
228
+ 3, 3
229
+ 4, 4
230
+ 5, 196
231
+ 6, 124599
232
+ 7, 964
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G1
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=14336
11
+ num_experts=8
12
+ topk=2
13
+ experts_per_rank=1
14
+ expert_params=176160768
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=8192
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=2.40625
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2/train.log ADDED
@@ -0,0 +1,1057 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G1 / ultraep (slots/rank=1 policy=one)
3
+ E=8 K=2 H'=14336 m=1
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=2.40625 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:01:58.361000 696450 torch/distributed/run.py:766]
10
+ W0830 07:01:58.361000 696450 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:01:58.361000 696450 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:01:58.361000 696450 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
16
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
17
+ Number of virtual stages per pipeline stage: None
18
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
19
+ using torch.bfloat16 for parameters ...
20
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
21
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
22
+ ------------------------ arguments ------------------------
23
+ account_for_embedding_in_pipeline_split ......... False
24
+ account_for_loss_in_pipeline_split .............. False
25
+ accumulate_allreduce_grads_in_fp32 .............. True
26
+ adam_beta1 ...................................... 0.9
27
+ adam_beta2 ...................................... 0.999
28
+ adam_eps ........................................ 1e-08
29
+ add_bias_linear ................................. False
30
+ add_position_embedding .......................... False
31
+ add_qkv_bias .................................... False
32
+ adlr_autoresume ................................. False
33
+ adlr_autoresume_interval ........................ 1000
34
+ align_grad_reduce ............................... True
35
+ align_param_gather .............................. False
36
+ app_tag_run_name ................................ None
37
+ app_tag_run_version ............................. 0.0.0
38
+ apply_layernorm_1p .............................. False
39
+ apply_query_key_layer_scaling ................... False
40
+ apply_residual_connection_post_layernorm ........ False
41
+ apply_rope_fusion ............................... True
42
+ async_save ...................................... None
43
+ async_tensor_model_parallel_allreduce ........... True
44
+ attention_backend ............................... AttnBackend.auto
45
+ attention_dropout ............................... 0.0
46
+ attention_softmax_in_fp32 ....................... False
47
+ auto_detect_ckpt_format ......................... False
48
+ barrier_with_L1_time ............................ True
49
+ bert_binary_head ................................ True
50
+ bert_embedder_type .............................. megatron
51
+ bert_load ....................................... None
52
+ bf16 ............................................ True
53
+ bias_dropout_fusion ............................. True
54
+ bias_gelu_fusion ................................ False
55
+ bias_swiglu_fusion .............................. True
56
+ biencoder_projection_dim ........................ 0
57
+ biencoder_shared_query_context_model ............ False
58
+ block_data_path ................................. None
59
+ calc_ft_timeouts ................................ False
60
+ calculate_per_token_loss ........................ False
61
+ check_for_large_grads ........................... False
62
+ check_for_nan_in_loss_and_grad .................. False
63
+ check_for_spiky_loss ............................ False
64
+ check_weight_hash_across_dp_replicas_interval ... None
65
+ ckpt_assume_constant_structure .................. False
66
+ ckpt_convert_format ............................. None
67
+ ckpt_convert_save ............................... None
68
+ ckpt_convert_update_legacy_dist_opt_format ...... False
69
+ ckpt_format ..................................... torch_dist
70
+ ckpt_fully_parallel_load ........................ False
71
+ ckpt_fully_parallel_save ........................ True
72
+ ckpt_fully_parallel_save_deprecated ............. False
73
+ ckpt_step ....................................... None
74
+ classes_fraction ................................ 1.0
75
+ clip_grad ....................................... 1.0
76
+ clone_scatter_output_in_embedding ............... True
77
+ config_logger_dir ...............................
78
+ consumed_train_samples .......................... 0
79
+ consumed_valid_samples .......................... 0
80
+ context_parallel_size ........................... 1
81
+ cp_comm_type .................................... ['p2p']
82
+ create_attention_mask_in_dataloader ............. True
83
+ cross_entropy_fusion_impl ....................... native
84
+ cross_entropy_loss_fusion ....................... False
85
+ cuda_graph_scope ................................ full_iteration
86
+ cuda_graph_warmup_steps ......................... 3
87
+ data_args_path .................................. None
88
+ data_cache_path ................................. None
89
+ data_parallel_random_init ....................... False
90
+ data_parallel_sharding_strategy ................. no_shard
91
+ data_parallel_size .............................. 8
92
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
93
+ data_per_class_fraction ......................... 1.0
94
+ data_sharding ................................... True
95
+ dataloader_type ................................. single
96
+ ddp_average_in_collective ....................... False
97
+ ddp_bucket_size ................................. None
98
+ ddp_num_buckets ................................. None
99
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
100
+ decoder_first_pipeline_num_layers ............... None
101
+ decoder_last_pipeline_num_layers ................ None
102
+ decoder_num_layers .............................. None
103
+ decoder_seq_length .............................. None
104
+ decoupled_lr .................................... None
105
+ decoupled_min_lr ................................ None
106
+ decrease_batch_size_if_needed ................... False
107
+ defer_embedding_wgrad_compute ................... False
108
+ delay_wgrad_compute ............................. False
109
+ deprecated_use_mcore_models ..................... True
110
+ deterministic_mode .............................. False
111
+ dino_bottleneck_size ............................ 256
112
+ dino_freeze_last_layer .......................... 1
113
+ dino_head_hidden_size ........................... 2048
114
+ dino_local_crops_number ......................... 10
115
+ dino_local_img_size ............................. 96
116
+ dino_norm_last_layer ............................ False
117
+ dino_teacher_temp ............................... 0.07
118
+ dino_warmup_teacher_temp ........................ 0.04
119
+ dino_warmup_teacher_temp_epochs ................. 30
120
+ disable_bf16_reduced_precision_matmul ........... False
121
+ disable_mamba_mem_eff_path ...................... False
122
+ disable_straggler_on_startup .................... False
123
+ dist_ckpt_format_deprecated ..................... None
124
+ dist_ckpt_strictness ............................ assume_ok_unexpected
125
+ distribute_saved_activations .................... False
126
+ distributed_backend ............................. nccl
127
+ distributed_timeout_minutes ..................... 10
128
+ embedding_init_method_std ....................... None
129
+ embedding_path .................................. None
130
+ empty_unused_memory_level ....................... 0
131
+ enable_cuda_graph ............................... True
132
+ enable_experimental ............................. False
133
+ enable_ft_package ............................... False
134
+ enable_gloo_process_groups ...................... True
135
+ enable_msc ...................................... True
136
+ enable_one_logger ............................... True
137
+ encoder_num_layers .............................. 16
138
+ encoder_seq_length .............................. 4096
139
+ end_weight_decay ................................ 0.1
140
+ eod_mask_loss ................................... False
141
+ error_injection_rate ............................ 0
142
+ error_injection_type ............................ transient_error
143
+ eval_interval ................................... 1000000
144
+ eval_iters ...................................... 0
145
+ evidence_data_path .............................. None
146
+ exit_duration_in_mins ........................... None
147
+ exit_interval ................................... None
148
+ exit_on_missing_checkpoint ...................... False
149
+ exit_signal_handler ............................. False
150
+ exp_avg_dtype ................................... torch.float32
151
+ exp_avg_sq_dtype ................................ torch.float32
152
+ expert_model_parallel_size ...................... 8
153
+ expert_tensor_parallel_size ..................... 1
154
+ export_force_local_attention .................... False
155
+ export_kd_cfg ................................... None
156
+ export_kd_teacher_ckpt_format ................... None
157
+ export_kd_teacher_load .......................... None
158
+ export_kv_cache_quant ........................... False
159
+ export_legacy_megatron .......................... False
160
+ export_model_type ............................... GPTModel
161
+ export_moe_apply_probs_on_input ................. False
162
+ export_qk_l2_norm ............................... False
163
+ export_quant_cfg ................................ None
164
+ export_real_quant_cfg ........................... None
165
+ export_te_mcore_model ........................... False
166
+ external_cuda_graph ............................. False
167
+ ffn_hidden_size ................................. 14336
168
+ finetune ........................................ False
169
+ finetune_data_split ............................. train
170
+ finetune_hf_dataset ............................. None
171
+ first_last_layers_bf16 .......................... False
172
+ flash_decode .................................... False
173
+ fp16 ............................................ False
174
+ fp16_lm_cross_entropy ........................... False
175
+ fp32_residual_connection ........................ False
176
+ fp8 ............................................. e4m3
177
+ fp8_amax_compute_algo ........................... most_recent
178
+ fp8_amax_history_len ............................ 1
179
+ fp8_interval .................................... 1
180
+ fp8_margin ...................................... 0
181
+ fp8_param_gather ................................ False
182
+ fp8_recipe ...................................... mxfp8
183
+ fp8_wgrad ....................................... True
184
+ fsdp_double_buffer .............................. False
185
+ global_batch_size ............................... 16
186
+ grad_reduce_in_bf16 ............................. False
187
+ gradient_accumulation_fusion .................... True
188
+ gradient_reduce_div_fusion ...................... True
189
+ group_query_attention ........................... True
190
+ head_lr_mult .................................... 1.0
191
+ heterogeneous_layers_config_encoded_json ........ None
192
+ heterogeneous_layers_config_path ................ None
193
+ hidden_dropout .................................. 0.0
194
+ hidden_size ..................................... 4096
195
+ hierarchical_context_parallel_sizes ............. None
196
+ high_priority_stream_groups ..................... []
197
+ hybrid_attention_ratio .......................... 0.0
198
+ hybrid_mlp_ratio ................................ 0.0
199
+ hybrid_override_pattern ......................... None
200
+ hysteresis ...................................... 2
201
+ ict_head_size ................................... None
202
+ ict_load ........................................ None
203
+ img_h ........................................... 224
204
+ img_w ........................................... 224
205
+ indexer_batch_size .............................. 128
206
+ indexer_log_interval ............................ 1000
207
+ inference_batch_times_seqlen_threshold .......... -1
208
+ inference_dynamic_batching ...................... False
209
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
210
+ inference_dynamic_batching_buffer_overflow_factor None
211
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
212
+ inference_dynamic_batching_chunk_size ........... 256
213
+ inference_dynamic_batching_max_requests_override None
214
+ inference_dynamic_batching_max_tokens_override .. None
215
+ inference_dynamic_batching_num_cuda_graphs ...... 16
216
+ inference_max_batch_size ........................ 8
217
+ inference_max_seq_length ........................ 2560
218
+ inference_rng_tracker ........................... False
219
+ init_method_std ................................. 0.01
220
+ init_method_xavier_uniform ...................... False
221
+ init_model_with_meta_device ..................... False
222
+ initial_loss_scale .............................. 4294967296
223
+ inprocess_active_world_size ..................... 8
224
+ inprocess_barrier_timeout ....................... 120
225
+ inprocess_completion_timeout .................... 120
226
+ inprocess_empty_cuda_cache ...................... False
227
+ inprocess_granularity ........................... node
228
+ inprocess_hard_timeout .......................... 90
229
+ inprocess_heartbeat_interval .................... 30
230
+ inprocess_heartbeat_timeout ..................... 60
231
+ inprocess_last_call_wait ........................ 1
232
+ inprocess_max_iterations ........................ None
233
+ inprocess_monitor_process_interval .............. 1.0
234
+ inprocess_monitor_thread_interval ............... 1.0
235
+ inprocess_progress_watchdog_interval ............ 1.0
236
+ inprocess_restart ............................... False
237
+ inprocess_soft_timeout .......................... 60
238
+ inprocess_termination_grace_time ................ 1
239
+ is_hybrid_model ................................. False
240
+ iter_per_epoch .................................. 1250
241
+ iterations_to_skip .............................. []
242
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
243
+ kitchen_config_file ............................. None
244
+ kitchen_recipe_number ........................... None
245
+ kv_channels ..................................... 128
246
+ kv_lora_rank .................................... 32
247
+ lazy_mpu_init ................................... None
248
+ load ............................................ None
249
+ load_main_params_from_ckpt ...................... None
250
+ load_model_opt_format ........................... False
251
+ local_rank ...................................... 0
252
+ log_energy ...................................... False
253
+ log_interval .................................... 1
254
+ log_loss_scale_to_tensorboard ................... True
255
+ log_memory_to_tensorboard ....................... False
256
+ log_num_zeros_in_grad ........................... False
257
+ log_params_norm ................................. False
258
+ log_progress .................................... False
259
+ log_straggler ................................... False
260
+ log_throughput .................................. True
261
+ log_timers_to_tensorboard ....................... False
262
+ log_validation_ppl_to_tensorboard ............... False
263
+ log_world_size_to_tensorboard ................... False
264
+ logging_level ................................... None
265
+ loss_scale ...................................... None
266
+ loss_scale_window ............................... 1000
267
+ lr .............................................. 0.0001
268
+ lr_decay_iters .................................. 8
269
+ lr_decay_samples ................................ None
270
+ lr_decay_style .................................. cosine
271
+ lr_warmup_fraction .............................. None
272
+ lr_warmup_init .................................. 0.0
273
+ lr_warmup_iters ................................. 5
274
+ lr_warmup_samples ............................... 0
275
+ lr_wsd_decay_iters .............................. None
276
+ lr_wsd_decay_samples ............................ None
277
+ lr_wsd_decay_style .............................. exponential
278
+ main_grads_dtype ................................ torch.float32
279
+ main_params_dtype ............................... torch.float32
280
+ make_vocab_size_divisible_by .................... 128
281
+ mamba_head_dim .................................. 64
282
+ mamba_num_groups ................................ 8
283
+ mamba_num_heads ................................. None
284
+ mamba_state_dim ................................. 128
285
+ manual_gc ....................................... False
286
+ manual_gc_eval .................................. True
287
+ manual_gc_interval .............................. 0
288
+ mask_factor ..................................... 1.0
289
+ mask_prob ....................................... 0.15
290
+ mask_type ....................................... random
291
+ masked_softmax_fusion ........................... False
292
+ max_position_embeddings ......................... 32768
293
+ max_tokens_to_oom ............................... 12000
294
+ memory_snapshot_path ............................ snapshot.pickle
295
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
296
+ micro_batch_size ................................ 1
297
+ microbatch_group_size_per_vp_stage .............. None
298
+ mid_level_dataset_surplus ....................... 0.005
299
+ min_loss_scale .................................. 1.0
300
+ min_lr .......................................... 1e-05
301
+ mlp_chunks_for_prefill .......................... 1
302
+ mmap_bin_files .................................. True
303
+ mock_data ....................................... False
304
+ moe_apply_probs_on_input ........................ False
305
+ moe_aux_loss_coeff .............................. 0.01
306
+ moe_capacity_monitor ............................ True
307
+ moe_deepep_num_sms .............................. 20
308
+ moe_echo_enable_random_offloading ............... False
309
+ moe_echo_expert_dispatch_overlap ................ False
310
+ moe_echo_expert_dispatcher_type ................. hybridep
311
+ moe_echo_recompute_expert_dispatch .............. False
312
+ moe_enable_deepep ............................... False
313
+ moe_enable_echo ................................. False
314
+ moe_enable_hybridep ............................. True
315
+ moe_enable_ultraep .............................. True
316
+ moe_expert_capacity_factor ...................... None
317
+ moe_extended_tp ................................. False
318
+ moe_ffn_hidden_size ............................. 14336
319
+ moe_grouped_gemm ................................ True
320
+ moe_hybridep_num_sms ............................ 16
321
+ moe_input_jitter_eps ............................ None
322
+ moe_layer_freq .................................. 1
323
+ moe_layer_recompute ............................. False
324
+ moe_num_echo_experts ............................ None
325
+ moe_num_redundant_experts_per_rank .............. 1
326
+ moe_pad_expert_input_to_capacity ................ False
327
+ moe_per_layer_logging ........................... False
328
+ moe_permute_fusion .............................. False
329
+ moe_received_token_capacity ..................... 2.40625
330
+ moe_router_bias_update_rate ..................... 0.001
331
+ moe_router_dtype ................................ fp32
332
+ moe_router_enable_expert_bias ................... False
333
+ moe_router_force_load_balancing ................. False
334
+ moe_router_group_topk ........................... None
335
+ moe_router_load_balancing_type .................. aux_loss
336
+ moe_router_num_groups ........................... None
337
+ moe_router_padding_for_fp8 ...................... False
338
+ moe_router_pre_softmax .......................... False
339
+ moe_router_score_function ....................... softmax
340
+ moe_router_topk ................................. 2
341
+ moe_router_topk_scaling_factor .................. None
342
+ moe_shared_expert_intermediate_size ............. None
343
+ moe_shared_expert_overlap ....................... False
344
+ moe_token_dispatcher_type ....................... flex
345
+ moe_token_drop_policy ........................... probs
346
+ moe_ultraep_max_microbatches .................... None
347
+ moe_ultraep_overlap_weight_sync ................. True
348
+ moe_ultraep_weight_sync_join .................... before_dispatch
349
+ moe_upcycling_granularity ....................... 1
350
+ moe_use_device_initiated_grouped_gemm ........... True
351
+ moe_use_legacy_grouped_gemm ..................... False
352
+ moe_use_upcycling ............................... False
353
+ moe_z_loss_coeff ................................ None
354
+ mrope_section ................................... None
355
+ mscale .......................................... 1.0
356
+ mscale_all_dim .................................. 0.0
357
+ mtp_loss_scaling_factor ......................... 0.1
358
+ mtp_num_layers .................................. None
359
+ multi_latent_attention .......................... False
360
+ nccl_all_reduce_for_prefill ..................... False
361
+ nccl_communicator_config_path ................... None
362
+ nccl_ub ......................................... False
363
+ no_load_optim ................................... None
364
+ no_load_rng ..................................... None
365
+ no_persist_layer_norm ........................... False
366
+ no_rope_freq .................................... None
367
+ no_save_optim ................................... None
368
+ no_save_rng ..................................... None
369
+ non_persistent_ckpt_type ........................ None
370
+ non_persistent_global_ckpt_dir .................. None
371
+ non_persistent_local_ckpt_algo .................. fully_parallel
372
+ non_persistent_local_ckpt_dir ................... None
373
+ non_persistent_save_interval .................... None
374
+ norm_epsilon .................................... 1e-05
375
+ normalization ................................... RMSNorm
376
+ num_attention_heads ............................. 32
377
+ num_channels .................................... 3
378
+ num_classes ..................................... 1000
379
+ num_dataset_builder_threads ..................... 1
380
+ num_distributed_optimizer_instances ............. 1
381
+ num_experts ..................................... 8
382
+ num_layers ...................................... 16
383
+ num_layers_at_end_in_bf16 ....................... 1
384
+ num_layers_at_start_in_bf16 ..................... 1
385
+ num_layers_per_virtual_pipeline_stage ........... None
386
+ num_query_groups ................................ 8
387
+ num_virtual_stages_per_pipeline_rank ............ None
388
+ num_workers ..................................... 2
389
+ object_storage_cache_path ....................... None
390
+ one_logger_async ................................ False
391
+ one_logger_project .............................. megatron-lm
392
+ one_logger_run_name ............................. None
393
+ onnx_safe ....................................... None
394
+ openai_gelu ..................................... False
395
+ optimizer ....................................... adam
396
+ optimizer_cpu_offload ........................... False
397
+ optimizer_offload_fraction ...................... 1.0
398
+ output_bert_embeddings .......................... False
399
+ overlap_cpu_optimizer_d2h_h2d ................... False
400
+ overlap_grad_reduce ............................. True
401
+ overlap_moe_expert_parallel_comm ................ False
402
+ overlap_p2p_comm ................................ False
403
+ overlap_p2p_comm_warmup_flush ................... False
404
+ overlap_param_gather ............................ True
405
+ overlap_param_gather_with_optimizer_step ........ False
406
+ override_opt_param_scheduler .................... False
407
+ params_dtype .................................... torch.bfloat16
408
+ patch_dim ....................................... 16
409
+ per_split_data_args_path ........................ None
410
+ perform_initialization .......................... True
411
+ pin_cpu_grads ................................... True
412
+ pin_cpu_params .................................. True
413
+ pipeline_model_parallel_comm_backend ............ None
414
+ pipeline_model_parallel_layout .................. None
415
+ pipeline_model_parallel_size .................... 1
416
+ position_embedding_type ......................... rope
417
+ pretrained_checkpoint ........................... None
418
+ profile ......................................... False
419
+ profile_ranks ................................... [0]
420
+ profile_step_end ................................ 12
421
+ profile_step_start .............................. 10
422
+ q_lora_rank ..................................... None
423
+ qk_head_dim ..................................... 128
424
+ qk_l2_norm ...................................... False
425
+ qk_layernorm .................................... False
426
+ qk_pos_emb_head_dim ............................. 64
427
+ query_in_block_prob ............................. 0.1
428
+ rampup_batch_size ............................... None
429
+ rank ............................................ 0
430
+ recompute_granularity ........................... None
431
+ recompute_method ................................ None
432
+ recompute_modules ............................... None
433
+ recompute_num_layers ............................ None
434
+ record_memory_history ........................... False
435
+ relative_attention_max_distance ................. 128
436
+ relative_attention_num_buckets .................. 32
437
+ replication ..................................... False
438
+ replication_factor .............................. 2
439
+ replication_jump ................................ None
440
+ rerun_mode ...................................... validate_results
441
+ reset_attention_mask ............................ False
442
+ reset_position_ids .............................. False
443
+ result_rejected_tracker_filename ................ None
444
+ retriever_report_topk_accuracies ................ []
445
+ retriever_score_scaling ......................... False
446
+ retriever_seq_length ............................ 256
447
+ retro_add_retriever ............................. False
448
+ retro_attention_gate ............................ 1
449
+ retro_cyclic_train_iters ........................ None
450
+ retro_encoder_attention_dropout ................. 0.1
451
+ retro_encoder_hidden_dropout .................... 0.1
452
+ retro_encoder_layers ............................ 2
453
+ retro_num_neighbors ............................. 2
454
+ retro_num_retrieved_chunks ...................... 2
455
+ retro_project_dir ............................... None
456
+ retro_verify_neighbor_count ..................... True
457
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
458
+ rope_scaling_factor ............................. 8.0
459
+ rope_type ....................................... None
460
+ rotary_base ..................................... 1000000
461
+ rotary_interleaved .............................. False
462
+ rotary_percent .................................. 1.0
463
+ rotary_scaling_factor ........................... 1.0
464
+ rotary_seq_len_interpolation_factor ............. None
465
+ run_workload_inspector_server ................... False
466
+ sample_rate ..................................... 1.0
467
+ save ............................................ None
468
+ save_interval ................................... None
469
+ save_retain_interval ............................ None
470
+ scatter_gather_tensors_in_pipeline .............. True
471
+ seed ............................................ 1234
472
+ seq_length ...................................... 4096
473
+ sequence_parallel ............................... False
474
+ sft ............................................. False
475
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
476
+ sgd_momentum .................................... 0.9
477
+ short_seq_prob .................................. 0.1
478
+ skip_train ...................................... False
479
+ skipped_train_samples ........................... 0
480
+ spec ............................................ None
481
+ split ........................................... 100,0,0
482
+ squared_relu .................................... False
483
+ start_weight_decay .............................. 0.1
484
+ straggler_ctrlr_port ............................ 65535
485
+ straggler_minmax_count .......................... 1
486
+ suggested_communication_unit_size ............... None
487
+ swiglu .......................................... True
488
+ swin_backbone_type .............................. tiny
489
+ symmetric_ar_type ............................... None
490
+ te_rng_tracker .................................. True
491
+ tensor_model_parallel_size ...................... 1
492
+ tensorboard_dir ................................. None
493
+ tensorboard_log_interval ........................ 1
494
+ tensorboard_queue_size .......................... 1000
495
+ test_data_path .................................. None
496
+ test_mode ....................................... False
497
+ tiktoken_num_special_tokens ..................... 1000
498
+ tiktoken_pattern ................................ None
499
+ tiktoken_special_tokens ......................... None
500
+ timing_log_level ................................ 0
501
+ timing_log_option ............................... minmax
502
+ titles_data_path ................................ None
503
+ tokenizer_model ................................. None
504
+ tokenizer_type .................................. GPT2BPETokenizer
505
+ torch_fsdp2_reshard_after_forward ............... True
506
+ tp_comm_bootstrap_backend ....................... nccl
507
+ tp_comm_bulk_dgrad .............................. True
508
+ tp_comm_bulk_wgrad .............................. True
509
+ tp_comm_overlap ................................. False
510
+ tp_comm_overlap_ag .............................. True
511
+ tp_comm_overlap_cfg ............................. None
512
+ tp_comm_overlap_rs .............................. True
513
+ tp_comm_overlap_rs_dgrad ........................ False
514
+ tp_comm_split_ag ................................ True
515
+ tp_comm_split_rs ................................ True
516
+ train_data_path ................................. None
517
+ train_iters ..................................... 12
518
+ train_samples ................................... None
519
+ train_sync_interval ............................. None
520
+ transformer_impl ................................ transformer_engine
521
+ transformer_pipeline_model_parallel_size ........ 1
522
+ untie_embeddings_and_output_weights ............. True
523
+ use_checkpoint_args ............................. False
524
+ use_checkpoint_opt_param_scheduler .............. False
525
+ use_cpu_initialization .......................... None
526
+ use_custom_fsdp ................................. False
527
+ use_dist_ckpt ................................... True
528
+ use_dist_ckpt_deprecated ........................ False
529
+ use_distributed_optimizer ....................... True
530
+ use_flash_attn .................................. False
531
+ use_legacy_models ............................... False
532
+ use_mp_args_from_checkpoint_args ................ False
533
+ use_one_sent_docs ............................... False
534
+ use_persistent_ckpt_worker ...................... False
535
+ use_precision_aware_optimizer ................... False
536
+ use_pytorch_profiler ............................ False
537
+ use_ring_exchange_p2p ........................... False
538
+ use_rope_scaling ................................ False
539
+ use_rotary_position_embeddings .................. False
540
+ use_sharp ....................................... False
541
+ use_tokenizer_model_from_checkpoint_args ........ True
542
+ use_torch_fsdp2 ................................. False
543
+ use_torch_optimizer_for_cpu_offload ............. False
544
+ use_tp_pp_dp_mapping ............................ False
545
+ v_head_dim ...................................... 128
546
+ valid_data_path ................................. None
547
+ variable_seq_lengths ............................ False
548
+ virtual_pipeline_model_parallel_size ............ None
549
+ vision_backbone_type ............................ vit
550
+ vision_pretraining .............................. False
551
+ vision_pretraining_type ......................... classify
552
+ vocab_extra_ids ................................. 0
553
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
554
+ vocab_size ...................................... None
555
+ wandb_exp_name ..................................
556
+ wandb_project ...................................
557
+ wandb_save_dir ..................................
558
+ weight_decay .................................... 0.1
559
+ weight_decay_incr_style ......................... constant
560
+ wgrad_deferral_limit ............................ 0
561
+ world_size ...................................... 8
562
+ yaml_cfg ........................................ None
563
+ -------------------- end of arguments ---------------------
564
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
565
+ > building GPT2BPETokenizer tokenizer ...
566
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
567
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
568
+ > initializing torch distributed ...
569
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
570
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
571
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
572
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
573
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
574
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
575
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
576
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
577
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
578
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
579
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
580
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
581
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
582
+ [Gloo] Rank 0[Gloo] Rank is connected to 7 peer ranks. 5Expected number of connected peer ranks is : is connected to 77 peer ranks.
583
+ Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank 367 is connected to is connected to is connected to 777 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 777
588
+
589
+
590
+ [Gloo] Rank [Gloo] Rank 70[Gloo] Rank is connected to is connected to 477 is connected to peer ranks. peer ranks. 7Expected number of connected peer ranks is : Expected number of connected peer ranks is : peer ranks. 77Expected number of connected peer ranks is :
591
+
592
+ [Gloo] Rank [Gloo] Rank 71 is connected to
593
+ 67 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
594
+ 7
595
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank [Gloo] Rank 5 is connected to 27 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
597
+ 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
604
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
607
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank1]:[W830 07:02:16.555588723 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank3]:[W830 07:02:16.555602797 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank6]:[W830 07:02:16.555618987 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank4]:[W830 07:02:16.555643925 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank5]:[W830 07:02:16.555691044 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank7]:[W830 07:02:16.555716906 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank2]:[W830 07:02:16.557884249 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.035 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:02:16.720540862 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.690 seconds
625
+ time to initialize megatron (seconds): 7.180
626
+ [after megatron is initialized] datetime: 2026-08-30 07:02:22
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6720983040
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (41988096 elements, 41988096 padded size):
670
+ module.decoder.final_layernorm.weight
671
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
672
+ module.decoder.layers.15.mlp.router.weight
673
+ module.decoder.layers.15.self_attention.linear_proj.weight
674
+ module.decoder.layers.15.self_attention.linear_qkv.weight
675
+ module.decoder.layers.15.pre_mlp_layernorm.weight
676
+ Params for bucket 3 (41984000 elements, 41984000 padded size):
677
+ module.decoder.layers.14.self_attention.linear_qkv.weight
678
+ module.decoder.layers.14.pre_mlp_layernorm.weight
679
+ module.decoder.layers.14.mlp.router.weight
680
+ module.decoder.layers.14.self_attention.linear_proj.weight
681
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
682
+ Params for bucket 4 (41984000 elements, 41984000 padded size):
683
+ module.decoder.layers.13.pre_mlp_layernorm.weight
684
+ module.decoder.layers.13.self_attention.linear_qkv.weight
685
+ module.decoder.layers.13.mlp.router.weight
686
+ module.decoder.layers.13.self_attention.linear_proj.weight
687
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
688
+ Params for bucket 5 (41984000 elements, 41984000 padded size):
689
+ module.decoder.layers.12.self_attention.linear_qkv.weight
690
+ module.decoder.layers.12.pre_mlp_layernorm.weight
691
+ module.decoder.layers.12.mlp.router.weight
692
+ module.decoder.layers.12.self_attention.linear_proj.weight
693
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
694
+ Params for bucket 6 (41984000 elements, 41984000 padded size):
695
+ module.decoder.layers.11.pre_mlp_layernorm.weight
696
+ module.decoder.layers.11.self_attention.linear_qkv.weight
697
+ module.decoder.layers.11.mlp.router.weight
698
+ module.decoder.layers.11.self_attention.linear_proj.weight
699
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
700
+ Params for bucket 7 (41984000 elements, 41984000 padded size):
701
+ module.decoder.layers.10.mlp.router.weight
702
+ module.decoder.layers.10.self_attention.linear_proj.weight
703
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
704
+ module.decoder.layers.10.pre_mlp_layernorm.weight
705
+ module.decoder.layers.10.self_attention.linear_qkv.weight
706
+ Params for bucket 8 (41984000 elements, 41984000 padded size):
707
+ module.decoder.layers.9.pre_mlp_layernorm.weight
708
+ module.decoder.layers.9.self_attention.linear_qkv.weight
709
+ module.decoder.layers.9.mlp.router.weight
710
+ module.decoder.layers.9.self_attention.linear_proj.weight
711
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
712
+ Params for bucket 9 (41984000 elements, 41984000 padded size):
713
+ module.decoder.layers.8.mlp.router.weight
714
+ module.decoder.layers.8.self_attention.linear_proj.weight
715
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
716
+ module.decoder.layers.8.pre_mlp_layernorm.weight
717
+ module.decoder.layers.8.self_attention.linear_qkv.weight
718
+ Params for bucket 10 (41984000 elements, 41984000 padded size):
719
+ module.decoder.layers.7.mlp.router.weight
720
+ module.decoder.layers.7.self_attention.linear_proj.weight
721
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
722
+ module.decoder.layers.7.self_attention.linear_qkv.weight
723
+ module.decoder.layers.7.pre_mlp_layernorm.weight
724
+ Params for bucket 11 (41984000 elements, 41984000 padded size):
725
+ module.decoder.layers.6.mlp.router.weight
726
+ module.decoder.layers.6.self_attention.linear_proj.weight
727
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
728
+ module.decoder.layers.6.pre_mlp_layernorm.weight
729
+ module.decoder.layers.6.self_attention.linear_qkv.weight
730
+ Params for bucket 12 (41984000 elements, 41984000 padded size):
731
+ module.decoder.layers.5.mlp.router.weight
732
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
733
+ module.decoder.layers.5.self_attention.linear_qkv.weight
734
+ module.decoder.layers.5.self_attention.linear_proj.weight
735
+ module.decoder.layers.5.pre_mlp_layernorm.weight
736
+ Params for bucket 13 (41984000 elements, 41984000 padded size):
737
+ module.decoder.layers.4.pre_mlp_layernorm.weight
738
+ module.decoder.layers.4.self_attention.linear_qkv.weight
739
+ module.decoder.layers.4.mlp.router.weight
740
+ module.decoder.layers.4.self_attention.linear_proj.weight
741
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
742
+ Params for bucket 14 (41984000 elements, 41984000 padded size):
743
+ module.decoder.layers.3.self_attention.linear_qkv.weight
744
+ module.decoder.layers.3.pre_mlp_layernorm.weight
745
+ module.decoder.layers.3.mlp.router.weight
746
+ module.decoder.layers.3.self_attention.linear_proj.weight
747
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
748
+ Params for bucket 15 (41984000 elements, 41984000 padded size):
749
+ module.decoder.layers.2.pre_mlp_layernorm.weight
750
+ module.decoder.layers.2.self_attention.linear_qkv.weight
751
+ module.decoder.layers.2.mlp.router.weight
752
+ module.decoder.layers.2.self_attention.linear_proj.weight
753
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
754
+ Params for bucket 16 (41984000 elements, 41984000 padded size):
755
+ module.decoder.layers.1.pre_mlp_layernorm.weight
756
+ module.decoder.layers.1.self_attention.linear_qkv.weight
757
+ module.decoder.layers.1.self_attention.linear_proj.weight
758
+ module.decoder.layers.1.mlp.router.weight
759
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
760
+ Params for bucket 17 (41984000 elements, 41984000 padded size):
761
+ module.decoder.layers.0.mlp.router.weight
762
+ module.decoder.layers.0.self_attention.linear_proj.weight
763
+ module.decoder.layers.0.self_attention.linear_qkv.weight
764
+ module.decoder.layers.0.pre_mlp_layernorm.weight
765
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 32
769
+ Params for bucket 1 (58720256 elements, 58720256 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
771
+ Params for bucket 2 (117440512 elements, 117440512 padded size):
772
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
773
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
774
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
775
+ Params for bucket 4 (117440512 elements, 117440512 padded size):
776
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
777
+ Params for bucket 5 (58720256 elements, 58720256 padded size):
778
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
779
+ Params for bucket 6 (117440512 elements, 117440512 padded size):
780
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
781
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
782
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
783
+ Params for bucket 8 (117440512 elements, 117440512 padded size):
784
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
785
+ Params for bucket 9 (58720256 elements, 58720256 padded size):
786
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
787
+ Params for bucket 10 (117440512 elements, 117440512 padded size):
788
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 11 (58720256 elements, 58720256 padded size):
790
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
791
+ Params for bucket 12 (117440512 elements, 117440512 padded size):
792
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
793
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
794
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
795
+ Params for bucket 14 (117440512 elements, 117440512 padded size):
796
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
797
+ Params for bucket 15 (58720256 elements, 58720256 padded size):
798
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
799
+ Params for bucket 16 (117440512 elements, 117440512 padded size):
800
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
801
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
802
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
803
+ Params for bucket 18 (117440512 elements, 117440512 padded size):
804
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
805
+ Params for bucket 19 (58720256 elements, 58720256 padded size):
806
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
807
+ Params for bucket 20 (117440512 elements, 117440512 padded size):
808
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
809
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
811
+ Params for bucket 22 (117440512 elements, 117440512 padded size):
812
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
813
+ Params for bucket 23 (58720256 elements, 58720256 padded size):
814
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
815
+ Params for bucket 24 (117440512 elements, 117440512 padded size):
816
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
817
+ Params for bucket 25 (58720256 elements, 58720256 padded size):
818
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
819
+ Params for bucket 26 (117440512 elements, 117440512 padded size):
820
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
821
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
822
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
823
+ Params for bucket 28 (117440512 elements, 117440512 padded size):
824
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
825
+ Params for bucket 29 (58720256 elements, 58720256 padded size):
826
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
827
+ Params for bucket 30 (117440512 elements, 117440512 padded size):
828
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
829
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
830
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
831
+ Params for bucket 32 (117440512 elements, 117440512 padded size):
832
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
833
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x7003bcaa76e0>, config_logger_dir='')
834
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
835
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:02:25
836
+ > building train, validation, and test datasets ...
837
+ > datasets target sizes (minimum size):
838
+ train: 192
839
+ validation: 0
840
+ test: 0
841
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
842
+ > building train, validation, and test datasets for GPT ...
843
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x7003bce9eae0>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
844
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
845
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
846
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
847
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
848
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
849
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
850
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
851
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
852
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
853
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
854
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
855
+ > finished creating GPT datasets ...
856
+ [after dataloaders are built] datetime: 2026-08-30 07:02:25
857
+ done with setup ...
858
+ training ...
859
+ (min, max) time across ranks (ms):
860
+ model-and-optimizer-setup ......................: (3442.83, 3445.64)
861
+ train/valid/test-data-iterators-setup ..........: (73.94, 91.29)
862
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
863
+ [before the start of training step] datetime: 2026-08-30 07:02:25
864
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
865
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
866
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
867
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
868
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
869
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
870
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
871
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
872
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
873
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
874
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
875
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
876
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
877
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
878
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
879
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
880
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
881
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
882
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
883
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
884
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
885
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
886
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
887
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
888
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
889
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
890
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
891
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
892
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
893
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
894
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
895
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
896
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
897
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
898
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
899
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
900
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
901
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
902
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
903
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
904
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
905
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
906
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
907
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
908
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
909
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
910
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
911
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
912
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
913
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
914
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
915
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
916
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
917
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
918
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
919
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
920
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
921
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
922
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
923
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
924
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
925
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
926
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
927
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
928
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
929
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
930
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
931
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
932
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
933
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
934
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
935
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
936
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
937
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
938
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
939
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
940
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
941
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
942
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
943
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
944
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
945
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
946
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
947
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
948
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
949
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
950
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
951
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
952
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
953
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
954
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
955
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
956
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
957
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
958
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
959
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
960
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
961
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
962
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
963
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
964
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
965
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
966
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
967
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
968
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
969
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
970
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
971
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
972
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
973
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
974
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
975
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
976
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
977
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
978
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
979
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
980
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
981
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
982
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
983
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
985
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
989
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
991
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
992
+ Number of parameters in transformer block in billions: 23.22
993
+ [2026-08-30 07:02:52] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26671.5 | throughput per GPU (TFLOP/s/GPU): 12.5 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.104376E+01 | loss scale: 1.0 | grad norm: 35.708 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |Number of parameters in embedding layers in billions: 0.41
994
+
995
+ Total number of parameters in billions: 23.63
996
+ Number of parameters in most loaded shard in billions: 23.6320
997
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
998
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58364.48876953125 | max allocated: 85653.79150390625 | reserved: 87692.0 | max reserved: 87692.0
999
+ [2026-08-30 07:02:53] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 824.6 | throughput per GPU (TFLOP/s/GPU): 404.3 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.104935E+01 | loss scale: 1.0 | grad norm: 37.623 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1000
+ [2026-08-30 07:02:54] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 867.4 | throughput per GPU (TFLOP/s/GPU): 384.3 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.680600E+00 | loss scale: 1.0 | grad norm: 42.214 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1001
+ Capture CUDA graph for training!!!
1002
+ CUDA graph capture done!!!
1003
+ [2026-08-30 07:02:56] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2003.1 | throughput per GPU (TFLOP/s/GPU): 166.4 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.120139E+01 | loss scale: 1.0 | grad norm: 89.147 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1004
+ [2026-08-30 07:02:56] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 878.3 | throughput per GPU (TFLOP/s/GPU): 379.6 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.016333E+01 | loss scale: 1.0 | grad norm: 15.442 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1005
+ [2026-08-30 07:02:57] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 913.9 | throughput per GPU (TFLOP/s/GPU): 364.8 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.011091E+01 | loss scale: 1.0 | grad norm: 42.624 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1006
+ [2026-08-30 07:02:58] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 901.7 | throughput per GPU (TFLOP/s/GPU): 369.7 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.885600E+00 | loss scale: 1.0 | grad norm: 8.139 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1007
+ [2026-08-30 07:02:59] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 894.7 | throughput per GPU (TFLOP/s/GPU): 372.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.999231E+00 | loss scale: 1.0 | grad norm: 7.416 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1008
+ [2026-08-30 07:03:00] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 895.7 | throughput per GPU (TFLOP/s/GPU): 372.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.664997E+00 | loss scale: 1.0 | grad norm: 6.888 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1009
+ [2026-08-30 07:03:01] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 880.9 | throughput per GPU (TFLOP/s/GPU): 378.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.445807E+00 | loss scale: 1.0 | grad norm: 5.404 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1010
+ [2026-08-30 07:03:02] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 891.2 | throughput per GPU (TFLOP/s/GPU): 374.1 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.367374E+00 | loss scale: 1.0 | grad norm: 4.658 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1011
+ [2026-08-30 07:03:03] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 884.2 | throughput per GPU (TFLOP/s/GPU): 377.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.283400E+00 | loss scale: 1.0 | grad norm: 5.037 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 18432 | min safe capacity: 2.2500 |
1012
+ [after training is done] datetime: 2026-08-30 07:03:03
1013
+ [rank0]:[W830 07:03:04.643244956 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1014
+ [rank4]:[W830 07:03:05.671184824 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1015
+ [rank2]:[W830 07:03:05.772227762 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1016
+ [rank1]:[W830 07:03:05.772393176 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1017
+ [rank5]:[W830 07:03:05.772499799 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1018
+ [rank6]:[W830 07:03:05.132109138 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1019
+ [rank3]:[W830 07:03:06.292900748 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1020
+ [rank7]:[W830 07:03:06.322290155 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1021
+ W0830 07:03:07.044000 696450 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1022
+ W0830 07:03:07.044000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696538 closing signal SIGTERM
1023
+ W0830 07:03:07.045000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696539 closing signal SIGTERM
1024
+ W0830 07:03:07.046000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696540 closing signal SIGTERM
1025
+ W0830 07:03:07.047000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696541 closing signal SIGTERM
1026
+ W0830 07:03:07.048000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696542 closing signal SIGTERM
1027
+ W0830 07:03:07.049000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696543 closing signal SIGTERM
1028
+ W0830 07:03:07.050000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696544 closing signal SIGTERM
1029
+ W0830 07:03:07.052000 696450 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 696545 closing signal SIGTERM
1030
+ Traceback (most recent call last):
1031
+ File "/usr/local/bin/torchrun", line 8, in <module>
1032
+ sys.exit(main())
1033
+ ^^^^^^
1034
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1035
+ return f(*args, **kwargs)
1036
+ ^^^^^^^^^^^^^^^^^^
1037
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1038
+ run(args)
1039
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1040
+ elastic_launch(
1041
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1042
+ return launch_agent(self._config, self._entrypoint, list(args))
1043
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1044
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1045
+ result = agent.run()
1046
+ ^^^^^^^^^^^
1047
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1048
+ result = f(*args, **kwargs)
1049
+ ^^^^^^^^^^^^^^^^^^
1050
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1051
+ result = self._invoke_run(role)
1052
+ ^^^^^^^^^^^^^^^^^^^^^^
1053
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1054
+ time.sleep(monitor_interval)
1055
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1056
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1057
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 696450 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3/mem_samples.csv ADDED
@@ -0,0 +1,232 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 1
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 4
42
+ 1, 629
43
+ 2, 4
44
+ 3, 629
45
+ 4, 4
46
+ 5, 4
47
+ 6, 4
48
+ 7, 4
49
+ 0, 821
50
+ 1, 821
51
+ 2, 821
52
+ 3, 821
53
+ 4, 821
54
+ 5, 821
55
+ 6, 821
56
+ 7, 821
57
+ 0, 1895
58
+ 1, 1895
59
+ 2, 1895
60
+ 3, 1895
61
+ 4, 1895
62
+ 5, 1895
63
+ 6, 1895
64
+ 7, 1895
65
+ 0, 2367
66
+ 1, 2405
67
+ 2, 2375
68
+ 3, 2399
69
+ 4, 2403
70
+ 5, 2393
71
+ 6, 2411
72
+ 7, 2409
73
+ 0, 42451
74
+ 1, 42449
75
+ 2, 42467
76
+ 3, 42433
77
+ 4, 42449
78
+ 5, 42449
79
+ 6, 42467
80
+ 7, 42449
81
+ 0, 43259
82
+ 1, 43257
83
+ 2, 43257
84
+ 3, 43257
85
+ 4, 43257
86
+ 5, 43257
87
+ 6, 43257
88
+ 7, 43257
89
+ 0, 44167
90
+ 1, 44165
91
+ 2, 44165
92
+ 3, 44165
93
+ 4, 44165
94
+ 5, 44165
95
+ 6, 44165
96
+ 7, 44165
97
+ 0, 44167
98
+ 1, 44165
99
+ 2, 44165
100
+ 3, 44165
101
+ 4, 44165
102
+ 5, 44165
103
+ 6, 44165
104
+ 7, 44165
105
+ 0, 45939
106
+ 1, 45937
107
+ 2, 45937
108
+ 3, 45937
109
+ 4, 45937
110
+ 5, 45937
111
+ 6, 45937
112
+ 7, 45937
113
+ 0, 45939
114
+ 1, 45937
115
+ 2, 45937
116
+ 3, 45937
117
+ 4, 45937
118
+ 5, 45937
119
+ 6, 45937
120
+ 7, 45937
121
+ 0, 81289
122
+ 1, 81287
123
+ 2, 81287
124
+ 3, 81287
125
+ 4, 81287
126
+ 5, 81287
127
+ 6, 81287
128
+ 7, 81287
129
+ 0, 82367
130
+ 1, 82365
131
+ 2, 82365
132
+ 3, 82365
133
+ 4, 82365
134
+ 5, 82365
135
+ 6, 82365
136
+ 7, 82365
137
+ 0, 82381
138
+ 1, 82379
139
+ 2, 82379
140
+ 3, 82379
141
+ 4, 82379
142
+ 5, 82379
143
+ 6, 82379
144
+ 7, 82379
145
+ 0, 95479
146
+ 1, 95431
147
+ 2, 95377
148
+ 3, 95457
149
+ 4, 95537
150
+ 5, 95437
151
+ 6, 95413
152
+ 7, 95433
153
+ 0, 96441
154
+ 1, 96337
155
+ 2, 96399
156
+ 3, 96361
157
+ 4, 96439
158
+ 5, 96451
159
+ 6, 96397
160
+ 7, 96411
161
+ 0, 119935
162
+ 1, 120031
163
+ 2, 120087
164
+ 3, 120087
165
+ 4, 119877
166
+ 5, 120087
167
+ 6, 120087
168
+ 7, 120031
169
+ 0, 129221
170
+ 1, 129109
171
+ 2, 129109
172
+ 3, 129109
173
+ 4, 129475
174
+ 5, 129109
175
+ 6, 129109
176
+ 7, 129109
177
+ 0, 129221
178
+ 1, 129109
179
+ 2, 129109
180
+ 3, 129109
181
+ 4, 129475
182
+ 5, 129109
183
+ 6, 129109
184
+ 7, 129109
185
+ 0, 129221
186
+ 1, 129109
187
+ 2, 129109
188
+ 3, 129109
189
+ 4, 129475
190
+ 5, 129109
191
+ 6, 129109
192
+ 7, 129109
193
+ 0, 129221
194
+ 1, 129109
195
+ 2, 129109
196
+ 3, 129109
197
+ 4, 129475
198
+ 5, 129109
199
+ 6, 129109
200
+ 7, 129109
201
+ 0, 129221
202
+ 1, 129109
203
+ 2, 129109
204
+ 3, 129109
205
+ 4, 129475
206
+ 5, 129109
207
+ 6, 129109
208
+ 7, 129109
209
+ 0, 125099
210
+ 1, 124987
211
+ 2, 124987
212
+ 3, 124987
213
+ 4, 125353
214
+ 5, 124987
215
+ 6, 124987
216
+ 7, 124987
217
+ 0, 124065
218
+ 1, 123953
219
+ 2, 123565
220
+ 3, 123953
221
+ 4, 123931
222
+ 5, 123953
223
+ 6, 123565
224
+ 7, 123953
225
+ 0, 129
226
+ 1, 123950
227
+ 2, 641
228
+ 3, 3
229
+ 4, 4
230
+ 5, 4
231
+ 6, 4
232
+ 7, 4
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G1
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=14336
11
+ num_experts=8
12
+ topk=2
13
+ experts_per_rank=1
14
+ expert_params=176160768
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=8192
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=2.40625
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3/train.log ADDED
@@ -0,0 +1,1057 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G1 / ultraep (slots/rank=1 policy=one)
3
+ E=8 K=2 H'=14336 m=1
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=2.40625 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:03:12.559000 699441 torch/distributed/run.py:766]
10
+ W0830 07:03:12.559000 699441 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:03:12.559000 699441 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:03:12.559000 699441 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
18
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
19
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
20
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
21
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
22
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
23
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
24
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
25
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
26
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
27
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
28
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
29
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
30
+ Number of virtual stages per pipeline stage: None
31
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
32
+ using torch.bfloat16 for parameters ...
33
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
34
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
35
+ ------------------------ arguments ------------------------
36
+ account_for_embedding_in_pipeline_split ......... False
37
+ account_for_loss_in_pipeline_split .............. False
38
+ accumulate_allreduce_grads_in_fp32 .............. True
39
+ adam_beta1 ...................................... 0.9
40
+ adam_beta2 ...................................... 0.999
41
+ adam_eps ........................................ 1e-08
42
+ add_bias_linear ................................. False
43
+ add_position_embedding .......................... False
44
+ add_qkv_bias .................................... False
45
+ adlr_autoresume ................................. False
46
+ adlr_autoresume_interval ........................ 1000
47
+ align_grad_reduce ............................... True
48
+ align_param_gather .............................. False
49
+ app_tag_run_name ................................ None
50
+ app_tag_run_version ............................. 0.0.0
51
+ apply_layernorm_1p .............................. False
52
+ apply_query_key_layer_scaling ................... False
53
+ apply_residual_connection_post_layernorm ........ False
54
+ apply_rope_fusion ............................... True
55
+ async_save ...................................... None
56
+ async_tensor_model_parallel_allreduce ........... True
57
+ attention_backend ............................... AttnBackend.auto
58
+ attention_dropout ............................... 0.0
59
+ attention_softmax_in_fp32 ....................... False
60
+ auto_detect_ckpt_format ......................... False
61
+ barrier_with_L1_time ............................ True
62
+ bert_binary_head ................................ True
63
+ bert_embedder_type .............................. megatron
64
+ bert_load ....................................... None
65
+ bf16 ............................................ True
66
+ bias_dropout_fusion ............................. True
67
+ bias_gelu_fusion ................................ False
68
+ bias_swiglu_fusion .............................. True
69
+ biencoder_projection_dim ........................ 0
70
+ biencoder_shared_query_context_model ............ False
71
+ block_data_path ................................. None
72
+ calc_ft_timeouts ................................ False
73
+ calculate_per_token_loss ........................ False
74
+ check_for_large_grads ........................... False
75
+ check_for_nan_in_loss_and_grad .................. False
76
+ check_for_spiky_loss ............................ False
77
+ check_weight_hash_across_dp_replicas_interval ... None
78
+ ckpt_assume_constant_structure .................. False
79
+ ckpt_convert_format ............................. None
80
+ ckpt_convert_save ............................... None
81
+ ckpt_convert_update_legacy_dist_opt_format ...... False
82
+ ckpt_format ..................................... torch_dist
83
+ ckpt_fully_parallel_load ........................ False
84
+ ckpt_fully_parallel_save ........................ True
85
+ ckpt_fully_parallel_save_deprecated ............. False
86
+ ckpt_step ....................................... None
87
+ classes_fraction ................................ 1.0
88
+ clip_grad ....................................... 1.0
89
+ clone_scatter_output_in_embedding ............... True
90
+ config_logger_dir ...............................
91
+ consumed_train_samples .......................... 0
92
+ consumed_valid_samples .......................... 0
93
+ context_parallel_size ........................... 1
94
+ cp_comm_type .................................... ['p2p']
95
+ create_attention_mask_in_dataloader ............. True
96
+ cross_entropy_fusion_impl ....................... native
97
+ cross_entropy_loss_fusion ....................... False
98
+ cuda_graph_scope ................................ full_iteration
99
+ cuda_graph_warmup_steps ......................... 3
100
+ data_args_path .................................. None
101
+ data_cache_path ................................. None
102
+ data_parallel_random_init ....................... False
103
+ data_parallel_sharding_strategy ................. no_shard
104
+ data_parallel_size .............................. 8
105
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
106
+ data_per_class_fraction ......................... 1.0
107
+ data_sharding ................................... True
108
+ dataloader_type ................................. single
109
+ ddp_average_in_collective ....................... False
110
+ ddp_bucket_size ................................. None
111
+ ddp_num_buckets ................................. None
112
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
113
+ decoder_first_pipeline_num_layers ............... None
114
+ decoder_last_pipeline_num_layers ................ None
115
+ decoder_num_layers .............................. None
116
+ decoder_seq_length .............................. None
117
+ decoupled_lr .................................... None
118
+ decoupled_min_lr ................................ None
119
+ decrease_batch_size_if_needed ................... False
120
+ defer_embedding_wgrad_compute ................... False
121
+ delay_wgrad_compute ............................. False
122
+ deprecated_use_mcore_models ..................... True
123
+ deterministic_mode .............................. False
124
+ dino_bottleneck_size ............................ 256
125
+ dino_freeze_last_layer .......................... 1
126
+ dino_head_hidden_size ........................... 2048
127
+ dino_local_crops_number ......................... 10
128
+ dino_local_img_size ............................. 96
129
+ dino_norm_last_layer ............................ False
130
+ dino_teacher_temp ............................... 0.07
131
+ dino_warmup_teacher_temp ........................ 0.04
132
+ dino_warmup_teacher_temp_epochs ................. 30
133
+ disable_bf16_reduced_precision_matmul ........... False
134
+ disable_mamba_mem_eff_path ...................... False
135
+ disable_straggler_on_startup .................... False
136
+ dist_ckpt_format_deprecated ..................... None
137
+ dist_ckpt_strictness ............................ assume_ok_unexpected
138
+ distribute_saved_activations .................... False
139
+ distributed_backend ............................. nccl
140
+ distributed_timeout_minutes ..................... 10
141
+ embedding_init_method_std ....................... None
142
+ embedding_path .................................. None
143
+ empty_unused_memory_level ....................... 0
144
+ enable_cuda_graph ............................... True
145
+ enable_experimental ............................. False
146
+ enable_ft_package ............................... False
147
+ enable_gloo_process_groups ...................... True
148
+ enable_msc ...................................... True
149
+ enable_one_logger ............................... True
150
+ encoder_num_layers .............................. 16
151
+ encoder_seq_length .............................. 4096
152
+ end_weight_decay ................................ 0.1
153
+ eod_mask_loss ................................... False
154
+ error_injection_rate ............................ 0
155
+ error_injection_type ............................ transient_error
156
+ eval_interval ................................... 1000000
157
+ eval_iters ...................................... 0
158
+ evidence_data_path .............................. None
159
+ exit_duration_in_mins ........................... None
160
+ exit_interval ................................... None
161
+ exit_on_missing_checkpoint ...................... False
162
+ exit_signal_handler ............................. False
163
+ exp_avg_dtype ................................... torch.float32
164
+ exp_avg_sq_dtype ................................ torch.float32
165
+ expert_model_parallel_size ...................... 8
166
+ expert_tensor_parallel_size ..................... 1
167
+ export_force_local_attention .................... False
168
+ export_kd_cfg ................................... None
169
+ export_kd_teacher_ckpt_format ................... None
170
+ export_kd_teacher_load .......................... None
171
+ export_kv_cache_quant ........................... False
172
+ export_legacy_megatron .......................... False
173
+ export_model_type ............................... GPTModel
174
+ export_moe_apply_probs_on_input ................. False
175
+ export_qk_l2_norm ............................... False
176
+ export_quant_cfg ................................ None
177
+ export_real_quant_cfg ........................... None
178
+ export_te_mcore_model ........................... False
179
+ external_cuda_graph ............................. False
180
+ ffn_hidden_size ................................. 14336
181
+ finetune ........................................ False
182
+ finetune_data_split ............................. train
183
+ finetune_hf_dataset ............................. None
184
+ first_last_layers_bf16 .......................... False
185
+ flash_decode .................................... False
186
+ fp16 ............................................ False
187
+ fp16_lm_cross_entropy ........................... False
188
+ fp32_residual_connection ........................ False
189
+ fp8 ............................................. e4m3
190
+ fp8_amax_compute_algo ........................... most_recent
191
+ fp8_amax_history_len ............................ 1
192
+ fp8_interval .................................... 1
193
+ fp8_margin ...................................... 0
194
+ fp8_param_gather ................................ False
195
+ fp8_recipe ...................................... mxfp8
196
+ fp8_wgrad ....................................... True
197
+ fsdp_double_buffer .............................. False
198
+ global_batch_size ............................... 16
199
+ grad_reduce_in_bf16 ............................. False
200
+ gradient_accumulation_fusion .................... True
201
+ gradient_reduce_div_fusion ...................... True
202
+ group_query_attention ........................... True
203
+ head_lr_mult .................................... 1.0
204
+ heterogeneous_layers_config_encoded_json ........ None
205
+ heterogeneous_layers_config_path ................ None
206
+ hidden_dropout .................................. 0.0
207
+ hidden_size ..................................... 4096
208
+ hierarchical_context_parallel_sizes ............. None
209
+ high_priority_stream_groups ..................... []
210
+ hybrid_attention_ratio .......................... 0.0
211
+ hybrid_mlp_ratio ................................ 0.0
212
+ hybrid_override_pattern ......................... None
213
+ hysteresis ...................................... 2
214
+ ict_head_size ................................... None
215
+ ict_load ........................................ None
216
+ img_h ........................................... 224
217
+ img_w ........................................... 224
218
+ indexer_batch_size .............................. 128
219
+ indexer_log_interval ............................ 1000
220
+ inference_batch_times_seqlen_threshold .......... -1
221
+ inference_dynamic_batching ...................... False
222
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
223
+ inference_dynamic_batching_buffer_overflow_factor None
224
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
225
+ inference_dynamic_batching_chunk_size ........... 256
226
+ inference_dynamic_batching_max_requests_override None
227
+ inference_dynamic_batching_max_tokens_override .. None
228
+ inference_dynamic_batching_num_cuda_graphs ...... 16
229
+ inference_max_batch_size ........................ 8
230
+ inference_max_seq_length ........................ 2560
231
+ inference_rng_tracker ........................... False
232
+ init_method_std ................................. 0.01
233
+ init_method_xavier_uniform ...................... False
234
+ init_model_with_meta_device ..................... False
235
+ initial_loss_scale .............................. 4294967296
236
+ inprocess_active_world_size ..................... 8
237
+ inprocess_barrier_timeout ....................... 120
238
+ inprocess_completion_timeout .................... 120
239
+ inprocess_empty_cuda_cache ...................... False
240
+ inprocess_granularity ........................... node
241
+ inprocess_hard_timeout .......................... 90
242
+ inprocess_heartbeat_interval .................... 30
243
+ inprocess_heartbeat_timeout ..................... 60
244
+ inprocess_last_call_wait ........................ 1
245
+ inprocess_max_iterations ........................ None
246
+ inprocess_monitor_process_interval .............. 1.0
247
+ inprocess_monitor_thread_interval ............... 1.0
248
+ inprocess_progress_watchdog_interval ............ 1.0
249
+ inprocess_restart ............................... False
250
+ inprocess_soft_timeout .......................... 60
251
+ inprocess_termination_grace_time ................ 1
252
+ is_hybrid_model ................................. False
253
+ iter_per_epoch .................................. 1250
254
+ iterations_to_skip .............................. []
255
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
256
+ kitchen_config_file ............................. None
257
+ kitchen_recipe_number ........................... None
258
+ kv_channels ..................................... 128
259
+ kv_lora_rank .................................... 32
260
+ lazy_mpu_init ................................... None
261
+ load ............................................ None
262
+ load_main_params_from_ckpt ...................... None
263
+ load_model_opt_format ........................... False
264
+ local_rank ...................................... 0
265
+ log_energy ...................................... False
266
+ log_interval .................................... 1
267
+ log_loss_scale_to_tensorboard ................... True
268
+ log_memory_to_tensorboard ....................... False
269
+ log_num_zeros_in_grad ........................... False
270
+ log_params_norm ................................. False
271
+ log_progress .................................... False
272
+ log_straggler ................................... False
273
+ log_throughput .................................. True
274
+ log_timers_to_tensorboard ....................... False
275
+ log_validation_ppl_to_tensorboard ............... False
276
+ log_world_size_to_tensorboard ................... False
277
+ logging_level ................................... None
278
+ loss_scale ...................................... None
279
+ loss_scale_window ............................... 1000
280
+ lr .............................................. 0.0001
281
+ lr_decay_iters .................................. 8
282
+ lr_decay_samples ................................ None
283
+ lr_decay_style .................................. cosine
284
+ lr_warmup_fraction .............................. None
285
+ lr_warmup_init .................................. 0.0
286
+ lr_warmup_iters ................................. 5
287
+ lr_warmup_samples ............................... 0
288
+ lr_wsd_decay_iters .............................. None
289
+ lr_wsd_decay_samples ............................ None
290
+ lr_wsd_decay_style .............................. exponential
291
+ main_grads_dtype ................................ torch.float32
292
+ main_params_dtype ............................... torch.float32
293
+ make_vocab_size_divisible_by .................... 128
294
+ mamba_head_dim .................................. 64
295
+ mamba_num_groups ................................ 8
296
+ mamba_num_heads ................................. None
297
+ mamba_state_dim ................................. 128
298
+ manual_gc ....................................... False
299
+ manual_gc_eval .................................. True
300
+ manual_gc_interval .............................. 0
301
+ mask_factor ..................................... 1.0
302
+ mask_prob ....................................... 0.15
303
+ mask_type ....................................... random
304
+ masked_softmax_fusion ........................... False
305
+ max_position_embeddings ......................... 32768
306
+ max_tokens_to_oom ............................... 12000
307
+ memory_snapshot_path ............................ snapshot.pickle
308
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
309
+ micro_batch_size ................................ 1
310
+ microbatch_group_size_per_vp_stage .............. None
311
+ mid_level_dataset_surplus ....................... 0.005
312
+ min_loss_scale .................................. 1.0
313
+ min_lr .......................................... 1e-05
314
+ mlp_chunks_for_prefill .......................... 1
315
+ mmap_bin_files .................................. True
316
+ mock_data ....................................... False
317
+ moe_apply_probs_on_input ........................ False
318
+ moe_aux_loss_coeff .............................. 0.01
319
+ moe_capacity_monitor ............................ True
320
+ moe_deepep_num_sms .............................. 20
321
+ moe_echo_enable_random_offloading ............... False
322
+ moe_echo_expert_dispatch_overlap ................ False
323
+ moe_echo_expert_dispatcher_type ................. hybridep
324
+ moe_echo_recompute_expert_dispatch .............. False
325
+ moe_enable_deepep ............................... False
326
+ moe_enable_echo ................................. False
327
+ moe_enable_hybridep ............................. True
328
+ moe_enable_ultraep .............................. True
329
+ moe_expert_capacity_factor ...................... None
330
+ moe_extended_tp ................................. False
331
+ moe_ffn_hidden_size ............................. 14336
332
+ moe_grouped_gemm ................................ True
333
+ moe_hybridep_num_sms ............................ 16
334
+ moe_input_jitter_eps ............................ None
335
+ moe_layer_freq .................................. 1
336
+ moe_layer_recompute ............................. False
337
+ moe_num_echo_experts ............................ None
338
+ moe_num_redundant_experts_per_rank .............. 1
339
+ moe_pad_expert_input_to_capacity ................ False
340
+ moe_per_layer_logging ........................... False
341
+ moe_permute_fusion .............................. False
342
+ moe_received_token_capacity ..................... 2.40625
343
+ moe_router_bias_update_rate ..................... 0.001
344
+ moe_router_dtype ................................ fp32
345
+ moe_router_enable_expert_bias ................... False
346
+ moe_router_force_load_balancing ................. False
347
+ moe_router_group_topk ........................... None
348
+ moe_router_load_balancing_type .................. aux_loss
349
+ moe_router_num_groups ........................... None
350
+ moe_router_padding_for_fp8 ...................... False
351
+ moe_router_pre_softmax .......................... False
352
+ moe_router_score_function ....................... softmax
353
+ moe_router_topk ................................. 2
354
+ moe_router_topk_scaling_factor .................. None
355
+ moe_shared_expert_intermediate_size ............. None
356
+ moe_shared_expert_overlap ....................... False
357
+ moe_token_dispatcher_type ....................... flex
358
+ moe_token_drop_policy ........................... probs
359
+ moe_ultraep_max_microbatches .................... None
360
+ moe_ultraep_overlap_weight_sync ................. True
361
+ moe_ultraep_weight_sync_join .................... before_dispatch
362
+ moe_upcycling_granularity ....................... 1
363
+ moe_use_device_initiated_grouped_gemm ........... True
364
+ moe_use_legacy_grouped_gemm ..................... False
365
+ moe_use_upcycling ............................... False
366
+ moe_z_loss_coeff ................................ None
367
+ mrope_section ................................... None
368
+ mscale .......................................... 1.0
369
+ mscale_all_dim .................................. 0.0
370
+ mtp_loss_scaling_factor ......................... 0.1
371
+ mtp_num_layers .................................. None
372
+ multi_latent_attention .......................... False
373
+ nccl_all_reduce_for_prefill ..................... False
374
+ nccl_communicator_config_path ................... None
375
+ nccl_ub ......................................... False
376
+ no_load_optim ................................... None
377
+ no_load_rng ..................................... None
378
+ no_persist_layer_norm ........................... False
379
+ no_rope_freq .................................... None
380
+ no_save_optim ................................... None
381
+ no_save_rng ..................................... None
382
+ non_persistent_ckpt_type ........................ None
383
+ non_persistent_global_ckpt_dir .................. None
384
+ non_persistent_local_ckpt_algo .................. fully_parallel
385
+ non_persistent_local_ckpt_dir ................... None
386
+ non_persistent_save_interval .................... None
387
+ norm_epsilon .................................... 1e-05
388
+ normalization ................................... RMSNorm
389
+ num_attention_heads ............................. 32
390
+ num_channels .................................... 3
391
+ num_classes ..................................... 1000
392
+ num_dataset_builder_threads ..................... 1
393
+ num_distributed_optimizer_instances ............. 1
394
+ num_experts ..................................... 8
395
+ num_layers ...................................... 16
396
+ num_layers_at_end_in_bf16 ....................... 1
397
+ num_layers_at_start_in_bf16 ..................... 1
398
+ num_layers_per_virtual_pipeline_stage ........... None
399
+ num_query_groups ................................ 8
400
+ num_virtual_stages_per_pipeline_rank ............ None
401
+ num_workers ..................................... 2
402
+ object_storage_cache_path ....................... None
403
+ one_logger_async ................................ False
404
+ one_logger_project .............................. megatron-lm
405
+ one_logger_run_name ............................. None
406
+ onnx_safe ....................................... None
407
+ openai_gelu ..................................... False
408
+ optimizer ....................................... adam
409
+ optimizer_cpu_offload ........................... False
410
+ optimizer_offload_fraction ...................... 1.0
411
+ output_bert_embeddings .......................... False
412
+ overlap_cpu_optimizer_d2h_h2d ................... False
413
+ overlap_grad_reduce ............................. True
414
+ overlap_moe_expert_parallel_comm ................ False
415
+ overlap_p2p_comm ................................ False
416
+ overlap_p2p_comm_warmup_flush ................... False
417
+ overlap_param_gather ............................ True
418
+ overlap_param_gather_with_optimizer_step ........ False
419
+ override_opt_param_scheduler .................... False
420
+ params_dtype .................................... torch.bfloat16
421
+ patch_dim ....................................... 16
422
+ per_split_data_args_path ........................ None
423
+ perform_initialization .......................... True
424
+ pin_cpu_grads ................................... True
425
+ pin_cpu_params .................................. True
426
+ pipeline_model_parallel_comm_backend ............ None
427
+ pipeline_model_parallel_layout .................. None
428
+ pipeline_model_parallel_size .................... 1
429
+ position_embedding_type ......................... rope
430
+ pretrained_checkpoint ........................... None
431
+ profile ......................................... False
432
+ profile_ranks ................................... [0]
433
+ profile_step_end ................................ 12
434
+ profile_step_start .............................. 10
435
+ q_lora_rank ..................................... None
436
+ qk_head_dim ..................................... 128
437
+ qk_l2_norm ...................................... False
438
+ qk_layernorm .................................... False
439
+ qk_pos_emb_head_dim ............................. 64
440
+ query_in_block_prob ............................. 0.1
441
+ rampup_batch_size ............................... None
442
+ rank ............................................ 0
443
+ recompute_granularity ........................... None
444
+ recompute_method ................................ None
445
+ recompute_modules ............................... None
446
+ recompute_num_layers ............................ None
447
+ record_memory_history ........................... False
448
+ relative_attention_max_distance ................. 128
449
+ relative_attention_num_buckets .................. 32
450
+ replication ..................................... False
451
+ replication_factor .............................. 2
452
+ replication_jump ................................ None
453
+ rerun_mode ...................................... validate_results
454
+ reset_attention_mask ............................ False
455
+ reset_position_ids .............................. False
456
+ result_rejected_tracker_filename ................ None
457
+ retriever_report_topk_accuracies ................ []
458
+ retriever_score_scaling ......................... False
459
+ retriever_seq_length ............................ 256
460
+ retro_add_retriever ............................. False
461
+ retro_attention_gate ............................ 1
462
+ retro_cyclic_train_iters ........................ None
463
+ retro_encoder_attention_dropout ................. 0.1
464
+ retro_encoder_hidden_dropout .................... 0.1
465
+ retro_encoder_layers ............................ 2
466
+ retro_num_neighbors ............................. 2
467
+ retro_num_retrieved_chunks ...................... 2
468
+ retro_project_dir ............................... None
469
+ retro_verify_neighbor_count ..................... True
470
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
471
+ rope_scaling_factor ............................. 8.0
472
+ rope_type ....................................... None
473
+ rotary_base ..................................... 1000000
474
+ rotary_interleaved .............................. False
475
+ rotary_percent .................................. 1.0
476
+ rotary_scaling_factor ........................... 1.0
477
+ rotary_seq_len_interpolation_factor ............. None
478
+ run_workload_inspector_server ................... False
479
+ sample_rate ..................................... 1.0
480
+ save ............................................ None
481
+ save_interval ................................... None
482
+ save_retain_interval ............................ None
483
+ scatter_gather_tensors_in_pipeline .............. True
484
+ seed ............................................ 1234
485
+ seq_length ...................................... 4096
486
+ sequence_parallel ............................... False
487
+ sft ............................................. False
488
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
489
+ sgd_momentum .................................... 0.9
490
+ short_seq_prob .................................. 0.1
491
+ skip_train ...................................... False
492
+ skipped_train_samples ........................... 0
493
+ spec ............................................ None
494
+ split ........................................... 100,0,0
495
+ squared_relu .................................... False
496
+ start_weight_decay .............................. 0.1
497
+ straggler_ctrlr_port ............................ 65535
498
+ straggler_minmax_count .......................... 1
499
+ suggested_communication_unit_size ............... None
500
+ swiglu .......................................... True
501
+ swin_backbone_type .............................. tiny
502
+ symmetric_ar_type ............................... None
503
+ te_rng_tracker .................................. True
504
+ tensor_model_parallel_size ...................... 1
505
+ tensorboard_dir ................................. None
506
+ tensorboard_log_interval ........................ 1
507
+ tensorboard_queue_size .......................... 1000
508
+ test_data_path .................................. None
509
+ test_mode ....................................... False
510
+ tiktoken_num_special_tokens ..................... 1000
511
+ tiktoken_pattern ................................ None
512
+ tiktoken_special_tokens ......................... None
513
+ timing_log_level ................................ 0
514
+ timing_log_option ............................... minmax
515
+ titles_data_path ................................ None
516
+ tokenizer_model ................................. None
517
+ tokenizer_type .................................. GPT2BPETokenizer
518
+ torch_fsdp2_reshard_after_forward ............... True
519
+ tp_comm_bootstrap_backend ....................... nccl
520
+ tp_comm_bulk_dgrad .............................. True
521
+ tp_comm_bulk_wgrad .............................. True
522
+ tp_comm_overlap ................................. False
523
+ tp_comm_overlap_ag .............................. True
524
+ tp_comm_overlap_cfg ............................. None
525
+ tp_comm_overlap_rs .............................. True
526
+ tp_comm_overlap_rs_dgrad ........................ False
527
+ tp_comm_split_ag ................................ True
528
+ tp_comm_split_rs ................................ True
529
+ train_data_path ................................. None
530
+ train_iters ..................................... 12
531
+ train_samples ................................... None
532
+ train_sync_interval ............................. None
533
+ transformer_impl ................................ transformer_engine
534
+ transformer_pipeline_model_parallel_size ........ 1
535
+ untie_embeddings_and_output_weights ............. True
536
+ use_checkpoint_args ............................. False
537
+ use_checkpoint_opt_param_scheduler .............. False
538
+ use_cpu_initialization .......................... None
539
+ use_custom_fsdp ................................. False
540
+ use_dist_ckpt ................................... True
541
+ use_dist_ckpt_deprecated ........................ False
542
+ use_distributed_optimizer ....................... True
543
+ use_flash_attn .................................. False
544
+ use_legacy_models ............................... False
545
+ use_mp_args_from_checkpoint_args ................ False
546
+ use_one_sent_docs ............................... False
547
+ use_persistent_ckpt_worker ...................... False
548
+ use_precision_aware_optimizer ................... False
549
+ use_pytorch_profiler ............................ False
550
+ use_ring_exchange_p2p ........................... False
551
+ use_rope_scaling ................................ False
552
+ use_rotary_position_embeddings .................. False
553
+ use_sharp ....................................... False
554
+ use_tokenizer_model_from_checkpoint_args ........ True
555
+ use_torch_fsdp2 ................................. False
556
+ use_torch_optimizer_for_cpu_offload ............. False
557
+ use_tp_pp_dp_mapping ............................ False
558
+ v_head_dim ...................................... 128
559
+ valid_data_path ................................. None
560
+ variable_seq_lengths ............................ False
561
+ virtual_pipeline_model_parallel_size ............ None
562
+ vision_backbone_type ............................ vit
563
+ vision_pretraining .............................. False
564
+ vision_pretraining_type ......................... classify
565
+ vocab_extra_ids ................................. 0
566
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
567
+ vocab_size ...................................... None
568
+ wandb_exp_name ..................................
569
+ wandb_project ...................................
570
+ wandb_save_dir ..................................
571
+ weight_decay .................................... 0.1
572
+ weight_decay_incr_style ......................... constant
573
+ wgrad_deferral_limit ............................ 0
574
+ world_size ...................................... 8
575
+ yaml_cfg ........................................ None
576
+ -------------------- end of arguments ---------------------
577
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
578
+ > building GPT2BPETokenizer tokenizer ...
579
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
580
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
581
+ > initializing torch distributed ...
582
+ [Gloo] Rank 5 is connected to [Gloo] Rank 7 peer ranks. 0Expected number of connected peer ranks is : is connected to 77 peer ranks.
583
+ Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank [Gloo] Rank 64 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77
589
+
590
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank [Gloo] Rank 7 is connected to 67 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
597
+ 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank [Gloo] Rank 0[Gloo] Rank 0 is connected to 0 is connected to 0 is connected to 0 peer ranks. 0 peer ranks. Expected number of connected peer ranks is : peer ranks. Expected number of connected peer ranks is : 0Expected number of connected peer ranks is : 0
603
+
604
+ 0
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
607
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank4]:[W830 07:03:30.613739351 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank7]:[W830 07:03:30.613762883 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank1]:[W830 07:03:30.613808415 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank5]:[W830 07:03:30.613867957 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank3]:[W830 07:03:30.613913132 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank2]:[W830 07:03:30.613931913 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank6]:[W830 07:03:30.615567371 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.034 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:03:30.772747659 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.980 seconds
625
+ time to initialize megatron (seconds): 7.383
626
+ [after megatron is initialized] datetime: 2026-08-30 07:03:36
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6720983040
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (41988096 elements, 41988096 padded size):
670
+ module.decoder.layers.15.self_attention.linear_qkv.weight
671
+ module.decoder.layers.15.pre_mlp_layernorm.weight
672
+ module.decoder.final_layernorm.weight
673
+ module.decoder.layers.15.mlp.router.weight
674
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
675
+ module.decoder.layers.15.self_attention.linear_proj.weight
676
+ Params for bucket 3 (41984000 elements, 41984000 padded size):
677
+ module.decoder.layers.14.mlp.router.weight
678
+ module.decoder.layers.14.self_attention.linear_proj.weight
679
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
680
+ module.decoder.layers.14.self_attention.linear_qkv.weight
681
+ module.decoder.layers.14.pre_mlp_layernorm.weight
682
+ Params for bucket 4 (41984000 elements, 41984000 padded size):
683
+ module.decoder.layers.13.mlp.router.weight
684
+ module.decoder.layers.13.self_attention.linear_proj.weight
685
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
686
+ module.decoder.layers.13.pre_mlp_layernorm.weight
687
+ module.decoder.layers.13.self_attention.linear_qkv.weight
688
+ Params for bucket 5 (41984000 elements, 41984000 padded size):
689
+ module.decoder.layers.12.mlp.router.weight
690
+ module.decoder.layers.12.self_attention.linear_proj.weight
691
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
692
+ module.decoder.layers.12.self_attention.linear_qkv.weight
693
+ module.decoder.layers.12.pre_mlp_layernorm.weight
694
+ Params for bucket 6 (41984000 elements, 41984000 padded size):
695
+ module.decoder.layers.11.mlp.router.weight
696
+ module.decoder.layers.11.self_attention.linear_proj.weight
697
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.11.pre_mlp_layernorm.weight
699
+ module.decoder.layers.11.self_attention.linear_qkv.weight
700
+ Params for bucket 7 (41984000 elements, 41984000 padded size):
701
+ module.decoder.layers.10.mlp.router.weight
702
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
703
+ module.decoder.layers.10.self_attention.linear_proj.weight
704
+ module.decoder.layers.10.self_attention.linear_qkv.weight
705
+ module.decoder.layers.10.pre_mlp_layernorm.weight
706
+ Params for bucket 8 (41984000 elements, 41984000 padded size):
707
+ module.decoder.layers.9.self_attention.linear_qkv.weight
708
+ module.decoder.layers.9.pre_mlp_layernorm.weight
709
+ module.decoder.layers.9.mlp.router.weight
710
+ module.decoder.layers.9.self_attention.linear_proj.weight
711
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
712
+ Params for bucket 9 (41984000 elements, 41984000 padded size):
713
+ module.decoder.layers.8.pre_mlp_layernorm.weight
714
+ module.decoder.layers.8.self_attention.linear_qkv.weight
715
+ module.decoder.layers.8.mlp.router.weight
716
+ module.decoder.layers.8.self_attention.linear_proj.weight
717
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
718
+ Params for bucket 10 (41984000 elements, 41984000 padded size):
719
+ module.decoder.layers.7.self_attention.linear_qkv.weight
720
+ module.decoder.layers.7.pre_mlp_layernorm.weight
721
+ module.decoder.layers.7.mlp.router.weight
722
+ module.decoder.layers.7.self_attention.linear_proj.weight
723
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
724
+ Params for bucket 11 (41984000 elements, 41984000 padded size):
725
+ module.decoder.layers.6.pre_mlp_layernorm.weight
726
+ module.decoder.layers.6.self_attention.linear_qkv.weight
727
+ module.decoder.layers.6.mlp.router.weight
728
+ module.decoder.layers.6.self_attention.linear_proj.weight
729
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
730
+ Params for bucket 12 (41984000 elements, 41984000 padded size):
731
+ module.decoder.layers.5.self_attention.linear_qkv.weight
732
+ module.decoder.layers.5.self_attention.linear_proj.weight
733
+ module.decoder.layers.5.pre_mlp_layernorm.weight
734
+ module.decoder.layers.5.mlp.router.weight
735
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
736
+ Params for bucket 13 (41984000 elements, 41984000 padded size):
737
+ module.decoder.layers.4.mlp.router.weight
738
+ module.decoder.layers.4.self_attention.linear_proj.weight
739
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
740
+ module.decoder.layers.4.pre_mlp_layernorm.weight
741
+ module.decoder.layers.4.self_attention.linear_qkv.weight
742
+ Params for bucket 14 (41984000 elements, 41984000 padded size):
743
+ module.decoder.layers.3.self_attention.linear_qkv.weight
744
+ module.decoder.layers.3.pre_mlp_layernorm.weight
745
+ module.decoder.layers.3.mlp.router.weight
746
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
747
+ module.decoder.layers.3.self_attention.linear_proj.weight
748
+ Params for bucket 15 (41984000 elements, 41984000 padded size):
749
+ module.decoder.layers.2.self_attention.linear_qkv.weight
750
+ module.decoder.layers.2.pre_mlp_layernorm.weight
751
+ module.decoder.layers.2.mlp.router.weight
752
+ module.decoder.layers.2.self_attention.linear_proj.weight
753
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
754
+ Params for bucket 16 (41984000 elements, 41984000 padded size):
755
+ module.decoder.layers.1.self_attention.linear_qkv.weight
756
+ module.decoder.layers.1.mlp.router.weight
757
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
758
+ module.decoder.layers.1.pre_mlp_layernorm.weight
759
+ module.decoder.layers.1.self_attention.linear_proj.weight
760
+ Params for bucket 17 (41984000 elements, 41984000 padded size):
761
+ module.decoder.layers.0.mlp.router.weight
762
+ module.decoder.layers.0.self_attention.linear_proj.weight
763
+ module.decoder.layers.0.pre_mlp_layernorm.weight
764
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
765
+ module.decoder.layers.0.self_attention.linear_qkv.weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 32
769
+ Params for bucket 1 (58720256 elements, 58720256 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
771
+ Params for bucket 2 (117440512 elements, 117440512 padded size):
772
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
773
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
774
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
775
+ Params for bucket 4 (117440512 elements, 117440512 padded size):
776
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
777
+ Params for bucket 5 (58720256 elements, 58720256 padded size):
778
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
779
+ Params for bucket 6 (117440512 elements, 117440512 padded size):
780
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
781
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
782
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
783
+ Params for bucket 8 (117440512 elements, 117440512 padded size):
784
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
785
+ Params for bucket 9 (58720256 elements, 58720256 padded size):
786
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
787
+ Params for bucket 10 (117440512 elements, 117440512 padded size):
788
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 11 (58720256 elements, 58720256 padded size):
790
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
791
+ Params for bucket 12 (117440512 elements, 117440512 padded size):
792
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
793
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
794
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
795
+ Params for bucket 14 (117440512 elements, 117440512 padded size):
796
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
797
+ Params for bucket 15 (58720256 elements, 58720256 padded size):
798
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
799
+ Params for bucket 16 (117440512 elements, 117440512 padded size):
800
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
801
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
802
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
803
+ Params for bucket 18 (117440512 elements, 117440512 padded size):
804
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
805
+ Params for bucket 19 (58720256 elements, 58720256 padded size):
806
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
807
+ Params for bucket 20 (117440512 elements, 117440512 padded size):
808
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
809
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
811
+ Params for bucket 22 (117440512 elements, 117440512 padded size):
812
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
813
+ Params for bucket 23 (58720256 elements, 58720256 padded size):
814
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
815
+ Params for bucket 24 (117440512 elements, 117440512 padded size):
816
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
817
+ Params for bucket 25 (58720256 elements, 58720256 padded size):
818
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
819
+ Params for bucket 26 (117440512 elements, 117440512 padded size):
820
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
821
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
822
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
823
+ Params for bucket 28 (117440512 elements, 117440512 padded size):
824
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
825
+ Params for bucket 29 (58720256 elements, 58720256 padded size):
826
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
827
+ Params for bucket 30 (117440512 elements, 117440512 padded size):
828
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
829
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
830
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
831
+ Params for bucket 32 (117440512 elements, 117440512 padded size):
832
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
833
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x7cc895f490a0>, config_logger_dir='')
834
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
835
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:03:39
836
+ > building train, validation, and test datasets ...
837
+ > datasets target sizes (minimum size):
838
+ train: 192
839
+ validation: 0
840
+ test: 0
841
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
842
+ > building train, validation, and test datasets for GPT ...
843
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x7cc8960cfef0>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
844
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
845
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
846
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
847
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
848
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
849
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
850
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
851
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
852
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
853
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
854
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
855
+ > finished creating GPT datasets ...
856
+ [after dataloaders are built] datetime: 2026-08-30 07:03:39
857
+ done with setup ...
858
+ (min, max) time across ranks (ms):
859
+ model-and-optimizer-setup ......................: (3338.95, 3341.18)
860
+ train/valid/test-data-iterators-setup ..........: (75.05, 90.50)
861
+ training ...
862
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
863
+ [before the start of training step] datetime: 2026-08-30 07:03:39
864
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
865
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
866
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
867
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
868
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
869
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
870
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
871
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
872
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
873
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
874
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
875
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
876
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
877
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
878
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
879
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
880
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
881
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
882
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
883
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
884
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
885
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
886
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
887
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
888
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
889
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
890
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
891
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
892
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
893
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
894
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
895
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
896
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
897
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
898
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
899
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
900
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
901
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
902
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
903
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
904
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
905
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
906
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
907
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
908
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
909
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
910
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
911
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
912
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
913
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
914
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
915
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
916
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
917
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
918
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
919
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
920
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
921
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
922
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
923
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
924
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
925
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
926
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
927
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
928
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
929
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
930
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
931
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
932
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
933
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
934
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
935
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
936
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
937
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
938
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
939
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
940
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
941
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
942
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
943
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
944
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
945
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
946
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
947
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
948
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
949
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
950
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
951
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
952
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
953
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
954
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
955
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
956
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
957
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
958
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
959
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
960
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
961
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
962
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
963
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
964
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
965
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
966
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
967
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
968
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
969
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
970
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
971
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
972
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
973
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
974
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
975
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
976
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
977
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
978
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
979
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
980
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
981
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
982
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
983
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
985
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
989
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
991
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
992
+ Number of parameters in transformer block in billions: 23.22 [2026-08-30 07:04:05] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 25916.4 | throughput per GPU (TFLOP/s/GPU): 12.9 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.104376E+01 | loss scale: 1.0 | grad norm: 35.708 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
993
+
994
+ Number of parameters in embedding layers in billions: 0.41
995
+ Total number of parameters in billions: 23.63
996
+ Number of parameters in most loaded shard in billions: 23.6320
997
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
998
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58364.48876953125 | max allocated: 85653.79150390625 | reserved: 87692.0 | max reserved: 87692.0
999
+ [2026-08-30 07:04:06] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 820.2 | throughput per GPU (TFLOP/s/GPU): 406.4 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.104935E+01 | loss scale: 1.0 | grad norm: 37.626 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1000
+ [2026-08-30 07:04:07] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 867.8 | throughput per GPU (TFLOP/s/GPU): 384.2 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.680878E+00 | loss scale: 1.0 | grad norm: 42.175 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1001
+ Capture CUDA graph for training!!!
1002
+ CUDA graph capture done!!!
1003
+ [2026-08-30 07:04:09] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1850.8 | throughput per GPU (TFLOP/s/GPU): 180.1 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.117627E+01 | loss scale: 1.0 | grad norm: 88.260 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1004
+ [2026-08-30 07:04:10] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 877.3 | throughput per GPU (TFLOP/s/GPU): 380.0 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.016106E+01 | loss scale: 1.0 | grad norm: 15.274 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1005
+ [2026-08-30 07:04:11] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 906.5 | throughput per GPU (TFLOP/s/GPU): 367.8 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.027267E+01 | loss scale: 1.0 | grad norm: 48.118 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1006
+ [2026-08-30 07:04:12] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 896.4 | throughput per GPU (TFLOP/s/GPU): 371.9 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.905370E+00 | loss scale: 1.0 | grad norm: 8.158 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1007
+ [2026-08-30 07:04:13] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 894.0 | throughput per GPU (TFLOP/s/GPU): 372.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.990004E+00 | loss scale: 1.0 | grad norm: 6.785 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1008
+ [2026-08-30 07:04:13] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 889.7 | throughput per GPU (TFLOP/s/GPU): 374.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.678252E+00 | loss scale: 1.0 | grad norm: 6.735 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1009
+ [2026-08-30 07:04:14] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 872.8 | throughput per GPU (TFLOP/s/GPU): 382.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.479605E+00 | loss scale: 1.0 | grad norm: 5.572 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1010
+ [2026-08-30 07:04:15] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 883.6 | throughput per GPU (TFLOP/s/GPU): 377.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.383426E+00 | loss scale: 1.0 | grad norm: 5.079 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1011
+ [2026-08-30 07:04:16] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 885.6 | throughput per GPU (TFLOP/s/GPU): 376.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.282817E+00 | loss scale: 1.0 | grad norm: 4.803 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 13312 | min safe capacity: 1.6250 |
1012
+ [after training is done] datetime: 2026-08-30 07:04:16
1013
+ [rank0]:[W830 07:04:17.006097053 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1014
+ [rank4]:[W830 07:04:19.366106405 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1015
+ [rank2]:[W830 07:04:19.432924280 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1016
+ [rank3]:[W830 07:04:19.436010710 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1017
+ [rank6]:[W830 07:04:19.436652988 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1018
+ [rank5]:[W830 07:04:19.517278663 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1019
+ [rank7]:[W830 07:04:19.519961480 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1020
+ [rank1]:[W830 07:04:19.734759612 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1021
+ W0830 07:04:21.242000 699441 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1022
+ W0830 07:04:21.243000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699529 closing signal SIGTERM
1023
+ W0830 07:04:21.244000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699530 closing signal SIGTERM
1024
+ W0830 07:04:21.245000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699531 closing signal SIGTERM
1025
+ W0830 07:04:21.246000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699532 closing signal SIGTERM
1026
+ W0830 07:04:21.246000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699533 closing signal SIGTERM
1027
+ W0830 07:04:21.247000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699534 closing signal SIGTERM
1028
+ W0830 07:04:21.248000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699535 closing signal SIGTERM
1029
+ W0830 07:04:21.249000 699441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 699536 closing signal SIGTERM
1030
+ Traceback (most recent call last):
1031
+ File "/usr/local/bin/torchrun", line 8, in <module>
1032
+ sys.exit(main())
1033
+ ^^^^^^
1034
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1035
+ return f(*args, **kwargs)
1036
+ ^^^^^^^^^^^^^^^^^^
1037
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1038
+ run(args)
1039
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1040
+ elastic_launch(
1041
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1042
+ return launch_agent(self._config, self._entrypoint, list(args))
1043
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1044
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1045
+ result = agent.run()
1046
+ ^^^^^^^^^^^
1047
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1048
+ result = f(*args, **kwargs)
1049
+ ^^^^^^^^^^^^^^^^^^
1050
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1051
+ result = self._invoke_run(role)
1052
+ ^^^^^^^^^^^^^^^^^^^^^^
1053
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1054
+ time.sleep(monitor_interval)
1055
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1056
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1057
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 699441 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1/mem_samples.csv ADDED
@@ -0,0 +1,224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 2
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 4
42
+ 1, 4
43
+ 2, 4
44
+ 3, 4
45
+ 4, 4
46
+ 5, 4
47
+ 6, 4
48
+ 7, 4
49
+ 0, 821
50
+ 1, 821
51
+ 2, 821
52
+ 3, 819
53
+ 4, 819
54
+ 5, 821
55
+ 6, 819
56
+ 7, 821
57
+ 0, 1537
58
+ 1, 1537
59
+ 2, 1537
60
+ 3, 1537
61
+ 4, 1537
62
+ 5, 1537
63
+ 6, 1537
64
+ 7, 1537
65
+ 0, 2231
66
+ 1, 2231
67
+ 2, 2231
68
+ 3, 2231
69
+ 4, 2231
70
+ 5, 2231
71
+ 6, 2231
72
+ 7, 2231
73
+ 0, 2903
74
+ 1, 2903
75
+ 2, 2903
76
+ 3, 2903
77
+ 4, 2903
78
+ 5, 2903
79
+ 6, 2903
80
+ 7, 2903
81
+ 0, 41173
82
+ 1, 41177
83
+ 2, 41155
84
+ 3, 41165
85
+ 4, 41165
86
+ 5, 41169
87
+ 6, 41187
88
+ 7, 41201
89
+ 0, 41975
90
+ 1, 41961
91
+ 2, 41951
92
+ 3, 41961
93
+ 4, 41983
94
+ 5, 41989
95
+ 6, 41977
96
+ 7, 42015
97
+ 0, 42679
98
+ 1, 42671
99
+ 2, 42671
100
+ 3, 42671
101
+ 4, 42671
102
+ 5, 42671
103
+ 6, 42671
104
+ 7, 42671
105
+ 0, 42957
106
+ 1, 42949
107
+ 2, 42949
108
+ 3, 42949
109
+ 4, 42949
110
+ 5, 42949
111
+ 6, 42949
112
+ 7, 42949
113
+ 0, 44381
114
+ 1, 44373
115
+ 2, 44373
116
+ 3, 44373
117
+ 4, 44373
118
+ 5, 44373
119
+ 6, 44373
120
+ 7, 44373
121
+ 0, 58389
122
+ 1, 58363
123
+ 2, 58365
124
+ 3, 58365
125
+ 4, 58379
126
+ 5, 58367
127
+ 6, 58365
128
+ 7, 58367
129
+ 0, 65637
130
+ 1, 65615
131
+ 2, 65615
132
+ 3, 65615
133
+ 4, 65629
134
+ 5, 65615
135
+ 6, 65615
136
+ 7, 65615
137
+ 0, 66123
138
+ 1, 66101
139
+ 2, 66101
140
+ 3, 66101
141
+ 4, 66115
142
+ 5, 66101
143
+ 6, 66101
144
+ 7, 66101
145
+ 0, 74013
146
+ 1, 74013
147
+ 2, 74027
148
+ 3, 74047
149
+ 4, 74053
150
+ 5, 74025
151
+ 6, 74059
152
+ 7, 74013
153
+ 0, 74863
154
+ 1, 74839
155
+ 2, 74839
156
+ 3, 74839
157
+ 4, 74855
158
+ 5, 74839
159
+ 6, 74839
160
+ 7, 74839
161
+ 0, 76307
162
+ 1, 76291
163
+ 2, 76309
164
+ 3, 76315
165
+ 4, 76229
166
+ 5, 76309
167
+ 6, 76321
168
+ 7, 76341
169
+ 0, 74665
170
+ 1, 82161
171
+ 2, 78283
172
+ 3, 95135
173
+ 4, 74089
174
+ 5, 83939
175
+ 6, 75579
176
+ 7, 95133
177
+ 0, 102689
178
+ 1, 102733
179
+ 2, 102733
180
+ 3, 102733
181
+ 4, 102627
182
+ 5, 102733
183
+ 6, 102733
184
+ 7, 102733
185
+ 0, 102689
186
+ 1, 102733
187
+ 2, 102733
188
+ 3, 102733
189
+ 4, 102627
190
+ 5, 102733
191
+ 6, 102733
192
+ 7, 102733
193
+ 0, 102161
194
+ 1, 102467
195
+ 2, 102205
196
+ 3, 102147
197
+ 4, 102297
198
+ 5, 102141
199
+ 6, 101885
200
+ 7, 102467
201
+ 0, 98561
202
+ 1, 98605
203
+ 2, 98669
204
+ 3, 99251
205
+ 4, 98819
206
+ 5, 98669
207
+ 6, 99251
208
+ 7, 99313
209
+ 0, 96493
210
+ 1, 96925
211
+ 2, 96925
212
+ 3, 96925
213
+ 4, 96819
214
+ 5, 96989
215
+ 6, 97571
216
+ 7, 97245
217
+ 0, 4
218
+ 1, 4
219
+ 2, 4
220
+ 3, 96537
221
+ 4, 4
222
+ 5, 4
223
+ 6, 4
224
+ 7, 4
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G2
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=3584
11
+ num_experts=32
12
+ topk=8
13
+ experts_per_rank=4
14
+ expert_params=44040192
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=32768
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.08203125
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1/train.log ADDED
@@ -0,0 +1,1170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G2 / ultraep (slots/rank=1 policy=one)
3
+ E=32 K=8 H'=3584 m=4
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.08203125 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:04:26.747000 702460 torch/distributed/run.py:766]
10
+ W0830 07:04:26.747000 702460 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:04:26.747000 702460 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:04:26.747000 702460 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
18
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
19
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
20
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
21
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
22
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
23
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
24
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
25
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
26
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
27
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
28
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
29
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
30
+ Number of virtual stages per pipeline stage: None
31
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
32
+ using torch.bfloat16 for parameters ...
33
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
34
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
35
+ ------------------------ arguments ------------------------
36
+ account_for_embedding_in_pipeline_split ......... False
37
+ account_for_loss_in_pipeline_split .............. False
38
+ accumulate_allreduce_grads_in_fp32 .............. True
39
+ adam_beta1 ...................................... 0.9
40
+ adam_beta2 ...................................... 0.999
41
+ adam_eps ........................................ 1e-08
42
+ add_bias_linear ................................. False
43
+ add_position_embedding .......................... False
44
+ add_qkv_bias .................................... False
45
+ adlr_autoresume ................................. False
46
+ adlr_autoresume_interval ........................ 1000
47
+ align_grad_reduce ............................... True
48
+ align_param_gather .............................. False
49
+ app_tag_run_name ................................ None
50
+ app_tag_run_version ............................. 0.0.0
51
+ apply_layernorm_1p .............................. False
52
+ apply_query_key_layer_scaling ................... False
53
+ apply_residual_connection_post_layernorm ........ False
54
+ apply_rope_fusion ............................... True
55
+ async_save ...................................... None
56
+ async_tensor_model_parallel_allreduce ........... True
57
+ attention_backend ............................... AttnBackend.auto
58
+ attention_dropout ............................... 0.0
59
+ attention_softmax_in_fp32 ....................... False
60
+ auto_detect_ckpt_format ......................... False
61
+ barrier_with_L1_time ............................ True
62
+ bert_binary_head ................................ True
63
+ bert_embedder_type .............................. megatron
64
+ bert_load ....................................... None
65
+ bf16 ............................................ True
66
+ bias_dropout_fusion ............................. True
67
+ bias_gelu_fusion ................................ False
68
+ bias_swiglu_fusion .............................. True
69
+ biencoder_projection_dim ........................ 0
70
+ biencoder_shared_query_context_model ............ False
71
+ block_data_path ................................. None
72
+ calc_ft_timeouts ................................ False
73
+ calculate_per_token_loss ........................ False
74
+ check_for_large_grads ........................... False
75
+ check_for_nan_in_loss_and_grad .................. False
76
+ check_for_spiky_loss ............................ False
77
+ check_weight_hash_across_dp_replicas_interval ... None
78
+ ckpt_assume_constant_structure .................. False
79
+ ckpt_convert_format ............................. None
80
+ ckpt_convert_save ............................... None
81
+ ckpt_convert_update_legacy_dist_opt_format ...... False
82
+ ckpt_format ..................................... torch_dist
83
+ ckpt_fully_parallel_load ........................ False
84
+ ckpt_fully_parallel_save ........................ True
85
+ ckpt_fully_parallel_save_deprecated ............. False
86
+ ckpt_step ....................................... None
87
+ classes_fraction ................................ 1.0
88
+ clip_grad ....................................... 1.0
89
+ clone_scatter_output_in_embedding ............... True
90
+ config_logger_dir ...............................
91
+ consumed_train_samples .......................... 0
92
+ consumed_valid_samples .......................... 0
93
+ context_parallel_size ........................... 1
94
+ cp_comm_type .................................... ['p2p']
95
+ create_attention_mask_in_dataloader ............. True
96
+ cross_entropy_fusion_impl ....................... native
97
+ cross_entropy_loss_fusion ....................... False
98
+ cuda_graph_scope ................................ full_iteration
99
+ cuda_graph_warmup_steps ......................... 3
100
+ data_args_path .................................. None
101
+ data_cache_path ................................. None
102
+ data_parallel_random_init ....................... False
103
+ data_parallel_sharding_strategy ................. no_shard
104
+ data_parallel_size .............................. 8
105
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
106
+ data_per_class_fraction ......................... 1.0
107
+ data_sharding ................................... True
108
+ dataloader_type ................................. single
109
+ ddp_average_in_collective ....................... False
110
+ ddp_bucket_size ................................. None
111
+ ddp_num_buckets ................................. None
112
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
113
+ decoder_first_pipeline_num_layers ............... None
114
+ decoder_last_pipeline_num_layers ................ None
115
+ decoder_num_layers .............................. None
116
+ decoder_seq_length .............................. None
117
+ decoupled_lr .................................... None
118
+ decoupled_min_lr ................................ None
119
+ decrease_batch_size_if_needed ................... False
120
+ defer_embedding_wgrad_compute ................... False
121
+ delay_wgrad_compute ............................. False
122
+ deprecated_use_mcore_models ..................... True
123
+ deterministic_mode .............................. False
124
+ dino_bottleneck_size ............................ 256
125
+ dino_freeze_last_layer .......................... 1
126
+ dino_head_hidden_size ........................... 2048
127
+ dino_local_crops_number ......................... 10
128
+ dino_local_img_size ............................. 96
129
+ dino_norm_last_layer ............................ False
130
+ dino_teacher_temp ............................... 0.07
131
+ dino_warmup_teacher_temp ........................ 0.04
132
+ dino_warmup_teacher_temp_epochs ................. 30
133
+ disable_bf16_reduced_precision_matmul ........... False
134
+ disable_mamba_mem_eff_path ...................... False
135
+ disable_straggler_on_startup .................... False
136
+ dist_ckpt_format_deprecated ..................... None
137
+ dist_ckpt_strictness ............................ assume_ok_unexpected
138
+ distribute_saved_activations .................... False
139
+ distributed_backend ............................. nccl
140
+ distributed_timeout_minutes ..................... 10
141
+ embedding_init_method_std ....................... None
142
+ embedding_path .................................. None
143
+ empty_unused_memory_level ....................... 0
144
+ enable_cuda_graph ............................... True
145
+ enable_experimental ............................. False
146
+ enable_ft_package ............................... False
147
+ enable_gloo_process_groups ...................... True
148
+ enable_msc ...................................... True
149
+ enable_one_logger ............................... True
150
+ encoder_num_layers .............................. 16
151
+ encoder_seq_length .............................. 4096
152
+ end_weight_decay ................................ 0.1
153
+ eod_mask_loss ................................... False
154
+ error_injection_rate ............................ 0
155
+ error_injection_type ............................ transient_error
156
+ eval_interval ................................... 1000000
157
+ eval_iters ...................................... 0
158
+ evidence_data_path .............................. None
159
+ exit_duration_in_mins ........................... None
160
+ exit_interval ................................... None
161
+ exit_on_missing_checkpoint ...................... False
162
+ exit_signal_handler ............................. False
163
+ exp_avg_dtype ................................... torch.float32
164
+ exp_avg_sq_dtype ................................ torch.float32
165
+ expert_model_parallel_size ...................... 8
166
+ expert_tensor_parallel_size ..................... 1
167
+ export_force_local_attention .................... False
168
+ export_kd_cfg ................................... None
169
+ export_kd_teacher_ckpt_format ................... None
170
+ export_kd_teacher_load .......................... None
171
+ export_kv_cache_quant ........................... False
172
+ export_legacy_megatron .......................... False
173
+ export_model_type ............................... GPTModel
174
+ export_moe_apply_probs_on_input ................. False
175
+ export_qk_l2_norm ............................... False
176
+ export_quant_cfg ................................ None
177
+ export_real_quant_cfg ........................... None
178
+ export_te_mcore_model ........................... False
179
+ external_cuda_graph ............................. False
180
+ ffn_hidden_size ................................. 14336
181
+ finetune ........................................ False
182
+ finetune_data_split ............................. train
183
+ finetune_hf_dataset ............................. None
184
+ first_last_layers_bf16 .......................... False
185
+ flash_decode .................................... False
186
+ fp16 ............................................ False
187
+ fp16_lm_cross_entropy ........................... False
188
+ fp32_residual_connection ........................ False
189
+ fp8 ............................................. e4m3
190
+ fp8_amax_compute_algo ........................... most_recent
191
+ fp8_amax_history_len ............................ 1
192
+ fp8_interval .................................... 1
193
+ fp8_margin ...................................... 0
194
+ fp8_param_gather ................................ False
195
+ fp8_recipe ...................................... mxfp8
196
+ fp8_wgrad ....................................... True
197
+ fsdp_double_buffer .............................. False
198
+ global_batch_size ............................... 16
199
+ grad_reduce_in_bf16 ............................. False
200
+ gradient_accumulation_fusion .................... True
201
+ gradient_reduce_div_fusion ...................... True
202
+ group_query_attention ........................... True
203
+ head_lr_mult .................................... 1.0
204
+ heterogeneous_layers_config_encoded_json ........ None
205
+ heterogeneous_layers_config_path ................ None
206
+ hidden_dropout .................................. 0.0
207
+ hidden_size ..................................... 4096
208
+ hierarchical_context_parallel_sizes ............. None
209
+ high_priority_stream_groups ..................... []
210
+ hybrid_attention_ratio .......................... 0.0
211
+ hybrid_mlp_ratio ................................ 0.0
212
+ hybrid_override_pattern ......................... None
213
+ hysteresis ...................................... 2
214
+ ict_head_size ................................... None
215
+ ict_load ........................................ None
216
+ img_h ........................................... 224
217
+ img_w ........................................... 224
218
+ indexer_batch_size .............................. 128
219
+ indexer_log_interval ............................ 1000
220
+ inference_batch_times_seqlen_threshold .......... -1
221
+ inference_dynamic_batching ...................... False
222
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
223
+ inference_dynamic_batching_buffer_overflow_factor None
224
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
225
+ inference_dynamic_batching_chunk_size ........... 256
226
+ inference_dynamic_batching_max_requests_override None
227
+ inference_dynamic_batching_max_tokens_override .. None
228
+ inference_dynamic_batching_num_cuda_graphs ...... 16
229
+ inference_max_batch_size ........................ 8
230
+ inference_max_seq_length ........................ 2560
231
+ inference_rng_tracker ........................... False
232
+ init_method_std ................................. 0.01
233
+ init_method_xavier_uniform ...................... False
234
+ init_model_with_meta_device ..................... False
235
+ initial_loss_scale .............................. 4294967296
236
+ inprocess_active_world_size ..................... 8
237
+ inprocess_barrier_timeout ....................... 120
238
+ inprocess_completion_timeout .................... 120
239
+ inprocess_empty_cuda_cache ...................... False
240
+ inprocess_granularity ........................... node
241
+ inprocess_hard_timeout .......................... 90
242
+ inprocess_heartbeat_interval .................... 30
243
+ inprocess_heartbeat_timeout ..................... 60
244
+ inprocess_last_call_wait ........................ 1
245
+ inprocess_max_iterations ........................ None
246
+ inprocess_monitor_process_interval .............. 1.0
247
+ inprocess_monitor_thread_interval ............... 1.0
248
+ inprocess_progress_watchdog_interval ............ 1.0
249
+ inprocess_restart ............................... False
250
+ inprocess_soft_timeout .......................... 60
251
+ inprocess_termination_grace_time ................ 1
252
+ is_hybrid_model ................................. False
253
+ iter_per_epoch .................................. 1250
254
+ iterations_to_skip .............................. []
255
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
256
+ kitchen_config_file ............................. None
257
+ kitchen_recipe_number ........................... None
258
+ kv_channels ..................................... 128
259
+ kv_lora_rank .................................... 32
260
+ lazy_mpu_init ................................... None
261
+ load ............................................ None
262
+ load_main_params_from_ckpt ...................... None
263
+ load_model_opt_format ........................... False
264
+ local_rank ...................................... 0
265
+ log_energy ...................................... False
266
+ log_interval .................................... 1
267
+ log_loss_scale_to_tensorboard ................... True
268
+ log_memory_to_tensorboard ....................... False
269
+ log_num_zeros_in_grad ........................... False
270
+ log_params_norm ................................. False
271
+ log_progress .................................... False
272
+ log_straggler ................................... False
273
+ log_throughput .................................. True
274
+ log_timers_to_tensorboard ....................... False
275
+ log_validation_ppl_to_tensorboard ............... False
276
+ log_world_size_to_tensorboard ................... False
277
+ logging_level ................................... None
278
+ loss_scale ...................................... None
279
+ loss_scale_window ............................... 1000
280
+ lr .............................................. 0.0001
281
+ lr_decay_iters .................................. 8
282
+ lr_decay_samples ................................ None
283
+ lr_decay_style .................................. cosine
284
+ lr_warmup_fraction .............................. None
285
+ lr_warmup_init .................................. 0.0
286
+ lr_warmup_iters ................................. 5
287
+ lr_warmup_samples ............................... 0
288
+ lr_wsd_decay_iters .............................. None
289
+ lr_wsd_decay_samples ............................ None
290
+ lr_wsd_decay_style .............................. exponential
291
+ main_grads_dtype ................................ torch.float32
292
+ main_params_dtype ............................... torch.float32
293
+ make_vocab_size_divisible_by .................... 128
294
+ mamba_head_dim .................................. 64
295
+ mamba_num_groups ................................ 8
296
+ mamba_num_heads ................................. None
297
+ mamba_state_dim ................................. 128
298
+ manual_gc ....................................... False
299
+ manual_gc_eval .................................. True
300
+ manual_gc_interval .............................. 0
301
+ mask_factor ..................................... 1.0
302
+ mask_prob ....................................... 0.15
303
+ mask_type ....................................... random
304
+ masked_softmax_fusion ........................... False
305
+ max_position_embeddings ......................... 32768
306
+ max_tokens_to_oom ............................... 12000
307
+ memory_snapshot_path ............................ snapshot.pickle
308
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
309
+ micro_batch_size ................................ 1
310
+ microbatch_group_size_per_vp_stage .............. None
311
+ mid_level_dataset_surplus ....................... 0.005
312
+ min_loss_scale .................................. 1.0
313
+ min_lr .......................................... 1e-05
314
+ mlp_chunks_for_prefill .......................... 1
315
+ mmap_bin_files .................................. True
316
+ mock_data ....................................... False
317
+ moe_apply_probs_on_input ........................ False
318
+ moe_aux_loss_coeff .............................. 0.01
319
+ moe_capacity_monitor ............................ True
320
+ moe_deepep_num_sms .............................. 20
321
+ moe_echo_enable_random_offloading ............... False
322
+ moe_echo_expert_dispatch_overlap ................ False
323
+ moe_echo_expert_dispatcher_type ................. hybridep
324
+ moe_echo_recompute_expert_dispatch .............. False
325
+ moe_enable_deepep ............................... False
326
+ moe_enable_echo ................................. False
327
+ moe_enable_hybridep ............................. True
328
+ moe_enable_ultraep .............................. True
329
+ moe_expert_capacity_factor ...................... None
330
+ moe_extended_tp ................................. False
331
+ moe_ffn_hidden_size ............................. 3584
332
+ moe_grouped_gemm ................................ True
333
+ moe_hybridep_num_sms ............................ 16
334
+ moe_input_jitter_eps ............................ None
335
+ moe_layer_freq .................................. 1
336
+ moe_layer_recompute ............................. False
337
+ moe_num_echo_experts ............................ None
338
+ moe_num_redundant_experts_per_rank .............. 1
339
+ moe_pad_expert_input_to_capacity ................ False
340
+ moe_per_layer_logging ........................... False
341
+ moe_permute_fusion .............................. False
342
+ moe_received_token_capacity ..................... 1.08203125
343
+ moe_router_bias_update_rate ..................... 0.001
344
+ moe_router_dtype ................................ fp32
345
+ moe_router_enable_expert_bias ................... False
346
+ moe_router_force_load_balancing ................. False
347
+ moe_router_group_topk ........................... None
348
+ moe_router_load_balancing_type .................. aux_loss
349
+ moe_router_num_groups ........................... None
350
+ moe_router_padding_for_fp8 ...................... False
351
+ moe_router_pre_softmax .......................... False
352
+ moe_router_score_function ....................... softmax
353
+ moe_router_topk ................................. 8
354
+ moe_router_topk_scaling_factor .................. None
355
+ moe_shared_expert_intermediate_size ............. None
356
+ moe_shared_expert_overlap ....................... False
357
+ moe_token_dispatcher_type ....................... flex
358
+ moe_token_drop_policy ........................... probs
359
+ moe_ultraep_max_microbatches .................... None
360
+ moe_ultraep_overlap_weight_sync ................. True
361
+ moe_ultraep_weight_sync_join .................... before_dispatch
362
+ moe_upcycling_granularity ....................... 1
363
+ moe_use_device_initiated_grouped_gemm ........... True
364
+ moe_use_legacy_grouped_gemm ..................... False
365
+ moe_use_upcycling ............................... False
366
+ moe_z_loss_coeff ................................ None
367
+ mrope_section ................................... None
368
+ mscale .......................................... 1.0
369
+ mscale_all_dim .................................. 0.0
370
+ mtp_loss_scaling_factor ......................... 0.1
371
+ mtp_num_layers .................................. None
372
+ multi_latent_attention .......................... False
373
+ nccl_all_reduce_for_prefill ..................... False
374
+ nccl_communicator_config_path ................... None
375
+ nccl_ub ......................................... False
376
+ no_load_optim ................................... None
377
+ no_load_rng ..................................... None
378
+ no_persist_layer_norm ........................... False
379
+ no_rope_freq .................................... None
380
+ no_save_optim ................................... None
381
+ no_save_rng ..................................... None
382
+ non_persistent_ckpt_type ........................ None
383
+ non_persistent_global_ckpt_dir .................. None
384
+ non_persistent_local_ckpt_algo .................. fully_parallel
385
+ non_persistent_local_ckpt_dir ................... None
386
+ non_persistent_save_interval .................... None
387
+ norm_epsilon .................................... 1e-05
388
+ normalization ................................... RMSNorm
389
+ num_attention_heads ............................. 32
390
+ num_channels .................................... 3
391
+ num_classes ..................................... 1000
392
+ num_dataset_builder_threads ..................... 1
393
+ num_distributed_optimizer_instances ............. 1
394
+ num_experts ..................................... 32
395
+ num_layers ...................................... 16
396
+ num_layers_at_end_in_bf16 ....................... 1
397
+ num_layers_at_start_in_bf16 ..................... 1
398
+ num_layers_per_virtual_pipeline_stage ........... None
399
+ num_query_groups ................................ 8
400
+ num_virtual_stages_per_pipeline_rank ............ None
401
+ num_workers ..................................... 2
402
+ object_storage_cache_path ....................... None
403
+ one_logger_async ................................ False
404
+ one_logger_project .............................. megatron-lm
405
+ one_logger_run_name ............................. None
406
+ onnx_safe ....................................... None
407
+ openai_gelu ..................................... False
408
+ optimizer ....................................... adam
409
+ optimizer_cpu_offload ........................... False
410
+ optimizer_offload_fraction ...................... 1.0
411
+ output_bert_embeddings .......................... False
412
+ overlap_cpu_optimizer_d2h_h2d ................... False
413
+ overlap_grad_reduce ............................. True
414
+ overlap_moe_expert_parallel_comm ................ False
415
+ overlap_p2p_comm ................................ False
416
+ overlap_p2p_comm_warmup_flush ................... False
417
+ overlap_param_gather ............................ True
418
+ overlap_param_gather_with_optimizer_step ........ False
419
+ override_opt_param_scheduler .................... False
420
+ params_dtype .................................... torch.bfloat16
421
+ patch_dim ....................................... 16
422
+ per_split_data_args_path ........................ None
423
+ perform_initialization .......................... True
424
+ pin_cpu_grads ................................... True
425
+ pin_cpu_params .................................. True
426
+ pipeline_model_parallel_comm_backend ............ None
427
+ pipeline_model_parallel_layout .................. None
428
+ pipeline_model_parallel_size .................... 1
429
+ position_embedding_type ......................... rope
430
+ pretrained_checkpoint ........................... None
431
+ profile ......................................... False
432
+ profile_ranks ................................... [0]
433
+ profile_step_end ................................ 12
434
+ profile_step_start .............................. 10
435
+ q_lora_rank ..................................... None
436
+ qk_head_dim ..................................... 128
437
+ qk_l2_norm ...................................... False
438
+ qk_layernorm .................................... False
439
+ qk_pos_emb_head_dim ............................. 64
440
+ query_in_block_prob ............................. 0.1
441
+ rampup_batch_size ............................... None
442
+ rank ............................................ 0
443
+ recompute_granularity ........................... None
444
+ recompute_method ................................ None
445
+ recompute_modules ............................... None
446
+ recompute_num_layers ............................ None
447
+ record_memory_history ........................... False
448
+ relative_attention_max_distance ................. 128
449
+ relative_attention_num_buckets .................. 32
450
+ replication ..................................... False
451
+ replication_factor .............................. 2
452
+ replication_jump ................................ None
453
+ rerun_mode ...................................... validate_results
454
+ reset_attention_mask ............................ False
455
+ reset_position_ids .............................. False
456
+ result_rejected_tracker_filename ................ None
457
+ retriever_report_topk_accuracies ................ []
458
+ retriever_score_scaling ......................... False
459
+ retriever_seq_length ............................ 256
460
+ retro_add_retriever ............................. False
461
+ retro_attention_gate ............................ 1
462
+ retro_cyclic_train_iters ........................ None
463
+ retro_encoder_attention_dropout ................. 0.1
464
+ retro_encoder_hidden_dropout .................... 0.1
465
+ retro_encoder_layers ............................ 2
466
+ retro_num_neighbors ............................. 2
467
+ retro_num_retrieved_chunks ...................... 2
468
+ retro_project_dir ............................... None
469
+ retro_verify_neighbor_count ..................... True
470
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
471
+ rope_scaling_factor ............................. 8.0
472
+ rope_type ....................................... None
473
+ rotary_base ..................................... 1000000
474
+ rotary_interleaved .............................. False
475
+ rotary_percent .................................. 1.0
476
+ rotary_scaling_factor ........................... 1.0
477
+ rotary_seq_len_interpolation_factor ............. None
478
+ run_workload_inspector_server ................... False
479
+ sample_rate ..................................... 1.0
480
+ save ............................................ None
481
+ save_interval ................................... None
482
+ save_retain_interval ............................ None
483
+ scatter_gather_tensors_in_pipeline .............. True
484
+ seed ............................................ 1234
485
+ seq_length ...................................... 4096
486
+ sequence_parallel ............................... False
487
+ sft ............................................. False
488
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
489
+ sgd_momentum .................................... 0.9
490
+ short_seq_prob .................................. 0.1
491
+ skip_train ...................................... False
492
+ skipped_train_samples ........................... 0
493
+ spec ............................................ None
494
+ split ........................................... 100,0,0
495
+ squared_relu .................................... False
496
+ start_weight_decay .............................. 0.1
497
+ straggler_ctrlr_port ............................ 65535
498
+ straggler_minmax_count .......................... 1
499
+ suggested_communication_unit_size ............... None
500
+ swiglu .......................................... True
501
+ swin_backbone_type .............................. tiny
502
+ symmetric_ar_type ............................... None
503
+ te_rng_tracker .................................. True
504
+ tensor_model_parallel_size ...................... 1
505
+ tensorboard_dir ................................. None
506
+ tensorboard_log_interval ........................ 1
507
+ tensorboard_queue_size .......................... 1000
508
+ test_data_path .................................. None
509
+ test_mode ....................................... False
510
+ tiktoken_num_special_tokens ..................... 1000
511
+ tiktoken_pattern ................................ None
512
+ tiktoken_special_tokens ......................... None
513
+ timing_log_level ................................ 0
514
+ timing_log_option ............................... minmax
515
+ titles_data_path ................................ None
516
+ tokenizer_model ................................. None
517
+ tokenizer_type .................................. GPT2BPETokenizer
518
+ torch_fsdp2_reshard_after_forward ............... True
519
+ tp_comm_bootstrap_backend ....................... nccl
520
+ tp_comm_bulk_dgrad .............................. True
521
+ tp_comm_bulk_wgrad .............................. True
522
+ tp_comm_overlap ................................. False
523
+ tp_comm_overlap_ag .............................. True
524
+ tp_comm_overlap_cfg ............................. None
525
+ tp_comm_overlap_rs .............................. True
526
+ tp_comm_overlap_rs_dgrad ........................ False
527
+ tp_comm_split_ag ................................ True
528
+ tp_comm_split_rs ................................ True
529
+ train_data_path ................................. None
530
+ train_iters ..................................... 12
531
+ train_samples ................................... None
532
+ train_sync_interval ............................. None
533
+ transformer_impl ................................ transformer_engine
534
+ transformer_pipeline_model_parallel_size ........ 1
535
+ untie_embeddings_and_output_weights ............. True
536
+ use_checkpoint_args ............................. False
537
+ use_checkpoint_opt_param_scheduler .............. False
538
+ use_cpu_initialization .......................... None
539
+ use_custom_fsdp ................................. False
540
+ use_dist_ckpt ................................... True
541
+ use_dist_ckpt_deprecated ........................ False
542
+ use_distributed_optimizer ....................... True
543
+ use_flash_attn .................................. False
544
+ use_legacy_models ............................... False
545
+ use_mp_args_from_checkpoint_args ................ False
546
+ use_one_sent_docs ............................... False
547
+ use_persistent_ckpt_worker ...................... False
548
+ use_precision_aware_optimizer ................... False
549
+ use_pytorch_profiler ............................ False
550
+ use_ring_exchange_p2p ........................... False
551
+ use_rope_scaling ................................ False
552
+ use_rotary_position_embeddings .................. False
553
+ use_sharp ....................................... False
554
+ use_tokenizer_model_from_checkpoint_args ........ True
555
+ use_torch_fsdp2 ................................. False
556
+ use_torch_optimizer_for_cpu_offload ............. False
557
+ use_tp_pp_dp_mapping ............................ False
558
+ v_head_dim ...................................... 128
559
+ valid_data_path ................................. None
560
+ variable_seq_lengths ............................ False
561
+ virtual_pipeline_model_parallel_size ............ None
562
+ vision_backbone_type ............................ vit
563
+ vision_pretraining .............................. False
564
+ vision_pretraining_type ......................... classify
565
+ vocab_extra_ids ................................. 0
566
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
567
+ vocab_size ...................................... None
568
+ wandb_exp_name ..................................
569
+ wandb_project ...................................
570
+ wandb_save_dir ..................................
571
+ weight_decay .................................... 0.1
572
+ weight_decay_incr_style ......................... constant
573
+ wgrad_deferral_limit ............................ 0
574
+ world_size ...................................... 8
575
+ yaml_cfg ........................................ None
576
+ -------------------- end of arguments ---------------------
577
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
578
+ > building GPT2BPETokenizer tokenizer ...
579
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
580
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
581
+ > initializing torch distributed ...
582
+ [Gloo] Rank 1 is connected to [Gloo] Rank 7 peer ranks. 0Expected number of connected peer ranks is : is connected to 77 peer ranks.
583
+ Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank 3657 is connected to is connected to is connected to is connected to 7777 peer ranks. peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 7777
585
+
586
+
587
+
588
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank [Gloo] Rank 6 is connected to 77 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
597
+ 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank [Gloo] Rank 00 is connected to is connected to [Gloo] Rank 000 peer ranks. peer ranks. is connected to Expected number of connected peer ranks is : Expected number of connected peer ranks is : 000
603
+ peer ranks.
604
+ Expected number of connected peer ranks is : 0
605
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank 000 is connected to is connected to is connected to 000 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 000
606
+
607
+
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank3]:[W830 07:04:44.563399012 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank7]:[W830 07:04:44.563700145 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank1]:[W830 07:04:44.563745704 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank4]:[W830 07:04:44.563767888 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank5]:[W830 07:04:44.563821226 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank6]:[W830 07:04:44.568422457 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank2]:[W830 07:04:44.568427701 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.033 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:04:44.735733659 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.837 seconds
625
+ time to initialize megatron (seconds): 7.217
626
+ [after megatron is initialized] datetime: 2026-08-30 07:04:50
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 4608626688
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (42086400 elements, 42086400 padded size):
670
+ module.decoder.layers.15.self_attention.linear_qkv.weight
671
+ module.decoder.layers.15.pre_mlp_layernorm.weight
672
+ module.decoder.final_layernorm.weight
673
+ module.decoder.layers.15.mlp.router.weight
674
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
675
+ module.decoder.layers.15.self_attention.linear_proj.weight
676
+ Params for bucket 3 (42082304 elements, 42082304 padded size):
677
+ module.decoder.layers.14.self_attention.linear_qkv.weight
678
+ module.decoder.layers.14.pre_mlp_layernorm.weight
679
+ module.decoder.layers.14.mlp.router.weight
680
+ module.decoder.layers.14.self_attention.linear_proj.weight
681
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
682
+ Params for bucket 4 (42082304 elements, 42082304 padded size):
683
+ module.decoder.layers.13.mlp.router.weight
684
+ module.decoder.layers.13.self_attention.linear_proj.weight
685
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
686
+ module.decoder.layers.13.pre_mlp_layernorm.weight
687
+ module.decoder.layers.13.self_attention.linear_qkv.weight
688
+ Params for bucket 5 (42082304 elements, 42082304 padded size):
689
+ module.decoder.layers.12.mlp.router.weight
690
+ module.decoder.layers.12.self_attention.linear_proj.weight
691
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
692
+ module.decoder.layers.12.pre_mlp_layernorm.weight
693
+ module.decoder.layers.12.self_attention.linear_qkv.weight
694
+ Params for bucket 6 (42082304 elements, 42082304 padded size):
695
+ module.decoder.layers.11.mlp.router.weight
696
+ module.decoder.layers.11.self_attention.linear_proj.weight
697
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.11.self_attention.linear_qkv.weight
699
+ module.decoder.layers.11.pre_mlp_layernorm.weight
700
+ Params for bucket 7 (42082304 elements, 42082304 padded size):
701
+ module.decoder.layers.10.mlp.router.weight
702
+ module.decoder.layers.10.self_attention.linear_proj.weight
703
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
704
+ module.decoder.layers.10.self_attention.linear_qkv.weight
705
+ module.decoder.layers.10.pre_mlp_layernorm.weight
706
+ Params for bucket 8 (42082304 elements, 42082304 padded size):
707
+ module.decoder.layers.9.mlp.router.weight
708
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
709
+ module.decoder.layers.9.self_attention.linear_proj.weight
710
+ module.decoder.layers.9.self_attention.linear_qkv.weight
711
+ module.decoder.layers.9.pre_mlp_layernorm.weight
712
+ Params for bucket 9 (42082304 elements, 42082304 padded size):
713
+ module.decoder.layers.8.mlp.router.weight
714
+ module.decoder.layers.8.self_attention.linear_proj.weight
715
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
716
+ module.decoder.layers.8.self_attention.linear_qkv.weight
717
+ module.decoder.layers.8.pre_mlp_layernorm.weight
718
+ Params for bucket 10 (42082304 elements, 42082304 padded size):
719
+ module.decoder.layers.7.self_attention.linear_qkv.weight
720
+ module.decoder.layers.7.pre_mlp_layernorm.weight
721
+ module.decoder.layers.7.mlp.router.weight
722
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
723
+ module.decoder.layers.7.self_attention.linear_proj.weight
724
+ Params for bucket 11 (42082304 elements, 42082304 padded size):
725
+ module.decoder.layers.6.self_attention.linear_qkv.weight
726
+ module.decoder.layers.6.pre_mlp_layernorm.weight
727
+ module.decoder.layers.6.mlp.router.weight
728
+ module.decoder.layers.6.self_attention.linear_proj.weight
729
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
730
+ Params for bucket 12 (42082304 elements, 42082304 padded size):
731
+ module.decoder.layers.5.pre_mlp_layernorm.weight
732
+ module.decoder.layers.5.self_attention.linear_proj.weight
733
+ module.decoder.layers.5.self_attention.linear_qkv.weight
734
+ module.decoder.layers.5.mlp.router.weight
735
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
736
+ Params for bucket 13 (42082304 elements, 42082304 padded size):
737
+ module.decoder.layers.4.mlp.router.weight
738
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
739
+ module.decoder.layers.4.self_attention.linear_proj.weight
740
+ module.decoder.layers.4.self_attention.linear_qkv.weight
741
+ module.decoder.layers.4.pre_mlp_layernorm.weight
742
+ Params for bucket 14 (42082304 elements, 42082304 padded size):
743
+ module.decoder.layers.3.mlp.router.weight
744
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
745
+ module.decoder.layers.3.self_attention.linear_proj.weight
746
+ module.decoder.layers.3.self_attention.linear_qkv.weight
747
+ module.decoder.layers.3.pre_mlp_layernorm.weight
748
+ Params for bucket 15 (42082304 elements, 42082304 padded size):
749
+ module.decoder.layers.2.mlp.router.weight
750
+ module.decoder.layers.2.self_attention.linear_proj.weight
751
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
752
+ module.decoder.layers.2.pre_mlp_layernorm.weight
753
+ module.decoder.layers.2.self_attention.linear_qkv.weight
754
+ Params for bucket 16 (42082304 elements, 42082304 padded size):
755
+ module.decoder.layers.1.self_attention.linear_proj.weight
756
+ module.decoder.layers.1.pre_mlp_layernorm.weight
757
+ module.decoder.layers.1.self_attention.linear_qkv.weight
758
+ module.decoder.layers.1.mlp.router.weight
759
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
760
+ Params for bucket 17 (42082304 elements, 42082304 padded size):
761
+ module.decoder.layers.0.pre_mlp_layernorm.weight
762
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
763
+ module.decoder.layers.0.self_attention.linear_proj.weight
764
+ module.decoder.layers.0.mlp.router.weight
765
+ module.decoder.layers.0.self_attention.linear_qkv.weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56
769
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
771
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
772
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
773
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
774
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
775
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
776
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
777
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
778
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
779
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
780
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
781
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
782
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
783
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
784
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
785
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
786
+ Params for bucket 6 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
788
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
789
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
790
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
791
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
792
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
793
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
794
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
795
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
796
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
797
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
798
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
799
+ Params for bucket 10 (58720256 elements, 58720256 padded size):
800
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
801
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
802
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
803
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
804
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
805
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
806
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
807
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
808
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
809
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
811
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
812
+ Params for bucket 14 (58720256 elements, 58720256 padded size):
813
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
814
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
815
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
816
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
817
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
818
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
819
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
820
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
821
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
822
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
823
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
824
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
825
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
827
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
828
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
829
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
830
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
831
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
832
+ Params for bucket 20 (58720256 elements, 58720256 padded size):
833
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
834
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
835
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
836
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
837
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
838
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
839
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
840
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
841
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
842
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
843
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
844
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
845
+ Params for bucket 24 (58720256 elements, 58720256 padded size):
846
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
847
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
848
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
849
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
850
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
851
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
852
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
853
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
854
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
855
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
856
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
857
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
858
+ Params for bucket 28 (58720256 elements, 58720256 padded size):
859
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
860
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
861
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
862
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
863
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
864
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
865
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
866
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
867
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
868
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
869
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
870
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
871
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
872
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
873
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
874
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
875
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
876
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
877
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
878
+ Params for bucket 34 (58720256 elements, 58720256 padded size):
879
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
880
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
881
+ Params for bucket 35 (58720256 elements, 58720256 padded size):
882
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
883
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
884
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
885
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
886
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
887
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
888
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
889
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
890
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
891
+ Params for bucket 38 (58720256 elements, 58720256 padded size):
892
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
893
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
894
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
895
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
896
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
897
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
898
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
899
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
900
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
901
+ Params for bucket 41 (58720256 elements, 58720256 padded size):
902
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
903
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
904
+ Params for bucket 42 (58720256 elements, 58720256 padded size):
905
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
906
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
907
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
908
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
909
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
910
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
911
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
912
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
913
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
914
+ Params for bucket 45 (58720256 elements, 58720256 padded size):
915
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
916
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
917
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
918
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
919
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
920
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
921
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
922
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
923
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
924
+ Params for bucket 48 (58720256 elements, 58720256 padded size):
925
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
926
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
927
+ Params for bucket 49 (58720256 elements, 58720256 padded size):
928
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
929
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
930
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
931
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
932
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
933
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
934
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
935
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
936
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
937
+ Params for bucket 52 (58720256 elements, 58720256 padded size):
938
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
939
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
940
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
941
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
942
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
943
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
944
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
945
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
946
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
947
+ Params for bucket 55 (58720256 elements, 58720256 padded size):
948
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
949
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
950
+ Params for bucket 56 (58720256 elements, 58720256 padded size):
951
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
952
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
953
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x77182c21b770>, config_logger_dir='')
954
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
955
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:04:53
956
+ > building train, validation, and test datasets ...
957
+ > datasets target sizes (minimum size):
958
+ train: 192
959
+ validation: 0
960
+ test: 0
961
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
962
+ > building train, validation, and test datasets for GPT ...
963
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x77182c167c80>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
964
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
965
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
966
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
967
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
968
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
969
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
970
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
971
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
972
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
973
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
974
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
975
+ > finished creating GPT datasets ...
976
+ [after dataloaders are built] datetime: 2026-08-30 07:04:53
977
+ done with setup ...
978
+ (min, max) time across ranks (ms):
979
+ model-and-optimizer-setup ......................: (3534.01, 3536.24)
980
+ train/valid/test-data-iterators-setup ..........: (75.28, 94.05)
981
+ training ...
982
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
983
+ [before the start of training step] datetime: 2026-08-30 07:04:53
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
985
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
989
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
991
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
992
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
993
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
994
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
995
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
996
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
997
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
998
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
999
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1000
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1001
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1002
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1003
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1004
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1005
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1006
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1007
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1008
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1009
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1010
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1011
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1012
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1013
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1014
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1015
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1016
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1017
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1018
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1019
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1020
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1021
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1022
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1023
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1024
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1025
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1026
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1027
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1028
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1029
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1030
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1031
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1032
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1033
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1034
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1035
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1036
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1037
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1038
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1039
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1040
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1041
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1042
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1043
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1044
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1045
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1046
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1047
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1048
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1049
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1050
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1051
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1052
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1053
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1054
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1055
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1056
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1057
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1058
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1059
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1060
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1061
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1062
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1063
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1064
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1065
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1066
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1067
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1068
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1069
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1070
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1071
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1072
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1073
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1074
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1075
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1076
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1077
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1078
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1079
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1080
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1081
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1082
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1083
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1084
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1085
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1086
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1087
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1088
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1089
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1090
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1091
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1092
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1093
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1094
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1095
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1096
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1097
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1098
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1099
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1100
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1101
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1102
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1103
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1104
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1105
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1106
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1107
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1108
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1109
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1110
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1111
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1112
+ [2026-08-30 07:05:20] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26364.9 | throughput per GPU (TFLOP/s/GPU): 12.6 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103065E+01 | loss scale: 1.0 | grad norm: 44.386 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1113
+ Number of parameters in transformer block in billions: 23.22
1114
+ Number of parameters in embedding layers in billions: 0.41
1115
+ Total number of parameters in billions: 23.63
1116
+ Number of parameters in most loaded shard in billions: 23.6320
1117
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1118
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.0380859375 | max allocated: 66160.3212890625 | reserved: 68232.0 | max reserved: 68232.0
1119
+ [2026-08-30 07:05:20] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 544.2 | throughput per GPU (TFLOP/s/GPU): 612.6 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101734E+01 | loss scale: 1.0 | grad norm: 45.894 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1120
+ [2026-08-30 07:05:21] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 501.3 | throughput per GPU (TFLOP/s/GPU): 665.0 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.018338E+01 | loss scale: 1.0 | grad norm: 186.457 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1121
+ Capture CUDA graph for training!!!
1122
+ CUDA graph capture done!!!
1123
+ [2026-08-30 07:05:23] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2053.7 | throughput per GPU (TFLOP/s/GPU): 162.3 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.327626E+01 | loss scale: 1.0 | grad norm: 28.601 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1124
+ [2026-08-30 07:05:23] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 457.1 | throughput per GPU (TFLOP/s/GPU): 729.4 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.039188E+01 | loss scale: 1.0 | grad norm: 15.993 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1125
+ [2026-08-30 07:05:24] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 441.7 | throughput per GPU (TFLOP/s/GPU): 754.8 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.068047E+01 | loss scale: 1.0 | grad norm: 50.936 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1126
+ [2026-08-30 07:05:24] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 471.7 | throughput per GPU (TFLOP/s/GPU): 706.7 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.011516E+01 | loss scale: 1.0 | grad norm: 9.322 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1127
+ [2026-08-30 07:05:25] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 462.1 | throughput per GPU (TFLOP/s/GPU): 721.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.618699E+00 | loss scale: 1.0 | grad norm: 5.912 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1128
+ [2026-08-30 07:05:25] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 462.9 | throughput per GPU (TFLOP/s/GPU): 720.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.348099E+00 | loss scale: 1.0 | grad norm: 5.760 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1129
+ [2026-08-30 07:05:26] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 486.6 | throughput per GPU (TFLOP/s/GPU): 685.1 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.177194E+00 | loss scale: 1.0 | grad norm: 5.540 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1130
+ [2026-08-30 07:05:26] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 469.2 | throughput per GPU (TFLOP/s/GPU): 710.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.036544E+00 | loss scale: 1.0 | grad norm: 5.187 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1131
+ [2026-08-30 07:05:27] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 471.8 | throughput per GPU (TFLOP/s/GPU): 706.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.913195E+00 | loss scale: 1.0 | grad norm: 4.850 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1132
+ [after training is done] datetime: 2026-08-30 07:05:27
1133
+ [rank0]:[W830 07:05:28.535162945 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1134
+ [rank5]:[W830 07:05:29.907847489 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1135
+ [rank3]:[W830 07:05:29.927304227 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1136
+ [rank2]:[W830 07:05:29.929927607 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1137
+ [rank4]:[W830 07:05:29.930086365 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1138
+ [rank6]:[W830 07:05:29.930794444 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1139
+ [rank7]:[W830 07:05:29.077380676 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1140
+ [rank1]:[W830 07:05:30.457507130 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1141
+ W0830 07:05:35.454000 702460 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1142
+ W0830 07:05:35.455000 702460 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 702551 closing signal SIGTERM
1143
+ Traceback (most recent call last):
1144
+ File "/usr/local/bin/torchrun", line 8, in <module>
1145
+ sys.exit(main())
1146
+ ^^^^^^
1147
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1148
+ return f(*args, **kwargs)
1149
+ ^^^^^^^^^^^^^^^^^^
1150
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1151
+ run(args)
1152
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1153
+ elastic_launch(
1154
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1155
+ return launch_agent(self._config, self._entrypoint, list(args))
1156
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1157
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1158
+ result = agent.run()
1159
+ ^^^^^^^^^^^
1160
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1161
+ result = f(*args, **kwargs)
1162
+ ^^^^^^^^^^^^^^^^^^
1163
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1164
+ result = self._invoke_run(role)
1165
+ ^^^^^^^^^^^^^^^^^^^^^^
1166
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1167
+ time.sleep(monitor_interval)
1168
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1169
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1170
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 702460 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2/mem_samples.csv ADDED
@@ -0,0 +1,224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 2
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 821
42
+ 1, 819
43
+ 2, 819
44
+ 3, 821
45
+ 4, 819
46
+ 5, 819
47
+ 6, 819
48
+ 7, 819
49
+ 0, 1537
50
+ 1, 1537
51
+ 2, 1537
52
+ 3, 1537
53
+ 4, 1537
54
+ 5, 1537
55
+ 6, 1537
56
+ 7, 1537
57
+ 0, 2231
58
+ 1, 2231
59
+ 2, 2231
60
+ 3, 2231
61
+ 4, 2231
62
+ 5, 2231
63
+ 6, 2231
64
+ 7, 2231
65
+ 0, 2903
66
+ 1, 2903
67
+ 2, 2903
68
+ 3, 2903
69
+ 4, 2903
70
+ 5, 2903
71
+ 6, 2903
72
+ 7, 2903
73
+ 0, 41191
74
+ 1, 41227
75
+ 2, 41149
76
+ 3, 41203
77
+ 4, 41221
78
+ 5, 41205
79
+ 6, 41193
80
+ 7, 41213
81
+ 0, 41929
82
+ 1, 41963
83
+ 2, 41953
84
+ 3, 41929
85
+ 4, 41929
86
+ 5, 41951
87
+ 6, 41955
88
+ 7, 41953
89
+ 0, 42679
90
+ 1, 42671
91
+ 2, 42671
92
+ 3, 42671
93
+ 4, 42671
94
+ 5, 42671
95
+ 6, 42671
96
+ 7, 42671
97
+ 0, 42957
98
+ 1, 42949
99
+ 2, 42949
100
+ 3, 42949
101
+ 4, 42949
102
+ 5, 42949
103
+ 6, 42949
104
+ 7, 42949
105
+ 0, 44381
106
+ 1, 44373
107
+ 2, 44373
108
+ 3, 44373
109
+ 4, 44373
110
+ 5, 44373
111
+ 6, 44373
112
+ 7, 44373
113
+ 0, 59505
114
+ 1, 59555
115
+ 2, 59527
116
+ 3, 59541
117
+ 4, 59511
118
+ 5, 59849
119
+ 6, 59849
120
+ 7, 59851
121
+ 0, 65637
122
+ 1, 65615
123
+ 2, 65615
124
+ 3, 65615
125
+ 4, 65629
126
+ 5, 65615
127
+ 6, 65615
128
+ 7, 65615
129
+ 0, 66123
130
+ 1, 66101
131
+ 2, 66101
132
+ 3, 66101
133
+ 4, 66115
134
+ 5, 66101
135
+ 6, 66101
136
+ 7, 66101
137
+ 0, 73933
138
+ 1, 73899
139
+ 2, 73901
140
+ 3, 73905
141
+ 4, 73911
142
+ 5, 73913
143
+ 6, 73929
144
+ 7, 73927
145
+ 0, 74997
146
+ 1, 74971
147
+ 2, 74957
148
+ 3, 74953
149
+ 4, 74965
150
+ 5, 74975
151
+ 6, 74965
152
+ 7, 74987
153
+ 0, 76383
154
+ 1, 76353
155
+ 2, 76349
156
+ 3, 76357
157
+ 4, 76251
158
+ 5, 76343
159
+ 6, 76361
160
+ 7, 76349
161
+ 0, 79657
162
+ 1, 91011
163
+ 2, 80759
164
+ 3, 75493
165
+ 4, 73973
166
+ 5, 95021
167
+ 6, 95179
168
+ 7, 74135
169
+ 0, 102691
170
+ 1, 102733
171
+ 2, 102731
172
+ 3, 102733
173
+ 4, 102627
174
+ 5, 102733
175
+ 6, 102731
176
+ 7, 102733
177
+ 0, 102691
178
+ 1, 102733
179
+ 2, 102731
180
+ 3, 102733
181
+ 4, 102627
182
+ 5, 102733
183
+ 6, 102731
184
+ 7, 102733
185
+ 0, 102105
186
+ 1, 102467
187
+ 2, 102203
188
+ 3, 102147
189
+ 4, 102361
190
+ 5, 102205
191
+ 6, 102465
192
+ 7, 102205
193
+ 0, 98883
194
+ 1, 98993
195
+ 2, 98603
196
+ 3, 98605
197
+ 4, 98887
198
+ 5, 98605
199
+ 6, 98667
200
+ 7, 99251
201
+ 0, 97267
202
+ 1, 97571
203
+ 2, 97243
204
+ 3, 96989
205
+ 4, 97465
206
+ 5, 97245
207
+ 6, 96987
208
+ 7, 97571
209
+ 0, 4
210
+ 1, 96537
211
+ 2, 4
212
+ 3, 4
213
+ 4, 4
214
+ 5, 4
215
+ 6, 4
216
+ 7, 96537
217
+ 0, 0
218
+ 1, 0
219
+ 2, 0
220
+ 3, 0
221
+ 4, 0
222
+ 5, 0
223
+ 6, 0
224
+ 7, 0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G2
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=3584
11
+ num_experts=32
12
+ topk=8
13
+ experts_per_rank=4
14
+ expert_params=44040192
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=32768
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.08203125
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2/train.log ADDED
@@ -0,0 +1,1171 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G2 / ultraep (slots/rank=1 policy=one)
3
+ E=32 K=8 H'=3584 m=4
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.08203125 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:05:38.954000 705505 torch/distributed/run.py:766]
10
+ W0830 07:05:38.954000 705505 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:05:38.954000 705505 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:05:38.954000 705505 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
18
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
19
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
20
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
21
+ Number of virtual stages per pipeline stage: None
22
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
23
+ using torch.bfloat16 for parameters ...
24
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
25
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
26
+ ------------------------ arguments ------------------------
27
+ account_for_embedding_in_pipeline_split ......... False
28
+ account_for_loss_in_pipeline_split .............. False
29
+ accumulate_allreduce_grads_in_fp32 .............. True
30
+ adam_beta1 ...................................... 0.9
31
+ adam_beta2 ...................................... 0.999
32
+ adam_eps ........................................ 1e-08
33
+ add_bias_linear ................................. False
34
+ add_position_embedding .......................... False
35
+ add_qkv_bias .................................... False
36
+ adlr_autoresume ................................. False
37
+ adlr_autoresume_interval ........................ 1000
38
+ align_grad_reduce ............................... True
39
+ align_param_gather .............................. False
40
+ app_tag_run_name ................................ None
41
+ app_tag_run_version ............................. 0.0.0
42
+ apply_layernorm_1p .............................. False
43
+ apply_query_key_layer_scaling ................... False
44
+ apply_residual_connection_post_layernorm ........ False
45
+ apply_rope_fusion ............................... True
46
+ async_save ...................................... None
47
+ async_tensor_model_parallel_allreduce ........... True
48
+ attention_backend ............................... AttnBackend.auto
49
+ attention_dropout ............................... 0.0
50
+ attention_softmax_in_fp32 ....................... False
51
+ auto_detect_ckpt_format ......................... False
52
+ barrier_with_L1_time ............................ True
53
+ bert_binary_head ................................ True
54
+ bert_embedder_type .............................. megatron
55
+ bert_load ....................................... None
56
+ bf16 ............................................ True
57
+ bias_dropout_fusion ............................. True
58
+ bias_gelu_fusion ................................ False
59
+ bias_swiglu_fusion .............................. True
60
+ biencoder_projection_dim ........................ 0
61
+ biencoder_shared_query_context_model ............ False
62
+ block_data_path ................................. None
63
+ calc_ft_timeouts ................................ False
64
+ calculate_per_token_loss ........................ False
65
+ check_for_large_grads ........................... False
66
+ check_for_nan_in_loss_and_grad .................. False
67
+ check_for_spiky_loss ............................ False
68
+ check_weight_hash_across_dp_replicas_interval ... None
69
+ ckpt_assume_constant_structure .................. False
70
+ ckpt_convert_format ............................. None
71
+ ckpt_convert_save ............................... None
72
+ ckpt_convert_update_legacy_dist_opt_format ...... False
73
+ ckpt_format ..................................... torch_dist
74
+ ckpt_fully_parallel_load ........................ False
75
+ ckpt_fully_parallel_save ........................ True
76
+ ckpt_fully_parallel_save_deprecated ............. False
77
+ ckpt_step ....................................... None
78
+ classes_fraction ................................ 1.0
79
+ clip_grad ....................................... 1.0
80
+ clone_scatter_output_in_embedding ............... True
81
+ config_logger_dir ...............................
82
+ consumed_train_samples .......................... 0
83
+ consumed_valid_samples .......................... 0
84
+ context_parallel_size ........................... 1
85
+ cp_comm_type .................................... ['p2p']
86
+ create_attention_mask_in_dataloader ............. True
87
+ cross_entropy_fusion_impl ....................... native
88
+ cross_entropy_loss_fusion ....................... False
89
+ cuda_graph_scope ................................ full_iteration
90
+ cuda_graph_warmup_steps ......................... 3
91
+ data_args_path .................................. None
92
+ data_cache_path ................................. None
93
+ data_parallel_random_init ....................... False
94
+ data_parallel_sharding_strategy ................. no_shard
95
+ data_parallel_size .............................. 8
96
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
97
+ data_per_class_fraction ......................... 1.0
98
+ data_sharding ................................... True
99
+ dataloader_type ................................. single
100
+ ddp_average_in_collective ....................... False
101
+ ddp_bucket_size ................................. None
102
+ ddp_num_buckets ................................. None
103
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
104
+ decoder_first_pipeline_num_layers ............... None
105
+ decoder_last_pipeline_num_layers ................ None
106
+ decoder_num_layers .............................. None
107
+ decoder_seq_length .............................. None
108
+ decoupled_lr .................................... None
109
+ decoupled_min_lr ................................ None
110
+ decrease_batch_size_if_needed ................... False
111
+ defer_embedding_wgrad_compute ................... False
112
+ delay_wgrad_compute ............................. False
113
+ deprecated_use_mcore_models ..................... True
114
+ deterministic_mode .............................. False
115
+ dino_bottleneck_size ............................ 256
116
+ dino_freeze_last_layer .......................... 1
117
+ dino_head_hidden_size ........................... 2048
118
+ dino_local_crops_number ......................... 10
119
+ dino_local_img_size ............................. 96
120
+ dino_norm_last_layer ............................ False
121
+ dino_teacher_temp ............................... 0.07
122
+ dino_warmup_teacher_temp ........................ 0.04
123
+ dino_warmup_teacher_temp_epochs ................. 30
124
+ disable_bf16_reduced_precision_matmul ........... False
125
+ disable_mamba_mem_eff_path ...................... False
126
+ disable_straggler_on_startup .................... False
127
+ dist_ckpt_format_deprecated ..................... None
128
+ dist_ckpt_strictness ............................ assume_ok_unexpected
129
+ distribute_saved_activations .................... False
130
+ distributed_backend ............................. nccl
131
+ distributed_timeout_minutes ..................... 10
132
+ embedding_init_method_std ....................... None
133
+ embedding_path .................................. None
134
+ empty_unused_memory_level ....................... 0
135
+ enable_cuda_graph ............................... True
136
+ enable_experimental ............................. False
137
+ enable_ft_package ............................... False
138
+ enable_gloo_process_groups ...................... True
139
+ enable_msc ...................................... True
140
+ enable_one_logger ............................... True
141
+ encoder_num_layers .............................. 16
142
+ encoder_seq_length .............................. 4096
143
+ end_weight_decay ................................ 0.1
144
+ eod_mask_loss ................................... False
145
+ error_injection_rate ............................ 0
146
+ error_injection_type ............................ transient_error
147
+ eval_interval ................................... 1000000
148
+ eval_iters ...................................... 0
149
+ evidence_data_path .............................. None
150
+ exit_duration_in_mins ........................... None
151
+ exit_interval ................................... None
152
+ exit_on_missing_checkpoint ...................... False
153
+ exit_signal_handler ............................. False
154
+ exp_avg_dtype ................................... torch.float32
155
+ exp_avg_sq_dtype ................................ torch.float32
156
+ expert_model_parallel_size ...................... 8
157
+ expert_tensor_parallel_size ..................... 1
158
+ export_force_local_attention .................... False
159
+ export_kd_cfg ................................... None
160
+ export_kd_teacher_ckpt_format ................... None
161
+ export_kd_teacher_load .......................... None
162
+ export_kv_cache_quant ........................... False
163
+ export_legacy_megatron .......................... False
164
+ export_model_type ............................... GPTModel
165
+ export_moe_apply_probs_on_input ................. False
166
+ export_qk_l2_norm ............................... False
167
+ export_quant_cfg ................................ None
168
+ export_real_quant_cfg ........................... None
169
+ export_te_mcore_model ........................... False
170
+ external_cuda_graph ............................. False
171
+ ffn_hidden_size ................................. 14336
172
+ finetune ........................................ False
173
+ finetune_data_split ............................. train
174
+ finetune_hf_dataset ............................. None
175
+ first_last_layers_bf16 .......................... False
176
+ flash_decode .................................... False
177
+ fp16 ............................................ False
178
+ fp16_lm_cross_entropy ........................... False
179
+ fp32_residual_connection ........................ False
180
+ fp8 ............................................. e4m3
181
+ fp8_amax_compute_algo ........................... most_recent
182
+ fp8_amax_history_len ............................ 1
183
+ fp8_interval .................................... 1
184
+ fp8_margin ...................................... 0
185
+ fp8_param_gather ................................ False
186
+ fp8_recipe ...................................... mxfp8
187
+ fp8_wgrad ....................................... True
188
+ fsdp_double_buffer .............................. False
189
+ global_batch_size ............................... 16
190
+ grad_reduce_in_bf16 ............................. False
191
+ gradient_accumulation_fusion .................... True
192
+ gradient_reduce_div_fusion ...................... True
193
+ group_query_attention ........................... True
194
+ head_lr_mult .................................... 1.0
195
+ heterogeneous_layers_config_encoded_json ........ None
196
+ heterogeneous_layers_config_path ................ None
197
+ hidden_dropout .................................. 0.0
198
+ hidden_size ..................................... 4096
199
+ hierarchical_context_parallel_sizes ............. None
200
+ high_priority_stream_groups ..................... []
201
+ hybrid_attention_ratio .......................... 0.0
202
+ hybrid_mlp_ratio ................................ 0.0
203
+ hybrid_override_pattern ......................... None
204
+ hysteresis ...................................... 2
205
+ ict_head_size ................................... None
206
+ ict_load ........................................ None
207
+ img_h ........................................... 224
208
+ img_w ........................................... 224
209
+ indexer_batch_size .............................. 128
210
+ indexer_log_interval ............................ 1000
211
+ inference_batch_times_seqlen_threshold .......... -1
212
+ inference_dynamic_batching ...................... False
213
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
214
+ inference_dynamic_batching_buffer_overflow_factor None
215
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
216
+ inference_dynamic_batching_chunk_size ........... 256
217
+ inference_dynamic_batching_max_requests_override None
218
+ inference_dynamic_batching_max_tokens_override .. None
219
+ inference_dynamic_batching_num_cuda_graphs ...... 16
220
+ inference_max_batch_size ........................ 8
221
+ inference_max_seq_length ........................ 2560
222
+ inference_rng_tracker ........................... False
223
+ init_method_std ................................. 0.01
224
+ init_method_xavier_uniform ...................... False
225
+ init_model_with_meta_device ..................... False
226
+ initial_loss_scale .............................. 4294967296
227
+ inprocess_active_world_size ..................... 8
228
+ inprocess_barrier_timeout ....................... 120
229
+ inprocess_completion_timeout .................... 120
230
+ inprocess_empty_cuda_cache ...................... False
231
+ inprocess_granularity ........................... node
232
+ inprocess_hard_timeout .......................... 90
233
+ inprocess_heartbeat_interval .................... 30
234
+ inprocess_heartbeat_timeout ..................... 60
235
+ inprocess_last_call_wait ........................ 1
236
+ inprocess_max_iterations ........................ None
237
+ inprocess_monitor_process_interval .............. 1.0
238
+ inprocess_monitor_thread_interval ............... 1.0
239
+ inprocess_progress_watchdog_interval ............ 1.0
240
+ inprocess_restart ............................... False
241
+ inprocess_soft_timeout .......................... 60
242
+ inprocess_termination_grace_time ................ 1
243
+ is_hybrid_model ................................. False
244
+ iter_per_epoch .................................. 1250
245
+ iterations_to_skip .............................. []
246
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
247
+ kitchen_config_file ............................. None
248
+ kitchen_recipe_number ........................... None
249
+ kv_channels ..................................... 128
250
+ kv_lora_rank .................................... 32
251
+ lazy_mpu_init ................................... None
252
+ load ............................................ None
253
+ load_main_params_from_ckpt ...................... None
254
+ load_model_opt_format ........................... False
255
+ local_rank ...................................... 0
256
+ log_energy ...................................... False
257
+ log_interval .................................... 1
258
+ log_loss_scale_to_tensorboard ................... True
259
+ log_memory_to_tensorboard ....................... False
260
+ log_num_zeros_in_grad ........................... False
261
+ log_params_norm ................................. False
262
+ log_progress .................................... False
263
+ log_straggler ................................... False
264
+ log_throughput .................................. True
265
+ log_timers_to_tensorboard ....................... False
266
+ log_validation_ppl_to_tensorboard ............... False
267
+ log_world_size_to_tensorboard ................... False
268
+ logging_level ................................... None
269
+ loss_scale ...................................... None
270
+ loss_scale_window ............................... 1000
271
+ lr .............................................. 0.0001
272
+ lr_decay_iters .................................. 8
273
+ lr_decay_samples ................................ None
274
+ lr_decay_style .................................. cosine
275
+ lr_warmup_fraction .............................. None
276
+ lr_warmup_init .................................. 0.0
277
+ lr_warmup_iters ................................. 5
278
+ lr_warmup_samples ............................... 0
279
+ lr_wsd_decay_iters .............................. None
280
+ lr_wsd_decay_samples ............................ None
281
+ lr_wsd_decay_style .............................. exponential
282
+ main_grads_dtype ................................ torch.float32
283
+ main_params_dtype ............................... torch.float32
284
+ make_vocab_size_divisible_by .................... 128
285
+ mamba_head_dim .................................. 64
286
+ mamba_num_groups ................................ 8
287
+ mamba_num_heads ................................. None
288
+ mamba_state_dim ................................. 128
289
+ manual_gc ....................................... False
290
+ manual_gc_eval .................................. True
291
+ manual_gc_interval .............................. 0
292
+ mask_factor ..................................... 1.0
293
+ mask_prob ....................................... 0.15
294
+ mask_type ....................................... random
295
+ masked_softmax_fusion ........................... False
296
+ max_position_embeddings ......................... 32768
297
+ max_tokens_to_oom ............................... 12000
298
+ memory_snapshot_path ............................ snapshot.pickle
299
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
300
+ micro_batch_size ................................ 1
301
+ microbatch_group_size_per_vp_stage .............. None
302
+ mid_level_dataset_surplus ....................... 0.005
303
+ min_loss_scale .................................. 1.0
304
+ min_lr .......................................... 1e-05
305
+ mlp_chunks_for_prefill .......................... 1
306
+ mmap_bin_files .................................. True
307
+ mock_data ....................................... False
308
+ moe_apply_probs_on_input ........................ False
309
+ moe_aux_loss_coeff .............................. 0.01
310
+ moe_capacity_monitor ............................ True
311
+ moe_deepep_num_sms .............................. 20
312
+ moe_echo_enable_random_offloading ............... False
313
+ moe_echo_expert_dispatch_overlap ................ False
314
+ moe_echo_expert_dispatcher_type ................. hybridep
315
+ moe_echo_recompute_expert_dispatch .............. False
316
+ moe_enable_deepep ............................... False
317
+ moe_enable_echo ................................. False
318
+ moe_enable_hybridep ............................. True
319
+ moe_enable_ultraep .............................. True
320
+ moe_expert_capacity_factor ...................... None
321
+ moe_extended_tp ................................. False
322
+ moe_ffn_hidden_size ............................. 3584
323
+ moe_grouped_gemm ................................ True
324
+ moe_hybridep_num_sms ............................ 16
325
+ moe_input_jitter_eps ............................ None
326
+ moe_layer_freq .................................. 1
327
+ moe_layer_recompute ............................. False
328
+ moe_num_echo_experts ............................ None
329
+ moe_num_redundant_experts_per_rank .............. 1
330
+ moe_pad_expert_input_to_capacity ................ False
331
+ moe_per_layer_logging ........................... False
332
+ moe_permute_fusion .............................. False
333
+ moe_received_token_capacity ..................... 1.08203125
334
+ moe_router_bias_update_rate ..................... 0.001
335
+ moe_router_dtype ................................ fp32
336
+ moe_router_enable_expert_bias ................... False
337
+ moe_router_force_load_balancing ................. False
338
+ moe_router_group_topk ........................... None
339
+ moe_router_load_balancing_type .................. aux_loss
340
+ moe_router_num_groups ........................... None
341
+ moe_router_padding_for_fp8 ...................... False
342
+ moe_router_pre_softmax .......................... False
343
+ moe_router_score_function ....................... softmax
344
+ moe_router_topk ................................. 8
345
+ moe_router_topk_scaling_factor .................. None
346
+ moe_shared_expert_intermediate_size ............. None
347
+ moe_shared_expert_overlap ....................... False
348
+ moe_token_dispatcher_type ....................... flex
349
+ moe_token_drop_policy ........................... probs
350
+ moe_ultraep_max_microbatches .................... None
351
+ moe_ultraep_overlap_weight_sync ................. True
352
+ moe_ultraep_weight_sync_join .................... before_dispatch
353
+ moe_upcycling_granularity ....................... 1
354
+ moe_use_device_initiated_grouped_gemm ........... True
355
+ moe_use_legacy_grouped_gemm ..................... False
356
+ moe_use_upcycling ............................... False
357
+ moe_z_loss_coeff ................................ None
358
+ mrope_section ................................... None
359
+ mscale .......................................... 1.0
360
+ mscale_all_dim .................................. 0.0
361
+ mtp_loss_scaling_factor ......................... 0.1
362
+ mtp_num_layers .................................. None
363
+ multi_latent_attention .......................... False
364
+ nccl_all_reduce_for_prefill ..................... False
365
+ nccl_communicator_config_path ................... None
366
+ nccl_ub ......................................... False
367
+ no_load_optim ................................... None
368
+ no_load_rng ..................................... None
369
+ no_persist_layer_norm ........................... False
370
+ no_rope_freq .................................... None
371
+ no_save_optim ................................... None
372
+ no_save_rng ..................................... None
373
+ non_persistent_ckpt_type ........................ None
374
+ non_persistent_global_ckpt_dir .................. None
375
+ non_persistent_local_ckpt_algo .................. fully_parallel
376
+ non_persistent_local_ckpt_dir ................... None
377
+ non_persistent_save_interval .................... None
378
+ norm_epsilon .................................... 1e-05
379
+ normalization ................................... RMSNorm
380
+ num_attention_heads ............................. 32
381
+ num_channels .................................... 3
382
+ num_classes ..................................... 1000
383
+ num_dataset_builder_threads ..................... 1
384
+ num_distributed_optimizer_instances ............. 1
385
+ num_experts ..................................... 32
386
+ num_layers ...................................... 16
387
+ num_layers_at_end_in_bf16 ....................... 1
388
+ num_layers_at_start_in_bf16 ..................... 1
389
+ num_layers_per_virtual_pipeline_stage ........... None
390
+ num_query_groups ................................ 8
391
+ num_virtual_stages_per_pipeline_rank ............ None
392
+ num_workers ..................................... 2
393
+ object_storage_cache_path ....................... None
394
+ one_logger_async ................................ False
395
+ one_logger_project .............................. megatron-lm
396
+ one_logger_run_name ............................. None
397
+ onnx_safe ....................................... None
398
+ openai_gelu ..................................... False
399
+ optimizer ....................................... adam
400
+ optimizer_cpu_offload ........................... False
401
+ optimizer_offload_fraction ...................... 1.0
402
+ output_bert_embeddings .......................... False
403
+ overlap_cpu_optimizer_d2h_h2d ................... False
404
+ overlap_grad_reduce ............................. True
405
+ overlap_moe_expert_parallel_comm ................ False
406
+ overlap_p2p_comm ................................ False
407
+ overlap_p2p_comm_warmup_flush ................... False
408
+ overlap_param_gather ............................ True
409
+ overlap_param_gather_with_optimizer_step ........ False
410
+ override_opt_param_scheduler .................... False
411
+ params_dtype .................................... torch.bfloat16
412
+ patch_dim ....................................... 16
413
+ per_split_data_args_path ........................ None
414
+ perform_initialization .......................... True
415
+ pin_cpu_grads ................................... True
416
+ pin_cpu_params .................................. True
417
+ pipeline_model_parallel_comm_backend ............ None
418
+ pipeline_model_parallel_layout .................. None
419
+ pipeline_model_parallel_size .................... 1
420
+ position_embedding_type ......................... rope
421
+ pretrained_checkpoint ........................... None
422
+ profile ......................................... False
423
+ profile_ranks ................................... [0]
424
+ profile_step_end ................................ 12
425
+ profile_step_start .............................. 10
426
+ q_lora_rank ..................................... None
427
+ qk_head_dim ..................................... 128
428
+ qk_l2_norm ...................................... False
429
+ qk_layernorm .................................... False
430
+ qk_pos_emb_head_dim ............................. 64
431
+ query_in_block_prob ............................. 0.1
432
+ rampup_batch_size ............................... None
433
+ rank ............................................ 0
434
+ recompute_granularity ........................... None
435
+ recompute_method ................................ None
436
+ recompute_modules ............................... None
437
+ recompute_num_layers ............................ None
438
+ record_memory_history ........................... False
439
+ relative_attention_max_distance ................. 128
440
+ relative_attention_num_buckets .................. 32
441
+ replication ..................................... False
442
+ replication_factor .............................. 2
443
+ replication_jump ................................ None
444
+ rerun_mode ...................................... validate_results
445
+ reset_attention_mask ............................ False
446
+ reset_position_ids .............................. False
447
+ result_rejected_tracker_filename ................ None
448
+ retriever_report_topk_accuracies ................ []
449
+ retriever_score_scaling ......................... False
450
+ retriever_seq_length ............................ 256
451
+ retro_add_retriever ............................. False
452
+ retro_attention_gate ............................ 1
453
+ retro_cyclic_train_iters ........................ None
454
+ retro_encoder_attention_dropout ................. 0.1
455
+ retro_encoder_hidden_dropout .................... 0.1
456
+ retro_encoder_layers ............................ 2
457
+ retro_num_neighbors ............................. 2
458
+ retro_num_retrieved_chunks ...................... 2
459
+ retro_project_dir ............................... None
460
+ retro_verify_neighbor_count ..................... True
461
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
462
+ rope_scaling_factor ............................. 8.0
463
+ rope_type ....................................... None
464
+ rotary_base ..................................... 1000000
465
+ rotary_interleaved .............................. False
466
+ rotary_percent .................................. 1.0
467
+ rotary_scaling_factor ........................... 1.0
468
+ rotary_seq_len_interpolation_factor ............. None
469
+ run_workload_inspector_server ................... False
470
+ sample_rate ..................................... 1.0
471
+ save ............................................ None
472
+ save_interval ................................... None
473
+ save_retain_interval ............................ None
474
+ scatter_gather_tensors_in_pipeline .............. True
475
+ seed ............................................ 1234
476
+ seq_length ...................................... 4096
477
+ sequence_parallel ............................... False
478
+ sft ............................................. False
479
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
480
+ sgd_momentum .................................... 0.9
481
+ short_seq_prob .................................. 0.1
482
+ skip_train ...................................... False
483
+ skipped_train_samples ........................... 0
484
+ spec ............................................ None
485
+ split ........................................... 100,0,0
486
+ squared_relu .................................... False
487
+ start_weight_decay .............................. 0.1
488
+ straggler_ctrlr_port ............................ 65535
489
+ straggler_minmax_count .......................... 1
490
+ suggested_communication_unit_size ............... None
491
+ swiglu .......................................... True
492
+ swin_backbone_type .............................. tiny
493
+ symmetric_ar_type ............................... None
494
+ te_rng_tracker .................................. True
495
+ tensor_model_parallel_size ...................... 1
496
+ tensorboard_dir ................................. None
497
+ tensorboard_log_interval ........................ 1
498
+ tensorboard_queue_size .......................... 1000
499
+ test_data_path .................................. None
500
+ test_mode ....................................... False
501
+ tiktoken_num_special_tokens ..................... 1000
502
+ tiktoken_pattern ................................ None
503
+ tiktoken_special_tokens ......................... None
504
+ timing_log_level ................................ 0
505
+ timing_log_option ............................... minmax
506
+ titles_data_path ................................ None
507
+ tokenizer_model ................................. None
508
+ tokenizer_type .................................. GPT2BPETokenizer
509
+ torch_fsdp2_reshard_after_forward ............... True
510
+ tp_comm_bootstrap_backend ....................... nccl
511
+ tp_comm_bulk_dgrad .............................. True
512
+ tp_comm_bulk_wgrad .............................. True
513
+ tp_comm_overlap ................................. False
514
+ tp_comm_overlap_ag .............................. True
515
+ tp_comm_overlap_cfg ............................. None
516
+ tp_comm_overlap_rs .............................. True
517
+ tp_comm_overlap_rs_dgrad ........................ False
518
+ tp_comm_split_ag ................................ True
519
+ tp_comm_split_rs ................................ True
520
+ train_data_path ................................. None
521
+ train_iters ..................................... 12
522
+ train_samples ................................... None
523
+ train_sync_interval ............................. None
524
+ transformer_impl ................................ transformer_engine
525
+ transformer_pipeline_model_parallel_size ........ 1
526
+ untie_embeddings_and_output_weights ............. True
527
+ use_checkpoint_args ............................. False
528
+ use_checkpoint_opt_param_scheduler .............. False
529
+ use_cpu_initialization .......................... None
530
+ use_custom_fsdp ................................. False
531
+ use_dist_ckpt ................................... True
532
+ use_dist_ckpt_deprecated ........................ False
533
+ use_distributed_optimizer ....................... True
534
+ use_flash_attn .................................. False
535
+ use_legacy_models ............................... False
536
+ use_mp_args_from_checkpoint_args ................ False
537
+ use_one_sent_docs ............................... False
538
+ use_persistent_ckpt_worker ...................... False
539
+ use_precision_aware_optimizer ................... False
540
+ use_pytorch_profiler ............................ False
541
+ use_ring_exchange_p2p ........................... False
542
+ use_rope_scaling ................................ False
543
+ use_rotary_position_embeddings .................. False
544
+ use_sharp ....................................... False
545
+ use_tokenizer_model_from_checkpoint_args ........ True
546
+ use_torch_fsdp2 ................................. False
547
+ use_torch_optimizer_for_cpu_offload ............. False
548
+ use_tp_pp_dp_mapping ............................ False
549
+ v_head_dim ...................................... 128
550
+ valid_data_path ................................. None
551
+ variable_seq_lengths ............................ False
552
+ virtual_pipeline_model_parallel_size ............ None
553
+ vision_backbone_type ............................ vit
554
+ vision_pretraining .............................. False
555
+ vision_pretraining_type ......................... classify
556
+ vocab_extra_ids ................................. 0
557
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
558
+ vocab_size ...................................... None
559
+ wandb_exp_name ..................................
560
+ wandb_project ...................................
561
+ wandb_save_dir ..................................
562
+ weight_decay .................................... 0.1
563
+ weight_decay_incr_style ......................... constant
564
+ wgrad_deferral_limit ............................ 0
565
+ world_size ...................................... 8
566
+ yaml_cfg ........................................ None
567
+ -------------------- end of arguments ---------------------
568
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
569
+ > building GPT2BPETokenizer tokenizer ...
570
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
571
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
572
+ > initializing torch distributed ...
573
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
574
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
575
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
576
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
577
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
578
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
579
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
580
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
581
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
582
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank 352 is connected to is connected to is connected to 777 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 777
585
+
586
+
587
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
597
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [Gloo] Rank [Gloo] Rank 00 is connected to is connected to 00 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 00
604
+
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
607
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank4]:[W830 07:05:57.243999327 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank7]:[W830 07:05:57.244175294 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank6]:[W830 07:05:57.244200124 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank1]:[W830 07:05:57.244621967 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank3]:[W830 07:05:57.245805078 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank5]:[W830 07:05:57.245835243 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank2]:[W830 07:05:57.245862070 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.035 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:05:57.403346557 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.720 seconds
625
+ time to initialize megatron (seconds): 7.147
626
+ [after megatron is initialized] datetime: 2026-08-30 07:06:02
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 4608626688
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (42086400 elements, 42086400 padded size):
670
+ module.decoder.final_layernorm.weight
671
+ module.decoder.layers.15.mlp.router.weight
672
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
673
+ module.decoder.layers.15.self_attention.linear_proj.weight
674
+ module.decoder.layers.15.self_attention.linear_qkv.weight
675
+ module.decoder.layers.15.pre_mlp_layernorm.weight
676
+ Params for bucket 3 (42082304 elements, 42082304 padded size):
677
+ module.decoder.layers.14.self_attention.linear_qkv.weight
678
+ module.decoder.layers.14.pre_mlp_layernorm.weight
679
+ module.decoder.layers.14.mlp.router.weight
680
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
681
+ module.decoder.layers.14.self_attention.linear_proj.weight
682
+ Params for bucket 4 (42082304 elements, 42082304 padded size):
683
+ module.decoder.layers.13.pre_mlp_layernorm.weight
684
+ module.decoder.layers.13.self_attention.linear_qkv.weight
685
+ module.decoder.layers.13.mlp.router.weight
686
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
687
+ module.decoder.layers.13.self_attention.linear_proj.weight
688
+ Params for bucket 5 (42082304 elements, 42082304 padded size):
689
+ module.decoder.layers.12.pre_mlp_layernorm.weight
690
+ module.decoder.layers.12.self_attention.linear_qkv.weight
691
+ module.decoder.layers.12.mlp.router.weight
692
+ module.decoder.layers.12.self_attention.linear_proj.weight
693
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
694
+ Params for bucket 6 (42082304 elements, 42082304 padded size):
695
+ module.decoder.layers.11.self_attention.linear_qkv.weight
696
+ module.decoder.layers.11.pre_mlp_layernorm.weight
697
+ module.decoder.layers.11.mlp.router.weight
698
+ module.decoder.layers.11.self_attention.linear_proj.weight
699
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
700
+ Params for bucket 7 (42082304 elements, 42082304 padded size):
701
+ module.decoder.layers.10.self_attention.linear_qkv.weight
702
+ module.decoder.layers.10.pre_mlp_layernorm.weight
703
+ module.decoder.layers.10.mlp.router.weight
704
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
705
+ module.decoder.layers.10.self_attention.linear_proj.weight
706
+ Params for bucket 8 (42082304 elements, 42082304 padded size):
707
+ module.decoder.layers.9.self_attention.linear_qkv.weight
708
+ module.decoder.layers.9.pre_mlp_layernorm.weight
709
+ module.decoder.layers.9.mlp.router.weight
710
+ module.decoder.layers.9.self_attention.linear_proj.weight
711
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
712
+ Params for bucket 9 (42082304 elements, 42082304 padded size):
713
+ module.decoder.layers.8.self_attention.linear_qkv.weight
714
+ module.decoder.layers.8.pre_mlp_layernorm.weight
715
+ module.decoder.layers.8.mlp.router.weight
716
+ module.decoder.layers.8.self_attention.linear_proj.weight
717
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
718
+ Params for bucket 10 (42082304 elements, 42082304 padded size):
719
+ module.decoder.layers.7.pre_mlp_layernorm.weight
720
+ module.decoder.layers.7.self_attention.linear_qkv.weight
721
+ module.decoder.layers.7.mlp.router.weight
722
+ module.decoder.layers.7.self_attention.linear_proj.weight
723
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
724
+ Params for bucket 11 (42082304 elements, 42082304 padded size):
725
+ module.decoder.layers.6.self_attention.linear_qkv.weight
726
+ module.decoder.layers.6.pre_mlp_layernorm.weight
727
+ module.decoder.layers.6.mlp.router.weight
728
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
729
+ module.decoder.layers.6.self_attention.linear_proj.weight
730
+ Params for bucket 12 (42082304 elements, 42082304 padded size):
731
+ module.decoder.layers.5.self_attention.linear_qkv.weight
732
+ module.decoder.layers.5.self_attention.linear_proj.weight
733
+ module.decoder.layers.5.pre_mlp_layernorm.weight
734
+ module.decoder.layers.5.mlp.router.weight
735
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
736
+ Params for bucket 13 (42082304 elements, 42082304 padded size):
737
+ module.decoder.layers.4.mlp.router.weight
738
+ module.decoder.layers.4.self_attention.linear_proj.weight
739
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
740
+ module.decoder.layers.4.pre_mlp_layernorm.weight
741
+ module.decoder.layers.4.self_attention.linear_qkv.weight
742
+ Params for bucket 14 (42082304 elements, 42082304 padded size):
743
+ module.decoder.layers.3.self_attention.linear_qkv.weight
744
+ module.decoder.layers.3.pre_mlp_layernorm.weight
745
+ module.decoder.layers.3.mlp.router.weight
746
+ module.decoder.layers.3.self_attention.linear_proj.weight
747
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
748
+ Params for bucket 15 (42082304 elements, 42082304 padded size):
749
+ module.decoder.layers.2.pre_mlp_layernorm.weight
750
+ module.decoder.layers.2.self_attention.linear_qkv.weight
751
+ module.decoder.layers.2.mlp.router.weight
752
+ module.decoder.layers.2.self_attention.linear_proj.weight
753
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
754
+ Params for bucket 16 (42082304 elements, 42082304 padded size):
755
+ module.decoder.layers.1.pre_mlp_layernorm.weight
756
+ module.decoder.layers.1.self_attention.linear_proj.weight
757
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
758
+ module.decoder.layers.1.mlp.router.weight
759
+ module.decoder.layers.1.self_attention.linear_qkv.weight
760
+ Params for bucket 17 (42082304 elements, 42082304 padded size):
761
+ module.decoder.layers.0.self_attention.linear_qkv.weight
762
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
763
+ module.decoder.layers.0.mlp.router.weight
764
+ module.decoder.layers.0.pre_mlp_layernorm.weight
765
+ module.decoder.layers.0.self_attention.linear_proj.weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56
769
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
771
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
772
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
773
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
774
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
775
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
776
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
777
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
778
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
779
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
780
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
781
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
782
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
783
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
784
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
785
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
786
+ Params for bucket 6 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
788
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
789
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
790
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
791
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
792
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
793
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
794
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
795
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
796
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
797
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
798
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
799
+ Params for bucket 10 (58720256 elements, 58720256 padded size):
800
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
801
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
802
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
803
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
804
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
805
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
806
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
807
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
808
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
809
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
811
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
812
+ Params for bucket 14 (58720256 elements, 58720256 padded size):
813
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
814
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
815
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
816
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
817
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
818
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
819
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
820
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
821
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
822
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
823
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
824
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
825
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
827
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
828
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
829
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
830
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
831
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
832
+ Params for bucket 20 (58720256 elements, 58720256 padded size):
833
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
834
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
835
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
836
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
837
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
838
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
839
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
840
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
841
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
842
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
843
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
844
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
845
+ Params for bucket 24 (58720256 elements, 58720256 padded size):
846
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
847
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
848
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
849
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
850
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
851
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
852
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
853
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
854
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
855
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
856
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
857
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
858
+ Params for bucket 28 (58720256 elements, 58720256 padded size):
859
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
860
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
861
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
862
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
863
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
864
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
865
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
866
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
867
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
868
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
869
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
870
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
871
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
872
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
873
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
874
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
875
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
876
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
877
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
878
+ Params for bucket 34 (58720256 elements, 58720256 padded size):
879
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
880
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
881
+ Params for bucket 35 (58720256 elements, 58720256 padded size):
882
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
883
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
884
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
885
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
886
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
887
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
888
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
889
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
890
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
891
+ Params for bucket 38 (58720256 elements, 58720256 padded size):
892
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
893
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
894
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
895
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
896
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
897
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
898
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
899
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
900
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
901
+ Params for bucket 41 (58720256 elements, 58720256 padded size):
902
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
903
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
904
+ Params for bucket 42 (58720256 elements, 58720256 padded size):
905
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
906
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
907
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
908
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
909
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
910
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
911
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
912
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
913
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
914
+ Params for bucket 45 (58720256 elements, 58720256 padded size):
915
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
916
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
917
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
918
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
919
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
920
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
921
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
922
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
923
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
924
+ Params for bucket 48 (58720256 elements, 58720256 padded size):
925
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
926
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
927
+ Params for bucket 49 (58720256 elements, 58720256 padded size):
928
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
929
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
930
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
931
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
932
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
933
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
934
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
935
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
936
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
937
+ Params for bucket 52 (58720256 elements, 58720256 padded size):
938
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
939
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
940
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
941
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
942
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
943
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
944
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
945
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
946
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
947
+ Params for bucket 55 (58720256 elements, 58720256 padded size):
948
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
949
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
950
+ Params for bucket 56 (58720256 elements, 58720256 padded size):
951
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
952
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
953
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x73947901cfe0>, config_logger_dir='')
954
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
955
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:06:06
956
+ > building train, validation, and test datasets ...
957
+ > datasets target sizes (minimum size):
958
+ train: 192
959
+ validation: 0
960
+ test: 0
961
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
962
+ > building train, validation, and test datasets for GPT ...
963
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x739479859a30>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
964
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
965
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
966
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
967
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
968
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
969
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
970
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
971
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
972
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
973
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
974
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
975
+ > finished creating GPT datasets ...
976
+ [after dataloaders are built] datetime: 2026-08-30 07:06:06
977
+ done with setup ...
978
+ (min, max) time across ranks (ms):
979
+ model-and-optimizer-setup ......................: (3595.45, 3598.15)
980
+ train/valid/test-data-iterators-setup ..........: (73.52, 87.93)
981
+ training ...
982
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
983
+ [before the start of training step] datetime: 2026-08-30 07:06:06
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
985
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
989
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
991
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
992
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
993
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
994
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
995
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
996
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
997
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
998
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
999
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1000
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1001
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1002
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1003
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1004
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1005
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1006
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1007
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1008
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1009
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1010
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1011
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1012
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1013
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1014
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1015
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1016
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1017
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1018
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1019
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1020
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1021
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1022
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1023
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1024
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1025
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1026
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1027
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1028
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1029
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1030
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1031
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1032
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1033
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1034
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1035
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1036
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1037
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1038
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1039
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1040
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1041
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1042
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1043
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1044
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1045
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1046
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1047
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1048
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1049
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1050
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1051
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1052
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1053
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1054
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1055
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1056
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1057
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1058
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1059
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1060
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1061
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1062
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1063
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1064
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1065
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1066
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1067
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1068
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1069
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1070
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1071
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1072
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1073
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1074
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1075
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1076
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1077
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1078
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1079
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1080
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1081
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1082
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1083
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1084
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1085
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1086
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1087
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1088
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1089
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1090
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1091
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1092
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1093
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1094
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1095
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1096
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1097
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1098
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1099
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1100
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1101
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1102
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1103
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1104
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1105
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1106
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1107
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1108
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1109
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1110
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1111
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1112
+ [2026-08-30 07:06:32] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 25857.3 | throughput per GPU (TFLOP/s/GPU): 12.9 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103065E+01 | loss scale: 1.0 | grad norm: 44.385 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1113
+ Number of parameters in transformer block in billions: 23.22
1114
+ Number of parameters in embedding layers in billions: 0.41
1115
+ Total number of parameters in billions: 23.63
1116
+ Number of parameters in most loaded shard in billions: 23.6320
1117
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1118
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.0380859375 | max allocated: 66160.3212890625 | reserved: 68232.0 | max reserved: 68232.0
1119
+ [2026-08-30 07:06:33] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 559.2 | throughput per GPU (TFLOP/s/GPU): 596.2 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101734E+01 | loss scale: 1.0 | grad norm: 45.896 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1120
+ [2026-08-30 07:06:33] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 497.3 | throughput per GPU (TFLOP/s/GPU): 670.4 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.018462E+01 | loss scale: 1.0 | grad norm: 186.583 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1121
+ Capture CUDA graph for training!!!
1122
+ CUDA graph capture done!!!
1123
+ [2026-08-30 07:06:35] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2058.8 | throughput per GPU (TFLOP/s/GPU): 161.9 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.327620E+01 | loss scale: 1.0 | grad norm: 28.592 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1124
+ [2026-08-30 07:06:36] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 454.8 | throughput per GPU (TFLOP/s/GPU): 733.0 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.039153E+01 | loss scale: 1.0 | grad norm: 15.982 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1125
+ [2026-08-30 07:06:36] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 444.2 | throughput per GPU (TFLOP/s/GPU): 750.5 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.068049E+01 | loss scale: 1.0 | grad norm: 50.961 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1126
+ [2026-08-30 07:06:36] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 472.0 | throughput per GPU (TFLOP/s/GPU): 706.3 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.011438E+01 | loss scale: 1.0 | grad norm: 9.314 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1127
+ [2026-08-30 07:06:37] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 460.6 | throughput per GPU (TFLOP/s/GPU): 723.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.617073E+00 | loss scale: 1.0 | grad norm: 5.911 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1128
+ [2026-08-30 07:06:37] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 460.4 | throughput per GPU (TFLOP/s/GPU): 724.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.343796E+00 | loss scale: 1.0 | grad norm: 5.760 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1129
+ [2026-08-30 07:06:38] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 488.2 | throughput per GPU (TFLOP/s/GPU): 682.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.176932E+00 | loss scale: 1.0 | grad norm: 5.536 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1130
+ [2026-08-30 07:06:38] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 473.1 | throughput per GPU (TFLOP/s/GPU): 704.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.035903E+00 | loss scale: 1.0 | grad norm: 5.182 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1131
+ [2026-08-30 07:06:39] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 469.0 | throughput per GPU (TFLOP/s/GPU): 710.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.914059E+00 | loss scale: 1.0 | grad norm: 4.855 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1132
+ [after training is done] datetime: 2026-08-30 07:06:39
1133
+ [rank0]:[W830 07:06:40.679317228 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1134
+ [rank6]:[W830 07:06:41.050677300 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1135
+ [rank2]:[W830 07:06:41.053166456 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1136
+ [rank5]:[W830 07:06:41.053800639 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1137
+ [rank3]:[W830 07:06:41.054213364 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1138
+ [rank4]:[W830 07:06:41.055374114 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1139
+ [rank7]:[W830 07:06:42.245007472 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1140
+ [rank1]:[W830 07:06:42.613661641 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1141
+ W0830 07:06:47.651000 705505 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1142
+ W0830 07:06:47.651000 705505 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 705594 closing signal SIGTERM
1143
+ W0830 07:06:47.652000 705505 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 705600 closing signal SIGTERM
1144
+ Traceback (most recent call last):
1145
+ File "/usr/local/bin/torchrun", line 8, in <module>
1146
+ sys.exit(main())
1147
+ ^^^^^^
1148
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1149
+ return f(*args, **kwargs)
1150
+ ^^^^^^^^^^^^^^^^^^
1151
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1152
+ run(args)
1153
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1154
+ elastic_launch(
1155
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1156
+ return launch_agent(self._config, self._entrypoint, list(args))
1157
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1158
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1159
+ result = agent.run()
1160
+ ^^^^^^^^^^^
1161
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1162
+ result = f(*args, **kwargs)
1163
+ ^^^^^^^^^^^^^^^^^^
1164
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1165
+ result = self._invoke_run(role)
1166
+ ^^^^^^^^^^^^^^^^^^^^^^
1167
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1168
+ time.sleep(monitor_interval)
1169
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1170
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1171
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 705505 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3/mem_samples.csv ADDED
@@ -0,0 +1,224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 1
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 4
42
+ 1, 47
43
+ 2, 4
44
+ 3, 629
45
+ 4, 4
46
+ 5, 4
47
+ 6, 4
48
+ 7, 4
49
+ 0, 821
50
+ 1, 821
51
+ 2, 821
52
+ 3, 821
53
+ 4, 821
54
+ 5, 821
55
+ 6, 821
56
+ 7, 821
57
+ 0, 1895
58
+ 1, 1895
59
+ 2, 1895
60
+ 3, 1895
61
+ 4, 1895
62
+ 5, 1895
63
+ 6, 1895
64
+ 7, 1895
65
+ 0, 2379
66
+ 1, 2381
67
+ 2, 2391
68
+ 3, 2383
69
+ 4, 2369
70
+ 5, 2393
71
+ 6, 2411
72
+ 7, 2395
73
+ 0, 40937
74
+ 1, 40933
75
+ 2, 40933
76
+ 3, 40933
77
+ 4, 40933
78
+ 5, 40933
79
+ 6, 40933
80
+ 7, 40933
81
+ 0, 41377
82
+ 1, 41367
83
+ 2, 41367
84
+ 3, 41367
85
+ 4, 41367
86
+ 5, 41367
87
+ 6, 41367
88
+ 7, 41367
89
+ 0, 42679
90
+ 1, 42671
91
+ 2, 42671
92
+ 3, 42671
93
+ 4, 42671
94
+ 5, 42671
95
+ 6, 42671
96
+ 7, 42671
97
+ 0, 42679
98
+ 1, 42671
99
+ 2, 42671
100
+ 3, 42671
101
+ 4, 42671
102
+ 5, 42671
103
+ 6, 42671
104
+ 7, 42671
105
+ 0, 42957
106
+ 1, 42949
107
+ 2, 42949
108
+ 3, 42949
109
+ 4, 42949
110
+ 5, 42949
111
+ 6, 42949
112
+ 7, 42949
113
+ 0, 44381
114
+ 1, 44373
115
+ 2, 44373
116
+ 3, 44373
117
+ 4, 44373
118
+ 5, 44373
119
+ 6, 44373
120
+ 7, 44373
121
+ 0, 65637
122
+ 1, 65615
123
+ 2, 65615
124
+ 3, 65615
125
+ 4, 65629
126
+ 5, 65615
127
+ 6, 65615
128
+ 7, 65615
129
+ 0, 65637
130
+ 1, 65615
131
+ 2, 65615
132
+ 3, 65615
133
+ 4, 65629
134
+ 5, 65615
135
+ 6, 65615
136
+ 7, 65615
137
+ 0, 66123
138
+ 1, 66101
139
+ 2, 66101
140
+ 3, 66101
141
+ 4, 66115
142
+ 5, 66101
143
+ 6, 66101
144
+ 7, 66101
145
+ 0, 73981
146
+ 1, 73917
147
+ 2, 73913
148
+ 3, 73961
149
+ 4, 73975
150
+ 5, 73933
151
+ 6, 73967
152
+ 7, 73945
153
+ 0, 74983
154
+ 1, 74957
155
+ 2, 74981
156
+ 3, 74979
157
+ 4, 74973
158
+ 5, 74979
159
+ 6, 75013
160
+ 7, 74993
161
+ 0, 76325
162
+ 1, 76299
163
+ 2, 76299
164
+ 3, 76291
165
+ 4, 76239
166
+ 5, 76339
167
+ 6, 76353
168
+ 7, 76331
169
+ 0, 102681
170
+ 1, 70321
171
+ 2, 87803
172
+ 3, 71381
173
+ 4, 74847
174
+ 5, 75457
175
+ 6, 72305
176
+ 7, 70289
177
+ 0, 102691
178
+ 1, 102733
179
+ 2, 102733
180
+ 3, 102733
181
+ 4, 102627
182
+ 5, 102733
183
+ 6, 102733
184
+ 7, 102733
185
+ 0, 102691
186
+ 1, 102733
187
+ 2, 102733
188
+ 3, 102733
189
+ 4, 102627
190
+ 5, 102733
191
+ 6, 102733
192
+ 7, 102733
193
+ 0, 102163
194
+ 1, 102147
195
+ 2, 102467
196
+ 3, 102205
197
+ 4, 102361
198
+ 5, 102205
199
+ 6, 102147
200
+ 7, 102403
201
+ 0, 98857
202
+ 1, 99639
203
+ 2, 99313
204
+ 3, 98993
205
+ 4, 98951
206
+ 5, 99611
207
+ 6, 99251
208
+ 7, 99343
209
+ 0, 97267
210
+ 1, 97571
211
+ 2, 97571
212
+ 3, 97245
213
+ 4, 96883
214
+ 5, 97571
215
+ 6, 97309
216
+ 7, 97571
217
+ 0, 3
218
+ 1, 4
219
+ 2, 96537
220
+ 3, 4
221
+ 4, 4
222
+ 5, 4
223
+ 6, 4
224
+ 7, 4
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G2
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=3584
11
+ num_experts=32
12
+ topk=8
13
+ experts_per_rank=4
14
+ expert_params=44040192
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=32768
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.08203125
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3/train.log ADDED
@@ -0,0 +1,1170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G2 / ultraep (slots/rank=1 policy=one)
3
+ E=32 K=8 H'=3584 m=4
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.08203125 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:06:51.155000 708548 torch/distributed/run.py:766]
10
+ W0830 07:06:51.155000 708548 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:06:51.155000 708548 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:06:51.155000 708548 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
18
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
19
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
20
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
21
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
22
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
23
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
24
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
25
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
26
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
27
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
28
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
29
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
30
+ Number of virtual stages per pipeline stage: None
31
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
32
+ using torch.bfloat16 for parameters ...
33
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
34
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
35
+ ------------------------ arguments ------------------------
36
+ account_for_embedding_in_pipeline_split ......... False
37
+ account_for_loss_in_pipeline_split .............. False
38
+ accumulate_allreduce_grads_in_fp32 .............. True
39
+ adam_beta1 ...................................... 0.9
40
+ adam_beta2 ...................................... 0.999
41
+ adam_eps ........................................ 1e-08
42
+ add_bias_linear ................................. False
43
+ add_position_embedding .......................... False
44
+ add_qkv_bias .................................... False
45
+ adlr_autoresume ................................. False
46
+ adlr_autoresume_interval ........................ 1000
47
+ align_grad_reduce ............................... True
48
+ align_param_gather .............................. False
49
+ app_tag_run_name ................................ None
50
+ app_tag_run_version ............................. 0.0.0
51
+ apply_layernorm_1p .............................. False
52
+ apply_query_key_layer_scaling ................... False
53
+ apply_residual_connection_post_layernorm ........ False
54
+ apply_rope_fusion ............................... True
55
+ async_save ...................................... None
56
+ async_tensor_model_parallel_allreduce ........... True
57
+ attention_backend ............................... AttnBackend.auto
58
+ attention_dropout ............................... 0.0
59
+ attention_softmax_in_fp32 ....................... False
60
+ auto_detect_ckpt_format ......................... False
61
+ barrier_with_L1_time ............................ True
62
+ bert_binary_head ................................ True
63
+ bert_embedder_type .............................. megatron
64
+ bert_load ....................................... None
65
+ bf16 ............................................ True
66
+ bias_dropout_fusion ............................. True
67
+ bias_gelu_fusion ................................ False
68
+ bias_swiglu_fusion .............................. True
69
+ biencoder_projection_dim ........................ 0
70
+ biencoder_shared_query_context_model ............ False
71
+ block_data_path ................................. None
72
+ calc_ft_timeouts ................................ False
73
+ calculate_per_token_loss ........................ False
74
+ check_for_large_grads ........................... False
75
+ check_for_nan_in_loss_and_grad .................. False
76
+ check_for_spiky_loss ............................ False
77
+ check_weight_hash_across_dp_replicas_interval ... None
78
+ ckpt_assume_constant_structure .................. False
79
+ ckpt_convert_format ............................. None
80
+ ckpt_convert_save ............................... None
81
+ ckpt_convert_update_legacy_dist_opt_format ...... False
82
+ ckpt_format ..................................... torch_dist
83
+ ckpt_fully_parallel_load ........................ False
84
+ ckpt_fully_parallel_save ........................ True
85
+ ckpt_fully_parallel_save_deprecated ............. False
86
+ ckpt_step ....................................... None
87
+ classes_fraction ................................ 1.0
88
+ clip_grad ....................................... 1.0
89
+ clone_scatter_output_in_embedding ............... True
90
+ config_logger_dir ...............................
91
+ consumed_train_samples .......................... 0
92
+ consumed_valid_samples .......................... 0
93
+ context_parallel_size ........................... 1
94
+ cp_comm_type .................................... ['p2p']
95
+ create_attention_mask_in_dataloader ............. True
96
+ cross_entropy_fusion_impl ....................... native
97
+ cross_entropy_loss_fusion ....................... False
98
+ cuda_graph_scope ................................ full_iteration
99
+ cuda_graph_warmup_steps ......................... 3
100
+ data_args_path .................................. None
101
+ data_cache_path ................................. None
102
+ data_parallel_random_init ....................... False
103
+ data_parallel_sharding_strategy ................. no_shard
104
+ data_parallel_size .............................. 8
105
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
106
+ data_per_class_fraction ......................... 1.0
107
+ data_sharding ................................... True
108
+ dataloader_type ................................. single
109
+ ddp_average_in_collective ....................... False
110
+ ddp_bucket_size ................................. None
111
+ ddp_num_buckets ................................. None
112
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
113
+ decoder_first_pipeline_num_layers ............... None
114
+ decoder_last_pipeline_num_layers ................ None
115
+ decoder_num_layers .............................. None
116
+ decoder_seq_length .............................. None
117
+ decoupled_lr .................................... None
118
+ decoupled_min_lr ................................ None
119
+ decrease_batch_size_if_needed ................... False
120
+ defer_embedding_wgrad_compute ................... False
121
+ delay_wgrad_compute ............................. False
122
+ deprecated_use_mcore_models ..................... True
123
+ deterministic_mode .............................. False
124
+ dino_bottleneck_size ............................ 256
125
+ dino_freeze_last_layer .......................... 1
126
+ dino_head_hidden_size ........................... 2048
127
+ dino_local_crops_number ......................... 10
128
+ dino_local_img_size ............................. 96
129
+ dino_norm_last_layer ............................ False
130
+ dino_teacher_temp ............................... 0.07
131
+ dino_warmup_teacher_temp ........................ 0.04
132
+ dino_warmup_teacher_temp_epochs ................. 30
133
+ disable_bf16_reduced_precision_matmul ........... False
134
+ disable_mamba_mem_eff_path ...................... False
135
+ disable_straggler_on_startup .................... False
136
+ dist_ckpt_format_deprecated ..................... None
137
+ dist_ckpt_strictness ............................ assume_ok_unexpected
138
+ distribute_saved_activations .................... False
139
+ distributed_backend ............................. nccl
140
+ distributed_timeout_minutes ..................... 10
141
+ embedding_init_method_std ....................... None
142
+ embedding_path .................................. None
143
+ empty_unused_memory_level ....................... 0
144
+ enable_cuda_graph ............................... True
145
+ enable_experimental ............................. False
146
+ enable_ft_package ............................... False
147
+ enable_gloo_process_groups ...................... True
148
+ enable_msc ...................................... True
149
+ enable_one_logger ............................... True
150
+ encoder_num_layers .............................. 16
151
+ encoder_seq_length .............................. 4096
152
+ end_weight_decay ................................ 0.1
153
+ eod_mask_loss ................................... False
154
+ error_injection_rate ............................ 0
155
+ error_injection_type ............................ transient_error
156
+ eval_interval ................................... 1000000
157
+ eval_iters ...................................... 0
158
+ evidence_data_path .............................. None
159
+ exit_duration_in_mins ........................... None
160
+ exit_interval ................................... None
161
+ exit_on_missing_checkpoint ...................... False
162
+ exit_signal_handler ............................. False
163
+ exp_avg_dtype ................................... torch.float32
164
+ exp_avg_sq_dtype ................................ torch.float32
165
+ expert_model_parallel_size ...................... 8
166
+ expert_tensor_parallel_size ..................... 1
167
+ export_force_local_attention .................... False
168
+ export_kd_cfg ................................... None
169
+ export_kd_teacher_ckpt_format ................... None
170
+ export_kd_teacher_load .......................... None
171
+ export_kv_cache_quant ........................... False
172
+ export_legacy_megatron .......................... False
173
+ export_model_type ............................... GPTModel
174
+ export_moe_apply_probs_on_input ................. False
175
+ export_qk_l2_norm ............................... False
176
+ export_quant_cfg ................................ None
177
+ export_real_quant_cfg ........................... None
178
+ export_te_mcore_model ........................... False
179
+ external_cuda_graph ............................. False
180
+ ffn_hidden_size ................................. 14336
181
+ finetune ........................................ False
182
+ finetune_data_split ............................. train
183
+ finetune_hf_dataset ............................. None
184
+ first_last_layers_bf16 .......................... False
185
+ flash_decode .................................... False
186
+ fp16 ............................................ False
187
+ fp16_lm_cross_entropy ........................... False
188
+ fp32_residual_connection ........................ False
189
+ fp8 ............................................. e4m3
190
+ fp8_amax_compute_algo ........................... most_recent
191
+ fp8_amax_history_len ............................ 1
192
+ fp8_interval .................................... 1
193
+ fp8_margin ...................................... 0
194
+ fp8_param_gather ................................ False
195
+ fp8_recipe ...................................... mxfp8
196
+ fp8_wgrad ....................................... True
197
+ fsdp_double_buffer .............................. False
198
+ global_batch_size ............................... 16
199
+ grad_reduce_in_bf16 ............................. False
200
+ gradient_accumulation_fusion .................... True
201
+ gradient_reduce_div_fusion ...................... True
202
+ group_query_attention ........................... True
203
+ head_lr_mult .................................... 1.0
204
+ heterogeneous_layers_config_encoded_json ........ None
205
+ heterogeneous_layers_config_path ................ None
206
+ hidden_dropout .................................. 0.0
207
+ hidden_size ..................................... 4096
208
+ hierarchical_context_parallel_sizes ............. None
209
+ high_priority_stream_groups ..................... []
210
+ hybrid_attention_ratio .......................... 0.0
211
+ hybrid_mlp_ratio ................................ 0.0
212
+ hybrid_override_pattern ......................... None
213
+ hysteresis ...................................... 2
214
+ ict_head_size ................................... None
215
+ ict_load ........................................ None
216
+ img_h ........................................... 224
217
+ img_w ........................................... 224
218
+ indexer_batch_size .............................. 128
219
+ indexer_log_interval ............................ 1000
220
+ inference_batch_times_seqlen_threshold .......... -1
221
+ inference_dynamic_batching ...................... False
222
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
223
+ inference_dynamic_batching_buffer_overflow_factor None
224
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
225
+ inference_dynamic_batching_chunk_size ........... 256
226
+ inference_dynamic_batching_max_requests_override None
227
+ inference_dynamic_batching_max_tokens_override .. None
228
+ inference_dynamic_batching_num_cuda_graphs ...... 16
229
+ inference_max_batch_size ........................ 8
230
+ inference_max_seq_length ........................ 2560
231
+ inference_rng_tracker ........................... False
232
+ init_method_std ................................. 0.01
233
+ init_method_xavier_uniform ...................... False
234
+ init_model_with_meta_device ..................... False
235
+ initial_loss_scale .............................. 4294967296
236
+ inprocess_active_world_size ..................... 8
237
+ inprocess_barrier_timeout ....................... 120
238
+ inprocess_completion_timeout .................... 120
239
+ inprocess_empty_cuda_cache ...................... False
240
+ inprocess_granularity ........................... node
241
+ inprocess_hard_timeout .......................... 90
242
+ inprocess_heartbeat_interval .................... 30
243
+ inprocess_heartbeat_timeout ..................... 60
244
+ inprocess_last_call_wait ........................ 1
245
+ inprocess_max_iterations ........................ None
246
+ inprocess_monitor_process_interval .............. 1.0
247
+ inprocess_monitor_thread_interval ............... 1.0
248
+ inprocess_progress_watchdog_interval ............ 1.0
249
+ inprocess_restart ............................... False
250
+ inprocess_soft_timeout .......................... 60
251
+ inprocess_termination_grace_time ................ 1
252
+ is_hybrid_model ................................. False
253
+ iter_per_epoch .................................. 1250
254
+ iterations_to_skip .............................. []
255
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
256
+ kitchen_config_file ............................. None
257
+ kitchen_recipe_number ........................... None
258
+ kv_channels ..................................... 128
259
+ kv_lora_rank .................................... 32
260
+ lazy_mpu_init ................................... None
261
+ load ............................................ None
262
+ load_main_params_from_ckpt ...................... None
263
+ load_model_opt_format ........................... False
264
+ local_rank ...................................... 0
265
+ log_energy ...................................... False
266
+ log_interval .................................... 1
267
+ log_loss_scale_to_tensorboard ................... True
268
+ log_memory_to_tensorboard ....................... False
269
+ log_num_zeros_in_grad ........................... False
270
+ log_params_norm ................................. False
271
+ log_progress .................................... False
272
+ log_straggler ................................... False
273
+ log_throughput .................................. True
274
+ log_timers_to_tensorboard ....................... False
275
+ log_validation_ppl_to_tensorboard ............... False
276
+ log_world_size_to_tensorboard ................... False
277
+ logging_level ................................... None
278
+ loss_scale ...................................... None
279
+ loss_scale_window ............................... 1000
280
+ lr .............................................. 0.0001
281
+ lr_decay_iters .................................. 8
282
+ lr_decay_samples ................................ None
283
+ lr_decay_style .................................. cosine
284
+ lr_warmup_fraction .............................. None
285
+ lr_warmup_init .................................. 0.0
286
+ lr_warmup_iters ................................. 5
287
+ lr_warmup_samples ............................... 0
288
+ lr_wsd_decay_iters .............................. None
289
+ lr_wsd_decay_samples ............................ None
290
+ lr_wsd_decay_style .............................. exponential
291
+ main_grads_dtype ................................ torch.float32
292
+ main_params_dtype ............................... torch.float32
293
+ make_vocab_size_divisible_by .................... 128
294
+ mamba_head_dim .................................. 64
295
+ mamba_num_groups ................................ 8
296
+ mamba_num_heads ................................. None
297
+ mamba_state_dim ................................. 128
298
+ manual_gc ....................................... False
299
+ manual_gc_eval .................................. True
300
+ manual_gc_interval .............................. 0
301
+ mask_factor ..................................... 1.0
302
+ mask_prob ....................................... 0.15
303
+ mask_type ....................................... random
304
+ masked_softmax_fusion ........................... False
305
+ max_position_embeddings ......................... 32768
306
+ max_tokens_to_oom ............................... 12000
307
+ memory_snapshot_path ............................ snapshot.pickle
308
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
309
+ micro_batch_size ................................ 1
310
+ microbatch_group_size_per_vp_stage .............. None
311
+ mid_level_dataset_surplus ....................... 0.005
312
+ min_loss_scale .................................. 1.0
313
+ min_lr .......................................... 1e-05
314
+ mlp_chunks_for_prefill .......................... 1
315
+ mmap_bin_files .................................. True
316
+ mock_data ....................................... False
317
+ moe_apply_probs_on_input ........................ False
318
+ moe_aux_loss_coeff .............................. 0.01
319
+ moe_capacity_monitor ............................ True
320
+ moe_deepep_num_sms .............................. 20
321
+ moe_echo_enable_random_offloading ............... False
322
+ moe_echo_expert_dispatch_overlap ................ False
323
+ moe_echo_expert_dispatcher_type ................. hybridep
324
+ moe_echo_recompute_expert_dispatch .............. False
325
+ moe_enable_deepep ............................... False
326
+ moe_enable_echo ................................. False
327
+ moe_enable_hybridep ............................. True
328
+ moe_enable_ultraep .............................. True
329
+ moe_expert_capacity_factor ...................... None
330
+ moe_extended_tp ................................. False
331
+ moe_ffn_hidden_size ............................. 3584
332
+ moe_grouped_gemm ................................ True
333
+ moe_hybridep_num_sms ............................ 16
334
+ moe_input_jitter_eps ............................ None
335
+ moe_layer_freq .................................. 1
336
+ moe_layer_recompute ............................. False
337
+ moe_num_echo_experts ............................ None
338
+ moe_num_redundant_experts_per_rank .............. 1
339
+ moe_pad_expert_input_to_capacity ................ False
340
+ moe_per_layer_logging ........................... False
341
+ moe_permute_fusion .............................. False
342
+ moe_received_token_capacity ..................... 1.08203125
343
+ moe_router_bias_update_rate ..................... 0.001
344
+ moe_router_dtype ................................ fp32
345
+ moe_router_enable_expert_bias ................... False
346
+ moe_router_force_load_balancing ................. False
347
+ moe_router_group_topk ........................... None
348
+ moe_router_load_balancing_type .................. aux_loss
349
+ moe_router_num_groups ........................... None
350
+ moe_router_padding_for_fp8 ...................... False
351
+ moe_router_pre_softmax .......................... False
352
+ moe_router_score_function ....................... softmax
353
+ moe_router_topk ................................. 8
354
+ moe_router_topk_scaling_factor .................. None
355
+ moe_shared_expert_intermediate_size ............. None
356
+ moe_shared_expert_overlap ....................... False
357
+ moe_token_dispatcher_type ....................... flex
358
+ moe_token_drop_policy ........................... probs
359
+ moe_ultraep_max_microbatches .................... None
360
+ moe_ultraep_overlap_weight_sync ................. True
361
+ moe_ultraep_weight_sync_join .................... before_dispatch
362
+ moe_upcycling_granularity ....................... 1
363
+ moe_use_device_initiated_grouped_gemm ........... True
364
+ moe_use_legacy_grouped_gemm ..................... False
365
+ moe_use_upcycling ............................... False
366
+ moe_z_loss_coeff ................................ None
367
+ mrope_section ................................... None
368
+ mscale .......................................... 1.0
369
+ mscale_all_dim .................................. 0.0
370
+ mtp_loss_scaling_factor ......................... 0.1
371
+ mtp_num_layers .................................. None
372
+ multi_latent_attention .......................... False
373
+ nccl_all_reduce_for_prefill ..................... False
374
+ nccl_communicator_config_path ................... None
375
+ nccl_ub ......................................... False
376
+ no_load_optim ................................... None
377
+ no_load_rng ..................................... None
378
+ no_persist_layer_norm ........................... False
379
+ no_rope_freq .................................... None
380
+ no_save_optim ................................... None
381
+ no_save_rng ..................................... None
382
+ non_persistent_ckpt_type ........................ None
383
+ non_persistent_global_ckpt_dir .................. None
384
+ non_persistent_local_ckpt_algo .................. fully_parallel
385
+ non_persistent_local_ckpt_dir ................... None
386
+ non_persistent_save_interval .................... None
387
+ norm_epsilon .................................... 1e-05
388
+ normalization ................................... RMSNorm
389
+ num_attention_heads ............................. 32
390
+ num_channels .................................... 3
391
+ num_classes ..................................... 1000
392
+ num_dataset_builder_threads ..................... 1
393
+ num_distributed_optimizer_instances ............. 1
394
+ num_experts ..................................... 32
395
+ num_layers ...................................... 16
396
+ num_layers_at_end_in_bf16 ....................... 1
397
+ num_layers_at_start_in_bf16 ..................... 1
398
+ num_layers_per_virtual_pipeline_stage ........... None
399
+ num_query_groups ................................ 8
400
+ num_virtual_stages_per_pipeline_rank ............ None
401
+ num_workers ..................................... 2
402
+ object_storage_cache_path ....................... None
403
+ one_logger_async ................................ False
404
+ one_logger_project .............................. megatron-lm
405
+ one_logger_run_name ............................. None
406
+ onnx_safe ....................................... None
407
+ openai_gelu ..................................... False
408
+ optimizer ....................................... adam
409
+ optimizer_cpu_offload ........................... False
410
+ optimizer_offload_fraction ...................... 1.0
411
+ output_bert_embeddings .......................... False
412
+ overlap_cpu_optimizer_d2h_h2d ................... False
413
+ overlap_grad_reduce ............................. True
414
+ overlap_moe_expert_parallel_comm ................ False
415
+ overlap_p2p_comm ................................ False
416
+ overlap_p2p_comm_warmup_flush ................... False
417
+ overlap_param_gather ............................ True
418
+ overlap_param_gather_with_optimizer_step ........ False
419
+ override_opt_param_scheduler .................... False
420
+ params_dtype .................................... torch.bfloat16
421
+ patch_dim ....................................... 16
422
+ per_split_data_args_path ........................ None
423
+ perform_initialization .......................... True
424
+ pin_cpu_grads ................................... True
425
+ pin_cpu_params .................................. True
426
+ pipeline_model_parallel_comm_backend ............ None
427
+ pipeline_model_parallel_layout .................. None
428
+ pipeline_model_parallel_size .................... 1
429
+ position_embedding_type ......................... rope
430
+ pretrained_checkpoint ........................... None
431
+ profile ......................................... False
432
+ profile_ranks ................................... [0]
433
+ profile_step_end ................................ 12
434
+ profile_step_start .............................. 10
435
+ q_lora_rank ..................................... None
436
+ qk_head_dim ..................................... 128
437
+ qk_l2_norm ...................................... False
438
+ qk_layernorm .................................... False
439
+ qk_pos_emb_head_dim ............................. 64
440
+ query_in_block_prob ............................. 0.1
441
+ rampup_batch_size ............................... None
442
+ rank ............................................ 0
443
+ recompute_granularity ........................... None
444
+ recompute_method ................................ None
445
+ recompute_modules ............................... None
446
+ recompute_num_layers ............................ None
447
+ record_memory_history ........................... False
448
+ relative_attention_max_distance ................. 128
449
+ relative_attention_num_buckets .................. 32
450
+ replication ..................................... False
451
+ replication_factor .............................. 2
452
+ replication_jump ................................ None
453
+ rerun_mode ...................................... validate_results
454
+ reset_attention_mask ............................ False
455
+ reset_position_ids .............................. False
456
+ result_rejected_tracker_filename ................ None
457
+ retriever_report_topk_accuracies ................ []
458
+ retriever_score_scaling ......................... False
459
+ retriever_seq_length ............................ 256
460
+ retro_add_retriever ............................. False
461
+ retro_attention_gate ............................ 1
462
+ retro_cyclic_train_iters ........................ None
463
+ retro_encoder_attention_dropout ................. 0.1
464
+ retro_encoder_hidden_dropout .................... 0.1
465
+ retro_encoder_layers ............................ 2
466
+ retro_num_neighbors ............................. 2
467
+ retro_num_retrieved_chunks ...................... 2
468
+ retro_project_dir ............................... None
469
+ retro_verify_neighbor_count ..................... True
470
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
471
+ rope_scaling_factor ............................. 8.0
472
+ rope_type ....................................... None
473
+ rotary_base ..................................... 1000000
474
+ rotary_interleaved .............................. False
475
+ rotary_percent .................................. 1.0
476
+ rotary_scaling_factor ........................... 1.0
477
+ rotary_seq_len_interpolation_factor ............. None
478
+ run_workload_inspector_server ................... False
479
+ sample_rate ..................................... 1.0
480
+ save ............................................ None
481
+ save_interval ................................... None
482
+ save_retain_interval ............................ None
483
+ scatter_gather_tensors_in_pipeline .............. True
484
+ seed ............................................ 1234
485
+ seq_length ...................................... 4096
486
+ sequence_parallel ............................... False
487
+ sft ............................................. False
488
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
489
+ sgd_momentum .................................... 0.9
490
+ short_seq_prob .................................. 0.1
491
+ skip_train ...................................... False
492
+ skipped_train_samples ........................... 0
493
+ spec ............................................ None
494
+ split ........................................... 100,0,0
495
+ squared_relu .................................... False
496
+ start_weight_decay .............................. 0.1
497
+ straggler_ctrlr_port ............................ 65535
498
+ straggler_minmax_count .......................... 1
499
+ suggested_communication_unit_size ............... None
500
+ swiglu .......................................... True
501
+ swin_backbone_type .............................. tiny
502
+ symmetric_ar_type ............................... None
503
+ te_rng_tracker .................................. True
504
+ tensor_model_parallel_size ...................... 1
505
+ tensorboard_dir ................................. None
506
+ tensorboard_log_interval ........................ 1
507
+ tensorboard_queue_size .......................... 1000
508
+ test_data_path .................................. None
509
+ test_mode ....................................... False
510
+ tiktoken_num_special_tokens ..................... 1000
511
+ tiktoken_pattern ................................ None
512
+ tiktoken_special_tokens ......................... None
513
+ timing_log_level ................................ 0
514
+ timing_log_option ............................... minmax
515
+ titles_data_path ................................ None
516
+ tokenizer_model ................................. None
517
+ tokenizer_type .................................. GPT2BPETokenizer
518
+ torch_fsdp2_reshard_after_forward ............... True
519
+ tp_comm_bootstrap_backend ....................... nccl
520
+ tp_comm_bulk_dgrad .............................. True
521
+ tp_comm_bulk_wgrad .............................. True
522
+ tp_comm_overlap ................................. False
523
+ tp_comm_overlap_ag .............................. True
524
+ tp_comm_overlap_cfg ............................. None
525
+ tp_comm_overlap_rs .............................. True
526
+ tp_comm_overlap_rs_dgrad ........................ False
527
+ tp_comm_split_ag ................................ True
528
+ tp_comm_split_rs ................................ True
529
+ train_data_path ................................. None
530
+ train_iters ..................................... 12
531
+ train_samples ................................... None
532
+ train_sync_interval ............................. None
533
+ transformer_impl ................................ transformer_engine
534
+ transformer_pipeline_model_parallel_size ........ 1
535
+ untie_embeddings_and_output_weights ............. True
536
+ use_checkpoint_args ............................. False
537
+ use_checkpoint_opt_param_scheduler .............. False
538
+ use_cpu_initialization .......................... None
539
+ use_custom_fsdp ................................. False
540
+ use_dist_ckpt ................................... True
541
+ use_dist_ckpt_deprecated ........................ False
542
+ use_distributed_optimizer ....................... True
543
+ use_flash_attn .................................. False
544
+ use_legacy_models ............................... False
545
+ use_mp_args_from_checkpoint_args ................ False
546
+ use_one_sent_docs ............................... False
547
+ use_persistent_ckpt_worker ...................... False
548
+ use_precision_aware_optimizer ................... False
549
+ use_pytorch_profiler ............................ False
550
+ use_ring_exchange_p2p ........................... False
551
+ use_rope_scaling ................................ False
552
+ use_rotary_position_embeddings .................. False
553
+ use_sharp ....................................... False
554
+ use_tokenizer_model_from_checkpoint_args ........ True
555
+ use_torch_fsdp2 ................................. False
556
+ use_torch_optimizer_for_cpu_offload ............. False
557
+ use_tp_pp_dp_mapping ............................ False
558
+ v_head_dim ...................................... 128
559
+ valid_data_path ................................. None
560
+ variable_seq_lengths ............................ False
561
+ virtual_pipeline_model_parallel_size ............ None
562
+ vision_backbone_type ............................ vit
563
+ vision_pretraining .............................. False
564
+ vision_pretraining_type ......................... classify
565
+ vocab_extra_ids ................................. 0
566
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
567
+ vocab_size ...................................... None
568
+ wandb_exp_name ..................................
569
+ wandb_project ...................................
570
+ wandb_save_dir ..................................
571
+ weight_decay .................................... 0.1
572
+ weight_decay_incr_style ......................... constant
573
+ wgrad_deferral_limit ............................ 0
574
+ world_size ...................................... 8
575
+ yaml_cfg ........................................ None
576
+ -------------------- end of arguments ---------------------
577
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
578
+ > building GPT2BPETokenizer tokenizer ...
579
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
580
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
581
+ > initializing torch distributed ...
582
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank 475 is connected to is connected to is connected to 7767 peer ranks. peer ranks. is connected to peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 77 peer ranks. 7
587
+
588
+ Expected number of connected peer ranks is :
589
+ 7
590
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : [Gloo] Rank 70
591
+ is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
597
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
604
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
607
+ [Gloo] Rank [Gloo] Rank 00 is connected to is connected to 00 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 00
608
+
609
+ [rank7]:[W830 07:07:09.224128572 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank2]:[W830 07:07:09.224131452 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank5]:[W830 07:07:09.224138479 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank4]:[W830 07:07:09.224189373 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank3]:[W830 07:07:09.224230549 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank6]:[W830 07:07:09.224283256 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank1]:[W830 07:07:09.224320826 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.035 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:07:09.388227031 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.815 seconds
625
+ time to initialize megatron (seconds): 7.290
626
+ [after megatron is initialized] datetime: 2026-08-30 07:07:14
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 4608626688
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (42086400 elements, 42086400 padded size):
670
+ module.decoder.layers.15.self_attention.linear_qkv.weight
671
+ module.decoder.layers.15.pre_mlp_layernorm.weight
672
+ module.decoder.final_layernorm.weight
673
+ module.decoder.layers.15.mlp.router.weight
674
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
675
+ module.decoder.layers.15.self_attention.linear_proj.weight
676
+ Params for bucket 3 (42082304 elements, 42082304 padded size):
677
+ module.decoder.layers.14.self_attention.linear_qkv.weight
678
+ module.decoder.layers.14.pre_mlp_layernorm.weight
679
+ module.decoder.layers.14.mlp.router.weight
680
+ module.decoder.layers.14.self_attention.linear_proj.weight
681
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
682
+ Params for bucket 4 (42082304 elements, 42082304 padded size):
683
+ module.decoder.layers.13.pre_mlp_layernorm.weight
684
+ module.decoder.layers.13.self_attention.linear_qkv.weight
685
+ module.decoder.layers.13.mlp.router.weight
686
+ module.decoder.layers.13.self_attention.linear_proj.weight
687
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
688
+ Params for bucket 5 (42082304 elements, 42082304 padded size):
689
+ module.decoder.layers.12.mlp.router.weight
690
+ module.decoder.layers.12.self_attention.linear_proj.weight
691
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
692
+ module.decoder.layers.12.pre_mlp_layernorm.weight
693
+ module.decoder.layers.12.self_attention.linear_qkv.weight
694
+ Params for bucket 6 (42082304 elements, 42082304 padded size):
695
+ module.decoder.layers.11.mlp.router.weight
696
+ module.decoder.layers.11.self_attention.linear_proj.weight
697
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.11.self_attention.linear_qkv.weight
699
+ module.decoder.layers.11.pre_mlp_layernorm.weight
700
+ Params for bucket 7 (42082304 elements, 42082304 padded size):
701
+ module.decoder.layers.10.mlp.router.weight
702
+ module.decoder.layers.10.self_attention.linear_proj.weight
703
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
704
+ module.decoder.layers.10.self_attention.linear_qkv.weight
705
+ module.decoder.layers.10.pre_mlp_layernorm.weight
706
+ Params for bucket 8 (42082304 elements, 42082304 padded size):
707
+ module.decoder.layers.9.mlp.router.weight
708
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
709
+ module.decoder.layers.9.self_attention.linear_proj.weight
710
+ module.decoder.layers.9.self_attention.linear_qkv.weight
711
+ module.decoder.layers.9.pre_mlp_layernorm.weight
712
+ Params for bucket 9 (42082304 elements, 42082304 padded size):
713
+ module.decoder.layers.8.mlp.router.weight
714
+ module.decoder.layers.8.self_attention.linear_proj.weight
715
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
716
+ module.decoder.layers.8.self_attention.linear_qkv.weight
717
+ module.decoder.layers.8.pre_mlp_layernorm.weight
718
+ Params for bucket 10 (42082304 elements, 42082304 padded size):
719
+ module.decoder.layers.7.mlp.router.weight
720
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
721
+ module.decoder.layers.7.self_attention.linear_proj.weight
722
+ module.decoder.layers.7.pre_mlp_layernorm.weight
723
+ module.decoder.layers.7.self_attention.linear_qkv.weight
724
+ Params for bucket 11 (42082304 elements, 42082304 padded size):
725
+ module.decoder.layers.6.mlp.router.weight
726
+ module.decoder.layers.6.self_attention.linear_proj.weight
727
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
728
+ module.decoder.layers.6.self_attention.linear_qkv.weight
729
+ module.decoder.layers.6.pre_mlp_layernorm.weight
730
+ Params for bucket 12 (42082304 elements, 42082304 padded size):
731
+ module.decoder.layers.5.self_attention.linear_qkv.weight
732
+ module.decoder.layers.5.self_attention.linear_proj.weight
733
+ module.decoder.layers.5.pre_mlp_layernorm.weight
734
+ module.decoder.layers.5.mlp.router.weight
735
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
736
+ Params for bucket 13 (42082304 elements, 42082304 padded size):
737
+ module.decoder.layers.4.self_attention.linear_qkv.weight
738
+ module.decoder.layers.4.pre_mlp_layernorm.weight
739
+ module.decoder.layers.4.mlp.router.weight
740
+ module.decoder.layers.4.self_attention.linear_proj.weight
741
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
742
+ Params for bucket 14 (42082304 elements, 42082304 padded size):
743
+ module.decoder.layers.3.mlp.router.weight
744
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
745
+ module.decoder.layers.3.self_attention.linear_proj.weight
746
+ module.decoder.layers.3.self_attention.linear_qkv.weight
747
+ module.decoder.layers.3.pre_mlp_layernorm.weight
748
+ Params for bucket 15 (42082304 elements, 42082304 padded size):
749
+ module.decoder.layers.2.mlp.router.weight
750
+ module.decoder.layers.2.self_attention.linear_proj.weight
751
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
752
+ module.decoder.layers.2.pre_mlp_layernorm.weight
753
+ module.decoder.layers.2.self_attention.linear_qkv.weight
754
+ Params for bucket 16 (42082304 elements, 42082304 padded size):
755
+ module.decoder.layers.1.mlp.router.weight
756
+ module.decoder.layers.1.self_attention.linear_proj.weight
757
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
758
+ module.decoder.layers.1.pre_mlp_layernorm.weight
759
+ module.decoder.layers.1.self_attention.linear_qkv.weight
760
+ Params for bucket 17 (42082304 elements, 42082304 padded size):
761
+ module.decoder.layers.0.mlp.router.weight
762
+ module.decoder.layers.0.pre_mlp_layernorm.weight
763
+ module.decoder.layers.0.self_attention.linear_proj.weight
764
+ module.decoder.layers.0.self_attention.linear_qkv.weight
765
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56
769
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
771
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
772
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
773
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
774
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
775
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
776
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
777
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
778
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
779
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
780
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
781
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
782
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
783
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
784
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
785
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
786
+ Params for bucket 6 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
788
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
789
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
790
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
791
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
792
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
793
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
794
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
795
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
796
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
797
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
798
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
799
+ Params for bucket 10 (58720256 elements, 58720256 padded size):
800
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
801
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
802
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
803
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
804
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
805
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
806
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
807
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
808
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
809
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
811
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
812
+ Params for bucket 14 (58720256 elements, 58720256 padded size):
813
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
814
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
815
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
816
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
817
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
818
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
819
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
820
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
821
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
822
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
823
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
824
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
825
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
827
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
828
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
829
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
830
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
831
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
832
+ Params for bucket 20 (58720256 elements, 58720256 padded size):
833
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
834
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
835
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
836
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
837
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
838
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
839
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
840
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
841
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
842
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
843
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
844
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
845
+ Params for bucket 24 (58720256 elements, 58720256 padded size):
846
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
847
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
848
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
849
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
850
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
851
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
852
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
853
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
854
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
855
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
856
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
857
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
858
+ Params for bucket 28 (58720256 elements, 58720256 padded size):
859
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
860
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
861
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
862
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
863
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
864
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
865
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
866
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
867
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
868
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
869
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
870
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
871
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
872
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
873
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
874
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
875
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
876
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
877
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
878
+ Params for bucket 34 (58720256 elements, 58720256 padded size):
879
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
880
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
881
+ Params for bucket 35 (58720256 elements, 58720256 padded size):
882
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
883
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
884
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
885
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
886
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
887
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
888
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
889
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
890
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
891
+ Params for bucket 38 (58720256 elements, 58720256 padded size):
892
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
893
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
894
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
895
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
896
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
897
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
898
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
899
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
900
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
901
+ Params for bucket 41 (58720256 elements, 58720256 padded size):
902
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
903
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
904
+ Params for bucket 42 (58720256 elements, 58720256 padded size):
905
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
906
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
907
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
908
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
909
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
910
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
911
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
912
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
913
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
914
+ Params for bucket 45 (58720256 elements, 58720256 padded size):
915
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
916
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
917
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
918
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
919
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
920
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
921
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
922
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
923
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
924
+ Params for bucket 48 (58720256 elements, 58720256 padded size):
925
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
926
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
927
+ Params for bucket 49 (58720256 elements, 58720256 padded size):
928
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
929
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
930
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
931
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
932
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
933
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
934
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
935
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
936
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
937
+ Params for bucket 52 (58720256 elements, 58720256 padded size):
938
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
939
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
940
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
941
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
942
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
943
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
944
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
945
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
946
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
947
+ Params for bucket 55 (58720256 elements, 58720256 padded size):
948
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
949
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
950
+ Params for bucket 56 (58720256 elements, 58720256 padded size):
951
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
952
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
953
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x7a6c8351d100>, config_logger_dir='')
954
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
955
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:07:18
956
+ > building train, validation, and test datasets ...
957
+ > datasets target sizes (minimum size):
958
+ train: 192
959
+ validation: 0
960
+ test: 0
961
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
962
+ > building train, validation, and test datasets for GPT ...
963
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x7a6c8364e840>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
964
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
965
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
966
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
967
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
968
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
969
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
970
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
971
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
972
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
973
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
974
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
975
+ > finished creating GPT datasets ...
976
+ [after dataloaders are built] datetime: 2026-08-30 07:07:18
977
+ done with setup ...
978
+ training ...
979
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
980
+ (min, max) time across ranks (ms):
981
+ model-and-optimizer-setup ......................: (3546.01, 3548.48)
982
+ train/valid/test-data-iterators-setup ..........: (71.73, 90.26)
983
+ [before the start of training step] datetime: 2026-08-30 07:07:18
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
985
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
989
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
991
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
992
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
993
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
994
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
995
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
996
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
997
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
998
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
999
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1000
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1001
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1002
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1003
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1004
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1005
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1006
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1007
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1008
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1009
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1010
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1011
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1012
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1013
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1014
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1015
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1016
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1017
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1018
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1019
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1020
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1021
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1022
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1023
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1024
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1025
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1026
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1027
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1028
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1029
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1030
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1031
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1032
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1033
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1034
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1035
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1036
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1037
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1038
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1039
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1040
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1041
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1042
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1043
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1044
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1045
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1046
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1047
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1048
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1049
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1050
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1051
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1052
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1053
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1054
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1055
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1056
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1057
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1058
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1059
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1060
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1061
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1062
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1063
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1064
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1065
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1066
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1067
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1068
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1069
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1070
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1071
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1072
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1073
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1074
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1075
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1076
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1077
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1078
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1079
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1080
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1081
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1082
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1083
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1084
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1085
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1086
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1087
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1088
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1089
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1090
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1091
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1092
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1093
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1094
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1095
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1096
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1097
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1098
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1099
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1100
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1101
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1102
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1103
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1104
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1105
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1106
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1107
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1108
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1109
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1110
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1111
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1112
+ Number of parameters in transformer block in billions: 23.22 [2026-08-30 07:07:44] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 25768.3 | throughput per GPU (TFLOP/s/GPU): 12.9 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103065E+01 | loss scale: 1.0 | grad norm: 44.385 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1113
+
1114
+ Number of parameters in embedding layers in billions: 0.41
1115
+ Total number of parameters in billions: 23.63
1116
+ Number of parameters in most loaded shard in billions: 23.6320
1117
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1118
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.0380859375 | max allocated: 66160.3212890625 | reserved: 68232.0 | max reserved: 68232.0
1119
+ [2026-08-30 07:07:45] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 616.2 | throughput per GPU (TFLOP/s/GPU): 541.0 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101734E+01 | loss scale: 1.0 | grad norm: 45.897 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1120
+ [2026-08-30 07:07:45] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 497.8 | throughput per GPU (TFLOP/s/GPU): 669.7 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.018430E+01 | loss scale: 1.0 | grad norm: 186.616 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1121
+ Capture CUDA graph for training!!!
1122
+ CUDA graph capture done!!!
1123
+ [2026-08-30 07:07:47] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2062.9 | throughput per GPU (TFLOP/s/GPU): 161.6 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.327590E+01 | loss scale: 1.0 | grad norm: 28.609 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1124
+ [2026-08-30 07:07:48] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 454.1 | throughput per GPU (TFLOP/s/GPU): 734.2 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.039147E+01 | loss scale: 1.0 | grad norm: 15.985 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1125
+ [2026-08-30 07:07:48] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 444.7 | throughput per GPU (TFLOP/s/GPU): 749.7 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.068062E+01 | loss scale: 1.0 | grad norm: 50.942 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1126
+ [2026-08-30 07:07:48] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 471.6 | throughput per GPU (TFLOP/s/GPU): 706.9 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.011609E+01 | loss scale: 1.0 | grad norm: 9.320 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1127
+ [2026-08-30 07:07:49] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 459.3 | throughput per GPU (TFLOP/s/GPU): 725.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.619381E+00 | loss scale: 1.0 | grad norm: 5.912 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1128
+ [2026-08-30 07:07:49] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 467.3 | throughput per GPU (TFLOP/s/GPU): 713.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.344950E+00 | loss scale: 1.0 | grad norm: 5.757 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1129
+ [2026-08-30 07:07:50] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 486.7 | throughput per GPU (TFLOP/s/GPU): 685.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.178471E+00 | loss scale: 1.0 | grad norm: 5.543 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1130
+ [2026-08-30 07:07:50] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 463.4 | throughput per GPU (TFLOP/s/GPU): 719.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.037644E+00 | loss scale: 1.0 | grad norm: 5.172 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1131
+ [2026-08-30 07:07:51] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 473.1 | throughput per GPU (TFLOP/s/GPU): 704.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.917351E+00 | loss scale: 1.0 | grad norm: 4.862 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 35456 | min safe capacity: 1.0820 |
1132
+ [after training is done] datetime: 2026-08-30 07:07:51
1133
+ [rank0]:[W830 07:07:52.651989654 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1134
+ [rank6]:[W830 07:07:54.261946158 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1135
+ [rank1]:[W830 07:07:54.266108151 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1136
+ [rank5]:[W830 07:07:54.270284120 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1137
+ [rank2]:[W830 07:07:54.271271285 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1138
+ [rank4]:[W830 07:07:54.671261233 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1139
+ [rank3]:[W830 07:07:54.683205981 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1140
+ [rank7]:[W830 07:07:54.803309674 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1141
+ W0830 07:07:59.840000 708548 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1142
+ W0830 07:07:59.840000 708548 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 708638 closing signal SIGTERM
1143
+ Traceback (most recent call last):
1144
+ File "/usr/local/bin/torchrun", line 8, in <module>
1145
+ sys.exit(main())
1146
+ ^^^^^^
1147
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1148
+ return f(*args, **kwargs)
1149
+ ^^^^^^^^^^^^^^^^^^
1150
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1151
+ run(args)
1152
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1153
+ elastic_launch(
1154
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1155
+ return launch_agent(self._config, self._entrypoint, list(args))
1156
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1157
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1158
+ result = agent.run()
1159
+ ^^^^^^^^^^^
1160
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1161
+ result = f(*args, **kwargs)
1162
+ ^^^^^^^^^^^^^^^^^^
1163
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1164
+ result = self._invoke_run(role)
1165
+ ^^^^^^^^^^^^^^^^^^^^^^
1166
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1167
+ time.sleep(monitor_interval)
1168
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1169
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1170
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 708548 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1/mem_samples.csv ADDED
@@ -0,0 +1,232 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 1
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 4
42
+ 1, 629
43
+ 2, 629
44
+ 3, 629
45
+ 4, 629
46
+ 5, 629
47
+ 6, 629
48
+ 7, 165
49
+ 0, 867
50
+ 1, 873
51
+ 2, 879
52
+ 3, 863
53
+ 4, 877
54
+ 5, 871
55
+ 6, 877
56
+ 7, 901
57
+ 0, 1895
58
+ 1, 1895
59
+ 2, 1895
60
+ 3, 1895
61
+ 4, 1895
62
+ 5, 1895
63
+ 6, 1895
64
+ 7, 1895
65
+ 0, 2371
66
+ 1, 2353
67
+ 2, 2373
68
+ 3, 2363
69
+ 4, 2379
70
+ 5, 2365
71
+ 6, 2393
72
+ 7, 2375
73
+ 0, 33823
74
+ 1, 33823
75
+ 2, 33823
76
+ 3, 33823
77
+ 4, 33823
78
+ 5, 33823
79
+ 6, 33823
80
+ 7, 33823
81
+ 0, 40935
82
+ 1, 40925
83
+ 2, 40925
84
+ 3, 40925
85
+ 4, 40925
86
+ 5, 40925
87
+ 6, 40925
88
+ 7, 40925
89
+ 0, 41977
90
+ 1, 41961
91
+ 2, 41961
92
+ 3, 41961
93
+ 4, 41961
94
+ 5, 41961
95
+ 6, 41961
96
+ 7, 41961
97
+ 0, 42251
98
+ 1, 42235
99
+ 2, 42235
100
+ 3, 42235
101
+ 4, 42235
102
+ 5, 42235
103
+ 6, 42235
104
+ 7, 42235
105
+ 0, 42801
106
+ 1, 42785
107
+ 2, 42785
108
+ 3, 42785
109
+ 4, 42785
110
+ 5, 42785
111
+ 6, 42785
112
+ 7, 42785
113
+ 0, 44637
114
+ 1, 44621
115
+ 2, 44621
116
+ 3, 44621
117
+ 4, 44621
118
+ 5, 44621
119
+ 6, 44621
120
+ 7, 44621
121
+ 0, 63691
122
+ 1, 63695
123
+ 2, 63693
124
+ 3, 64219
125
+ 4, 63707
126
+ 5, 63723
127
+ 6, 64219
128
+ 7, 63709
129
+ 0, 67631
130
+ 1, 67649
131
+ 2, 67649
132
+ 3, 67649
133
+ 4, 67703
134
+ 5, 67649
135
+ 6, 67649
136
+ 7, 67649
137
+ 0, 67631
138
+ 1, 67649
139
+ 2, 67649
140
+ 3, 67649
141
+ 4, 67703
142
+ 5, 67649
143
+ 6, 67649
144
+ 7, 67649
145
+ 0, 74729
146
+ 1, 74733
147
+ 2, 74733
148
+ 3, 74733
149
+ 4, 74787
150
+ 5, 74733
151
+ 6, 74733
152
+ 7, 74733
153
+ 0, 75619
154
+ 1, 75629
155
+ 2, 75693
156
+ 3, 75655
157
+ 4, 75789
158
+ 5, 75645
159
+ 6, 75649
160
+ 7, 75667
161
+ 0, 76663
162
+ 1, 76647
163
+ 2, 76651
164
+ 3, 76621
165
+ 4, 76729
166
+ 5, 76679
167
+ 6, 76679
168
+ 7, 76665
169
+ 0, 99955
170
+ 1, 100037
171
+ 2, 100037
172
+ 3, 100037
173
+ 4, 100005
174
+ 5, 100037
175
+ 6, 100037
176
+ 7, 100037
177
+ 0, 101939
178
+ 1, 101827
179
+ 2, 101827
180
+ 3, 101827
181
+ 4, 102043
182
+ 5, 101827
183
+ 6, 101827
184
+ 7, 101827
185
+ 0, 101939
186
+ 1, 101827
187
+ 2, 101827
188
+ 3, 101827
189
+ 4, 102043
190
+ 5, 101827
191
+ 6, 101827
192
+ 7, 101827
193
+ 0, 101939
194
+ 1, 101827
195
+ 2, 101827
196
+ 3, 101827
197
+ 4, 102043
198
+ 5, 101827
199
+ 6, 101827
200
+ 7, 101827
201
+ 0, 99481
202
+ 1, 99369
203
+ 2, 99369
204
+ 3, 99369
205
+ 4, 99585
206
+ 5, 99369
207
+ 6, 99369
208
+ 7, 99369
209
+ 0, 97801
210
+ 1, 97689
211
+ 2, 97301
212
+ 3, 97689
213
+ 4, 97905
214
+ 5, 97301
215
+ 6, 97689
216
+ 7, 97301
217
+ 0, 95733
218
+ 1, 96009
219
+ 2, 4
220
+ 3, 95621
221
+ 4, 95837
222
+ 5, 95621
223
+ 6, 96009
224
+ 7, 4
225
+ 0, 1
226
+ 1, 95618
227
+ 2, 1
228
+ 3, 1
229
+ 4, 1
230
+ 5, 1
231
+ 6, 1
232
+ 7, 4
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G3
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=1792
11
+ num_experts=64
12
+ topk=16
13
+ experts_per_rank=8
14
+ expert_params=22020096
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=65536
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.072265625
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1/train.log ADDED
@@ -0,0 +1,1308 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G3 / ultraep (slots/rank=1 policy=one)
3
+ E=64 K=16 H'=1792 m=8
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.072265625 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:08:03.352000 711599 torch/distributed/run.py:766]
10
+ W0830 07:08:03.352000 711599 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:08:03.352000 711599 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:08:03.352000 711599 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
18
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
19
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
20
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
21
+ Number of virtual stages per pipeline stage: None
22
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
23
+ using torch.bfloat16 for parameters ...
24
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
25
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
26
+ ------------------------ arguments ------------------------
27
+ account_for_embedding_in_pipeline_split ......... False
28
+ account_for_loss_in_pipeline_split .............. False
29
+ accumulate_allreduce_grads_in_fp32 .............. True
30
+ adam_beta1 ...................................... 0.9
31
+ adam_beta2 ...................................... 0.999
32
+ adam_eps ........................................ 1e-08
33
+ add_bias_linear ................................. False
34
+ add_position_embedding .......................... False
35
+ add_qkv_bias .................................... False
36
+ adlr_autoresume ................................. False
37
+ adlr_autoresume_interval ........................ 1000
38
+ align_grad_reduce ............................... True
39
+ align_param_gather .............................. False
40
+ app_tag_run_name ................................ None
41
+ app_tag_run_version ............................. 0.0.0
42
+ apply_layernorm_1p .............................. False
43
+ apply_query_key_layer_scaling ................... False
44
+ apply_residual_connection_post_layernorm ........ False
45
+ apply_rope_fusion ............................... True
46
+ async_save ...................................... None
47
+ async_tensor_model_parallel_allreduce ........... True
48
+ attention_backend ............................... AttnBackend.auto
49
+ attention_dropout ............................... 0.0
50
+ attention_softmax_in_fp32 ....................... False
51
+ auto_detect_ckpt_format ......................... False
52
+ barrier_with_L1_time ............................ True
53
+ bert_binary_head ................................ True
54
+ bert_embedder_type .............................. megatron
55
+ bert_load ....................................... None
56
+ bf16 ............................................ True
57
+ bias_dropout_fusion ............................. True
58
+ bias_gelu_fusion ................................ False
59
+ bias_swiglu_fusion .............................. True
60
+ biencoder_projection_dim ........................ 0
61
+ biencoder_shared_query_context_model ............ False
62
+ block_data_path ................................. None
63
+ calc_ft_timeouts ................................ False
64
+ calculate_per_token_loss ........................ False
65
+ check_for_large_grads ........................... False
66
+ check_for_nan_in_loss_and_grad .................. False
67
+ check_for_spiky_loss ............................ False
68
+ check_weight_hash_across_dp_replicas_interval ... None
69
+ ckpt_assume_constant_structure .................. False
70
+ ckpt_convert_format ............................. None
71
+ ckpt_convert_save ............................... None
72
+ ckpt_convert_update_legacy_dist_opt_format ...... False
73
+ ckpt_format ..................................... torch_dist
74
+ ckpt_fully_parallel_load ........................ False
75
+ ckpt_fully_parallel_save ........................ True
76
+ ckpt_fully_parallel_save_deprecated ............. False
77
+ ckpt_step ....................................... None
78
+ classes_fraction ................................ 1.0
79
+ clip_grad ....................................... 1.0
80
+ clone_scatter_output_in_embedding ............... True
81
+ config_logger_dir ...............................
82
+ consumed_train_samples .......................... 0
83
+ consumed_valid_samples .......................... 0
84
+ context_parallel_size ........................... 1
85
+ cp_comm_type .................................... ['p2p']
86
+ create_attention_mask_in_dataloader ............. True
87
+ cross_entropy_fusion_impl ....................... native
88
+ cross_entropy_loss_fusion ....................... False
89
+ cuda_graph_scope ................................ full_iteration
90
+ cuda_graph_warmup_steps ......................... 3
91
+ data_args_path .................................. None
92
+ data_cache_path ................................. None
93
+ data_parallel_random_init ....................... False
94
+ data_parallel_sharding_strategy ................. no_shard
95
+ data_parallel_size .............................. 8
96
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
97
+ data_per_class_fraction ......................... 1.0
98
+ data_sharding ................................... True
99
+ dataloader_type ................................. single
100
+ ddp_average_in_collective ....................... False
101
+ ddp_bucket_size ................................. None
102
+ ddp_num_buckets ................................. None
103
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
104
+ decoder_first_pipeline_num_layers ............... None
105
+ decoder_last_pipeline_num_layers ................ None
106
+ decoder_num_layers .............................. None
107
+ decoder_seq_length .............................. None
108
+ decoupled_lr .................................... None
109
+ decoupled_min_lr ................................ None
110
+ decrease_batch_size_if_needed ................... False
111
+ defer_embedding_wgrad_compute ................... False
112
+ delay_wgrad_compute ............................. False
113
+ deprecated_use_mcore_models ..................... True
114
+ deterministic_mode .............................. False
115
+ dino_bottleneck_size ............................ 256
116
+ dino_freeze_last_layer .......................... 1
117
+ dino_head_hidden_size ........................... 2048
118
+ dino_local_crops_number ......................... 10
119
+ dino_local_img_size ............................. 96
120
+ dino_norm_last_layer ............................ False
121
+ dino_teacher_temp ............................... 0.07
122
+ dino_warmup_teacher_temp ........................ 0.04
123
+ dino_warmup_teacher_temp_epochs ................. 30
124
+ disable_bf16_reduced_precision_matmul ........... False
125
+ disable_mamba_mem_eff_path ...................... False
126
+ disable_straggler_on_startup .................... False
127
+ dist_ckpt_format_deprecated ..................... None
128
+ dist_ckpt_strictness ............................ assume_ok_unexpected
129
+ distribute_saved_activations .................... False
130
+ distributed_backend ............................. nccl
131
+ distributed_timeout_minutes ..................... 10
132
+ embedding_init_method_std ....................... None
133
+ embedding_path .................................. None
134
+ empty_unused_memory_level ....................... 0
135
+ enable_cuda_graph ............................... True
136
+ enable_experimental ............................. False
137
+ enable_ft_package ............................... False
138
+ enable_gloo_process_groups ...................... True
139
+ enable_msc ...................................... True
140
+ enable_one_logger ............................... True
141
+ encoder_num_layers .............................. 16
142
+ encoder_seq_length .............................. 4096
143
+ end_weight_decay ................................ 0.1
144
+ eod_mask_loss ................................... False
145
+ error_injection_rate ............................ 0
146
+ error_injection_type ............................ transient_error
147
+ eval_interval ................................... 1000000
148
+ eval_iters ...................................... 0
149
+ evidence_data_path .............................. None
150
+ exit_duration_in_mins ........................... None
151
+ exit_interval ................................... None
152
+ exit_on_missing_checkpoint ...................... False
153
+ exit_signal_handler ............................. False
154
+ exp_avg_dtype ................................... torch.float32
155
+ exp_avg_sq_dtype ................................ torch.float32
156
+ expert_model_parallel_size ...................... 8
157
+ expert_tensor_parallel_size ..................... 1
158
+ export_force_local_attention .................... False
159
+ export_kd_cfg ................................... None
160
+ export_kd_teacher_ckpt_format ................... None
161
+ export_kd_teacher_load .......................... None
162
+ export_kv_cache_quant ........................... False
163
+ export_legacy_megatron .......................... False
164
+ export_model_type ............................... GPTModel
165
+ export_moe_apply_probs_on_input ................. False
166
+ export_qk_l2_norm ............................... False
167
+ export_quant_cfg ................................ None
168
+ export_real_quant_cfg ........................... None
169
+ export_te_mcore_model ........................... False
170
+ external_cuda_graph ............................. False
171
+ ffn_hidden_size ................................. 14336
172
+ finetune ........................................ False
173
+ finetune_data_split ............................. train
174
+ finetune_hf_dataset ............................. None
175
+ first_last_layers_bf16 .......................... False
176
+ flash_decode .................................... False
177
+ fp16 ............................................ False
178
+ fp16_lm_cross_entropy ........................... False
179
+ fp32_residual_connection ........................ False
180
+ fp8 ............................................. e4m3
181
+ fp8_amax_compute_algo ........................... most_recent
182
+ fp8_amax_history_len ............................ 1
183
+ fp8_interval .................................... 1
184
+ fp8_margin ...................................... 0
185
+ fp8_param_gather ................................ False
186
+ fp8_recipe ...................................... mxfp8
187
+ fp8_wgrad ....................................... True
188
+ fsdp_double_buffer .............................. False
189
+ global_batch_size ............................... 16
190
+ grad_reduce_in_bf16 ............................. False
191
+ gradient_accumulation_fusion .................... True
192
+ gradient_reduce_div_fusion ...................... True
193
+ group_query_attention ........................... True
194
+ head_lr_mult .................................... 1.0
195
+ heterogeneous_layers_config_encoded_json ........ None
196
+ heterogeneous_layers_config_path ................ None
197
+ hidden_dropout .................................. 0.0
198
+ hidden_size ..................................... 4096
199
+ hierarchical_context_parallel_sizes ............. None
200
+ high_priority_stream_groups ..................... []
201
+ hybrid_attention_ratio .......................... 0.0
202
+ hybrid_mlp_ratio ................................ 0.0
203
+ hybrid_override_pattern ......................... None
204
+ hysteresis ...................................... 2
205
+ ict_head_size ................................... None
206
+ ict_load ........................................ None
207
+ img_h ........................................... 224
208
+ img_w ........................................... 224
209
+ indexer_batch_size .............................. 128
210
+ indexer_log_interval ............................ 1000
211
+ inference_batch_times_seqlen_threshold .......... -1
212
+ inference_dynamic_batching ...................... False
213
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
214
+ inference_dynamic_batching_buffer_overflow_factor None
215
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
216
+ inference_dynamic_batching_chunk_size ........... 256
217
+ inference_dynamic_batching_max_requests_override None
218
+ inference_dynamic_batching_max_tokens_override .. None
219
+ inference_dynamic_batching_num_cuda_graphs ...... 16
220
+ inference_max_batch_size ........................ 8
221
+ inference_max_seq_length ........................ 2560
222
+ inference_rng_tracker ........................... False
223
+ init_method_std ................................. 0.01
224
+ init_method_xavier_uniform ...................... False
225
+ init_model_with_meta_device ..................... False
226
+ initial_loss_scale .............................. 4294967296
227
+ inprocess_active_world_size ..................... 8
228
+ inprocess_barrier_timeout ....................... 120
229
+ inprocess_completion_timeout .................... 120
230
+ inprocess_empty_cuda_cache ...................... False
231
+ inprocess_granularity ........................... node
232
+ inprocess_hard_timeout .......................... 90
233
+ inprocess_heartbeat_interval .................... 30
234
+ inprocess_heartbeat_timeout ..................... 60
235
+ inprocess_last_call_wait ........................ 1
236
+ inprocess_max_iterations ........................ None
237
+ inprocess_monitor_process_interval .............. 1.0
238
+ inprocess_monitor_thread_interval ............... 1.0
239
+ inprocess_progress_watchdog_interval ............ 1.0
240
+ inprocess_restart ............................... False
241
+ inprocess_soft_timeout .......................... 60
242
+ inprocess_termination_grace_time ................ 1
243
+ is_hybrid_model ................................. False
244
+ iter_per_epoch .................................. 1250
245
+ iterations_to_skip .............................. []
246
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
247
+ kitchen_config_file ............................. None
248
+ kitchen_recipe_number ........................... None
249
+ kv_channels ..................................... 128
250
+ kv_lora_rank .................................... 32
251
+ lazy_mpu_init ................................... None
252
+ load ............................................ None
253
+ load_main_params_from_ckpt ...................... None
254
+ load_model_opt_format ........................... False
255
+ local_rank ...................................... 0
256
+ log_energy ...................................... False
257
+ log_interval .................................... 1
258
+ log_loss_scale_to_tensorboard ................... True
259
+ log_memory_to_tensorboard ....................... False
260
+ log_num_zeros_in_grad ........................... False
261
+ log_params_norm ................................. False
262
+ log_progress .................................... False
263
+ log_straggler ................................... False
264
+ log_throughput .................................. True
265
+ log_timers_to_tensorboard ....................... False
266
+ log_validation_ppl_to_tensorboard ............... False
267
+ log_world_size_to_tensorboard ................... False
268
+ logging_level ................................... None
269
+ loss_scale ...................................... None
270
+ loss_scale_window ............................... 1000
271
+ lr .............................................. 0.0001
272
+ lr_decay_iters .................................. 8
273
+ lr_decay_samples ................................ None
274
+ lr_decay_style .................................. cosine
275
+ lr_warmup_fraction .............................. None
276
+ lr_warmup_init .................................. 0.0
277
+ lr_warmup_iters ................................. 5
278
+ lr_warmup_samples ............................... 0
279
+ lr_wsd_decay_iters .............................. None
280
+ lr_wsd_decay_samples ............................ None
281
+ lr_wsd_decay_style .............................. exponential
282
+ main_grads_dtype ................................ torch.float32
283
+ main_params_dtype ............................... torch.float32
284
+ make_vocab_size_divisible_by .................... 128
285
+ mamba_head_dim .................................. 64
286
+ mamba_num_groups ................................ 8
287
+ mamba_num_heads ................................. None
288
+ mamba_state_dim ................................. 128
289
+ manual_gc ....................................... False
290
+ manual_gc_eval .................................. True
291
+ manual_gc_interval .............................. 0
292
+ mask_factor ..................................... 1.0
293
+ mask_prob ....................................... 0.15
294
+ mask_type ....................................... random
295
+ masked_softmax_fusion ........................... False
296
+ max_position_embeddings ......................... 32768
297
+ max_tokens_to_oom ............................... 12000
298
+ memory_snapshot_path ............................ snapshot.pickle
299
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
300
+ micro_batch_size ................................ 1
301
+ microbatch_group_size_per_vp_stage .............. None
302
+ mid_level_dataset_surplus ....................... 0.005
303
+ min_loss_scale .................................. 1.0
304
+ min_lr .......................................... 1e-05
305
+ mlp_chunks_for_prefill .......................... 1
306
+ mmap_bin_files .................................. True
307
+ mock_data ....................................... False
308
+ moe_apply_probs_on_input ........................ False
309
+ moe_aux_loss_coeff .............................. 0.01
310
+ moe_capacity_monitor ............................ True
311
+ moe_deepep_num_sms .............................. 20
312
+ moe_echo_enable_random_offloading ............... False
313
+ moe_echo_expert_dispatch_overlap ................ False
314
+ moe_echo_expert_dispatcher_type ................. hybridep
315
+ moe_echo_recompute_expert_dispatch .............. False
316
+ moe_enable_deepep ............................... False
317
+ moe_enable_echo ................................. False
318
+ moe_enable_hybridep ............................. True
319
+ moe_enable_ultraep .............................. True
320
+ moe_expert_capacity_factor ...................... None
321
+ moe_extended_tp ................................. False
322
+ moe_ffn_hidden_size ............................. 1792
323
+ moe_grouped_gemm ................................ True
324
+ moe_hybridep_num_sms ............................ 16
325
+ moe_input_jitter_eps ............................ None
326
+ moe_layer_freq .................................. 1
327
+ moe_layer_recompute ............................. False
328
+ moe_num_echo_experts ............................ None
329
+ moe_num_redundant_experts_per_rank .............. 1
330
+ moe_pad_expert_input_to_capacity ................ False
331
+ moe_per_layer_logging ........................... False
332
+ moe_permute_fusion .............................. False
333
+ moe_received_token_capacity ..................... 1.072265625
334
+ moe_router_bias_update_rate ..................... 0.001
335
+ moe_router_dtype ................................ fp32
336
+ moe_router_enable_expert_bias ................... False
337
+ moe_router_force_load_balancing ................. False
338
+ moe_router_group_topk ........................... None
339
+ moe_router_load_balancing_type .................. aux_loss
340
+ moe_router_num_groups ........................... None
341
+ moe_router_padding_for_fp8 ...................... False
342
+ moe_router_pre_softmax .......................... False
343
+ moe_router_score_function ....................... softmax
344
+ moe_router_topk ................................. 16
345
+ moe_router_topk_scaling_factor .................. None
346
+ moe_shared_expert_intermediate_size ............. None
347
+ moe_shared_expert_overlap ....................... False
348
+ moe_token_dispatcher_type ....................... flex
349
+ moe_token_drop_policy ........................... probs
350
+ moe_ultraep_max_microbatches .................... None
351
+ moe_ultraep_overlap_weight_sync ................. True
352
+ moe_ultraep_weight_sync_join .................... before_dispatch
353
+ moe_upcycling_granularity ....................... 1
354
+ moe_use_device_initiated_grouped_gemm ........... True
355
+ moe_use_legacy_grouped_gemm ..................... False
356
+ moe_use_upcycling ............................... False
357
+ moe_z_loss_coeff ................................ None
358
+ mrope_section ................................... None
359
+ mscale .......................................... 1.0
360
+ mscale_all_dim .................................. 0.0
361
+ mtp_loss_scaling_factor ......................... 0.1
362
+ mtp_num_layers .................................. None
363
+ multi_latent_attention .......................... False
364
+ nccl_all_reduce_for_prefill ..................... False
365
+ nccl_communicator_config_path ................... None
366
+ nccl_ub ......................................... False
367
+ no_load_optim ................................... None
368
+ no_load_rng ..................................... None
369
+ no_persist_layer_norm ........................... False
370
+ no_rope_freq .................................... None
371
+ no_save_optim ................................... None
372
+ no_save_rng ..................................... None
373
+ non_persistent_ckpt_type ........................ None
374
+ non_persistent_global_ckpt_dir .................. None
375
+ non_persistent_local_ckpt_algo .................. fully_parallel
376
+ non_persistent_local_ckpt_dir ................... None
377
+ non_persistent_save_interval .................... None
378
+ norm_epsilon .................................... 1e-05
379
+ normalization ................................... RMSNorm
380
+ num_attention_heads ............................. 32
381
+ num_channels .................................... 3
382
+ num_classes ..................................... 1000
383
+ num_dataset_builder_threads ..................... 1
384
+ num_distributed_optimizer_instances ............. 1
385
+ num_experts ..................................... 64
386
+ num_layers ...................................... 16
387
+ num_layers_at_end_in_bf16 ....................... 1
388
+ num_layers_at_start_in_bf16 ..................... 1
389
+ num_layers_per_virtual_pipeline_stage ........... None
390
+ num_query_groups ................................ 8
391
+ num_virtual_stages_per_pipeline_rank ............ None
392
+ num_workers ..................................... 2
393
+ object_storage_cache_path ....................... None
394
+ one_logger_async ................................ False
395
+ one_logger_project .............................. megatron-lm
396
+ one_logger_run_name ............................. None
397
+ onnx_safe ....................................... None
398
+ openai_gelu ..................................... False
399
+ optimizer ....................................... adam
400
+ optimizer_cpu_offload ........................... False
401
+ optimizer_offload_fraction ...................... 1.0
402
+ output_bert_embeddings .......................... False
403
+ overlap_cpu_optimizer_d2h_h2d ................... False
404
+ overlap_grad_reduce ............................. True
405
+ overlap_moe_expert_parallel_comm ................ False
406
+ overlap_p2p_comm ................................ False
407
+ overlap_p2p_comm_warmup_flush ................... False
408
+ overlap_param_gather ............................ True
409
+ overlap_param_gather_with_optimizer_step ........ False
410
+ override_opt_param_scheduler .................... False
411
+ params_dtype .................................... torch.bfloat16
412
+ patch_dim ....................................... 16
413
+ per_split_data_args_path ........................ None
414
+ perform_initialization .......................... True
415
+ pin_cpu_grads ................................... True
416
+ pin_cpu_params .................................. True
417
+ pipeline_model_parallel_comm_backend ............ None
418
+ pipeline_model_parallel_layout .................. None
419
+ pipeline_model_parallel_size .................... 1
420
+ position_embedding_type ......................... rope
421
+ pretrained_checkpoint ........................... None
422
+ profile ......................................... False
423
+ profile_ranks ................................... [0]
424
+ profile_step_end ................................ 12
425
+ profile_step_start .............................. 10
426
+ q_lora_rank ..................................... None
427
+ qk_head_dim ..................................... 128
428
+ qk_l2_norm ...................................... False
429
+ qk_layernorm .................................... False
430
+ qk_pos_emb_head_dim ............................. 64
431
+ query_in_block_prob ............................. 0.1
432
+ rampup_batch_size ............................... None
433
+ rank ............................................ 0
434
+ recompute_granularity ........................... None
435
+ recompute_method ................................ None
436
+ recompute_modules ............................... None
437
+ recompute_num_layers ............................ None
438
+ record_memory_history ........................... False
439
+ relative_attention_max_distance ................. 128
440
+ relative_attention_num_buckets .................. 32
441
+ replication ..................................... False
442
+ replication_factor .............................. 2
443
+ replication_jump ................................ None
444
+ rerun_mode ...................................... validate_results
445
+ reset_attention_mask ............................ False
446
+ reset_position_ids .............................. False
447
+ result_rejected_tracker_filename ................ None
448
+ retriever_report_topk_accuracies ................ []
449
+ retriever_score_scaling ......................... False
450
+ retriever_seq_length ............................ 256
451
+ retro_add_retriever ............................. False
452
+ retro_attention_gate ............................ 1
453
+ retro_cyclic_train_iters ........................ None
454
+ retro_encoder_attention_dropout ................. 0.1
455
+ retro_encoder_hidden_dropout .................... 0.1
456
+ retro_encoder_layers ............................ 2
457
+ retro_num_neighbors ............................. 2
458
+ retro_num_retrieved_chunks ...................... 2
459
+ retro_project_dir ............................... None
460
+ retro_verify_neighbor_count ..................... True
461
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
462
+ rope_scaling_factor ............................. 8.0
463
+ rope_type ....................................... None
464
+ rotary_base ..................................... 1000000
465
+ rotary_interleaved .............................. False
466
+ rotary_percent .................................. 1.0
467
+ rotary_scaling_factor ........................... 1.0
468
+ rotary_seq_len_interpolation_factor ............. None
469
+ run_workload_inspector_server ................... False
470
+ sample_rate ..................................... 1.0
471
+ save ............................................ None
472
+ save_interval ................................... None
473
+ save_retain_interval ............................ None
474
+ scatter_gather_tensors_in_pipeline .............. True
475
+ seed ............................................ 1234
476
+ seq_length ...................................... 4096
477
+ sequence_parallel ............................... False
478
+ sft ............................................. False
479
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
480
+ sgd_momentum .................................... 0.9
481
+ short_seq_prob .................................. 0.1
482
+ skip_train ...................................... False
483
+ skipped_train_samples ........................... 0
484
+ spec ............................................ None
485
+ split ........................................... 100,0,0
486
+ squared_relu .................................... False
487
+ start_weight_decay .............................. 0.1
488
+ straggler_ctrlr_port ............................ 65535
489
+ straggler_minmax_count .......................... 1
490
+ suggested_communication_unit_size ............... None
491
+ swiglu .......................................... True
492
+ swin_backbone_type .............................. tiny
493
+ symmetric_ar_type ............................... None
494
+ te_rng_tracker .................................. True
495
+ tensor_model_parallel_size ...................... 1
496
+ tensorboard_dir ................................. None
497
+ tensorboard_log_interval ........................ 1
498
+ tensorboard_queue_size .......................... 1000
499
+ test_data_path .................................. None
500
+ test_mode ....................................... False
501
+ tiktoken_num_special_tokens ..................... 1000
502
+ tiktoken_pattern ................................ None
503
+ tiktoken_special_tokens ......................... None
504
+ timing_log_level ................................ 0
505
+ timing_log_option ............................... minmax
506
+ titles_data_path ................................ None
507
+ tokenizer_model ................................. None
508
+ tokenizer_type .................................. GPT2BPETokenizer
509
+ torch_fsdp2_reshard_after_forward ............... True
510
+ tp_comm_bootstrap_backend ....................... nccl
511
+ tp_comm_bulk_dgrad .............................. True
512
+ tp_comm_bulk_wgrad .............................. True
513
+ tp_comm_overlap ................................. False
514
+ tp_comm_overlap_ag .............................. True
515
+ tp_comm_overlap_cfg ............................. None
516
+ tp_comm_overlap_rs .............................. True
517
+ tp_comm_overlap_rs_dgrad ........................ False
518
+ tp_comm_split_ag ................................ True
519
+ tp_comm_split_rs ................................ True
520
+ train_data_path ................................. None
521
+ train_iters ..................................... 12
522
+ train_samples ................................... None
523
+ train_sync_interval ............................. None
524
+ transformer_impl ................................ transformer_engine
525
+ transformer_pipeline_model_parallel_size ........ 1
526
+ untie_embeddings_and_output_weights ............. True
527
+ use_checkpoint_args ............................. False
528
+ use_checkpoint_opt_param_scheduler .............. False
529
+ use_cpu_initialization .......................... None
530
+ use_custom_fsdp ................................. False
531
+ use_dist_ckpt ................................... True
532
+ use_dist_ckpt_deprecated ........................ False
533
+ use_distributed_optimizer ....................... True
534
+ use_flash_attn .................................. False
535
+ use_legacy_models ............................... False
536
+ use_mp_args_from_checkpoint_args ................ False
537
+ use_one_sent_docs ............................... False
538
+ use_persistent_ckpt_worker ...................... False
539
+ use_precision_aware_optimizer ................... False
540
+ use_pytorch_profiler ............................ False
541
+ use_ring_exchange_p2p ........................... False
542
+ use_rope_scaling ................................ False
543
+ use_rotary_position_embeddings .................. False
544
+ use_sharp ....................................... False
545
+ use_tokenizer_model_from_checkpoint_args ........ True
546
+ use_torch_fsdp2 ................................. False
547
+ use_torch_optimizer_for_cpu_offload ............. False
548
+ use_tp_pp_dp_mapping ............................ False
549
+ v_head_dim ...................................... 128
550
+ valid_data_path ................................. None
551
+ variable_seq_lengths ............................ False
552
+ virtual_pipeline_model_parallel_size ............ None
553
+ vision_backbone_type ............................ vit
554
+ vision_pretraining .............................. False
555
+ vision_pretraining_type ......................... classify
556
+ vocab_extra_ids ................................. 0
557
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
558
+ vocab_size ...................................... None
559
+ wandb_exp_name ..................................
560
+ wandb_project ...................................
561
+ wandb_save_dir ..................................
562
+ weight_decay .................................... 0.1
563
+ weight_decay_incr_style ......................... constant
564
+ wgrad_deferral_limit ............................ 0
565
+ world_size ...................................... 8
566
+ yaml_cfg ........................................ None
567
+ -------------------- end of arguments ---------------------
568
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
569
+ > building GPT2BPETokenizer tokenizer ...
570
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
571
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
572
+ > initializing torch distributed ...
573
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
574
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
575
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
576
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
577
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
578
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
579
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
580
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
581
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
582
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank 012 is connected to 3 is connected to is connected to 7 is connected to 77 peer ranks. 7 peer ranks. peer ranks. Expected number of connected peer ranks is : peer ranks. [Gloo] Rank Expected number of connected peer ranks is : Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 747
584
+
585
+ 7
586
+ is connected to
587
+ 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank [Gloo] Rank 25 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77
594
+
595
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank [Gloo] Rank 7 is connected to 67 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
597
+ 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
604
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
607
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank6]:[W830 07:08:21.576415366 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank3]:[W830 07:08:21.578756987 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank7]:[W830 07:08:21.578769311 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank5]:[W830 07:08:21.584423291 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank2]:[W830 07:08:21.584425377 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank1]:[W830 07:08:21.584462554 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank4]:[W830 07:08:21.584487382 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.035 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:08:21.741419669 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.838 seconds
625
+ time to initialize megatron (seconds): 7.356
626
+ [after megatron is initialized] datetime: 2026-08-30 07:08:27
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 4258402304
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (42217472 elements, 42217472 padded size):
670
+ module.decoder.final_layernorm.weight
671
+ module.decoder.layers.15.self_attention.linear_qkv.weight
672
+ module.decoder.layers.15.pre_mlp_layernorm.weight
673
+ module.decoder.layers.15.mlp.router.weight
674
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
675
+ module.decoder.layers.15.self_attention.linear_proj.weight
676
+ Params for bucket 3 (42213376 elements, 42213376 padded size):
677
+ module.decoder.layers.14.pre_mlp_layernorm.weight
678
+ module.decoder.layers.14.self_attention.linear_qkv.weight
679
+ module.decoder.layers.14.mlp.router.weight
680
+ module.decoder.layers.14.self_attention.linear_proj.weight
681
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
682
+ Params for bucket 4 (42213376 elements, 42213376 padded size):
683
+ module.decoder.layers.13.self_attention.linear_qkv.weight
684
+ module.decoder.layers.13.pre_mlp_layernorm.weight
685
+ module.decoder.layers.13.mlp.router.weight
686
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
687
+ module.decoder.layers.13.self_attention.linear_proj.weight
688
+ Params for bucket 5 (42213376 elements, 42213376 padded size):
689
+ module.decoder.layers.12.self_attention.linear_qkv.weight
690
+ module.decoder.layers.12.pre_mlp_layernorm.weight
691
+ module.decoder.layers.12.mlp.router.weight
692
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
693
+ module.decoder.layers.12.self_attention.linear_proj.weight
694
+ Params for bucket 6 (42213376 elements, 42213376 padded size):
695
+ module.decoder.layers.11.self_attention.linear_qkv.weight
696
+ module.decoder.layers.11.pre_mlp_layernorm.weight
697
+ module.decoder.layers.11.mlp.router.weight
698
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
699
+ module.decoder.layers.11.self_attention.linear_proj.weight
700
+ Params for bucket 7 (42213376 elements, 42213376 padded size):
701
+ module.decoder.layers.10.pre_mlp_layernorm.weight
702
+ module.decoder.layers.10.self_attention.linear_qkv.weight
703
+ module.decoder.layers.10.mlp.router.weight
704
+ module.decoder.layers.10.self_attention.linear_proj.weight
705
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
706
+ Params for bucket 8 (42213376 elements, 42213376 padded size):
707
+ module.decoder.layers.9.pre_mlp_layernorm.weight
708
+ module.decoder.layers.9.self_attention.linear_qkv.weight
709
+ module.decoder.layers.9.mlp.router.weight
710
+ module.decoder.layers.9.self_attention.linear_proj.weight
711
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
712
+ Params for bucket 9 (42213376 elements, 42213376 padded size):
713
+ module.decoder.layers.8.pre_mlp_layernorm.weight
714
+ module.decoder.layers.8.self_attention.linear_qkv.weight
715
+ module.decoder.layers.8.mlp.router.weight
716
+ module.decoder.layers.8.self_attention.linear_proj.weight
717
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
718
+ Params for bucket 10 (42213376 elements, 42213376 padded size):
719
+ module.decoder.layers.7.mlp.router.weight
720
+ module.decoder.layers.7.self_attention.linear_proj.weight
721
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
722
+ module.decoder.layers.7.self_attention.linear_qkv.weight
723
+ module.decoder.layers.7.pre_mlp_layernorm.weight
724
+ Params for bucket 11 (42213376 elements, 42213376 padded size):
725
+ module.decoder.layers.6.pre_mlp_layernorm.weight
726
+ module.decoder.layers.6.self_attention.linear_qkv.weight
727
+ module.decoder.layers.6.mlp.router.weight
728
+ module.decoder.layers.6.self_attention.linear_proj.weight
729
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
730
+ Params for bucket 12 (42213376 elements, 42213376 padded size):
731
+ module.decoder.layers.5.pre_mlp_layernorm.weight
732
+ module.decoder.layers.5.self_attention.linear_qkv.weight
733
+ module.decoder.layers.5.self_attention.linear_proj.weight
734
+ module.decoder.layers.5.mlp.router.weight
735
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
736
+ Params for bucket 13 (42213376 elements, 42213376 padded size):
737
+ module.decoder.layers.4.self_attention.linear_qkv.weight
738
+ module.decoder.layers.4.pre_mlp_layernorm.weight
739
+ module.decoder.layers.4.mlp.router.weight
740
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
741
+ module.decoder.layers.4.self_attention.linear_proj.weight
742
+ Params for bucket 14 (42213376 elements, 42213376 padded size):
743
+ module.decoder.layers.3.mlp.router.weight
744
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
745
+ module.decoder.layers.3.self_attention.linear_proj.weight
746
+ module.decoder.layers.3.self_attention.linear_qkv.weight
747
+ module.decoder.layers.3.pre_mlp_layernorm.weight
748
+ Params for bucket 15 (42213376 elements, 42213376 padded size):
749
+ module.decoder.layers.2.mlp.router.weight
750
+ module.decoder.layers.2.self_attention.linear_proj.weight
751
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
752
+ module.decoder.layers.2.self_attention.linear_qkv.weight
753
+ module.decoder.layers.2.pre_mlp_layernorm.weight
754
+ Params for bucket 16 (42213376 elements, 42213376 padded size):
755
+ module.decoder.layers.1.mlp.router.weight
756
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
757
+ module.decoder.layers.1.self_attention.linear_proj.weight
758
+ module.decoder.layers.1.pre_mlp_layernorm.weight
759
+ module.decoder.layers.1.self_attention.linear_qkv.weight
760
+ Params for bucket 17 (42213376 elements, 42213376 padded size):
761
+ module.decoder.layers.0.mlp.router.weight
762
+ module.decoder.layers.0.self_attention.linear_proj.weight
763
+ module.decoder.layers.0.self_attention.linear_qkv.weight
764
+ module.decoder.layers.0.pre_mlp_layernorm.weight
765
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 64
769
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight7
771
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
772
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight5
773
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight4
774
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight6
775
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
776
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
777
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
778
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
779
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight6
780
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight7
781
+ Params for bucket 3 (44040192 elements, 44040192 padded size):
782
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight5
783
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight4
784
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
785
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
786
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
787
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
788
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
790
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight6
791
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight4
792
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
793
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight7
794
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight5
795
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
796
+ Params for bucket 6 (44040192 elements, 44040192 padded size):
797
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
798
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight7
799
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
800
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight6
801
+ Params for bucket 7 (44040192 elements, 44040192 padded size):
802
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight5
803
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight4
804
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
805
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
806
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
807
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
808
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
809
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
810
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight6
811
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight4
812
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
813
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight7
814
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
815
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight5
816
+ Params for bucket 10 (44040192 elements, 44040192 padded size):
817
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
818
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight7
819
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
820
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight6
821
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
822
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight5
823
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight4
824
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
825
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
827
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
828
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
829
+ Params for bucket 13 (44040192 elements, 44040192 padded size):
830
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight4
831
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight6
832
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
833
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight7
834
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight5
835
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
836
+ Params for bucket 14 (44040192 elements, 44040192 padded size):
837
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
838
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
839
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight6
840
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight7
841
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
842
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight5
843
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
844
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight4
845
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
846
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
847
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
848
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
849
+ Params for bucket 17 (44040192 elements, 44040192 padded size):
850
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight4
851
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight6
852
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
853
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight7
854
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
855
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight5
856
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
857
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
858
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight6
859
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
860
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight7
861
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
862
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight5
863
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
864
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight4
865
+ Params for bucket 20 (44040192 elements, 44040192 padded size):
866
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
867
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
868
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
869
+ Params for bucket 21 (44040192 elements, 44040192 padded size):
870
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight4
871
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight6
872
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
873
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight7
874
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight5
875
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
876
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
877
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
878
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight6
879
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight7
880
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
881
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
882
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight5
883
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
884
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight4
885
+ Params for bucket 24 (44040192 elements, 44040192 padded size):
886
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
887
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
888
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
889
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
890
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight6
891
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight4
892
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
893
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight7
894
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
895
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight5
896
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
897
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
898
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight7
899
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
900
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight6
901
+ Params for bucket 27 (44040192 elements, 44040192 padded size):
902
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight5
903
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight4
904
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
905
+ Params for bucket 28 (44040192 elements, 44040192 padded size):
906
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
907
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
908
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
909
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
910
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight6
911
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight4
912
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
913
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight7
914
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
915
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight5
916
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
917
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
918
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
919
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight6
920
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight7
921
+ Params for bucket 31 (44040192 elements, 44040192 padded size):
922
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight5
923
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
924
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight4
925
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
926
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
927
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
928
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
929
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
930
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight6
931
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight4
932
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
933
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight7
934
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight5
935
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
936
+ Params for bucket 34 (44040192 elements, 44040192 padded size):
937
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
938
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
939
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight6
940
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight7
941
+ Params for bucket 35 (44040192 elements, 44040192 padded size):
942
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight5
943
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
944
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight4
945
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
946
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
947
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
948
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
949
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
950
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight6
951
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight4
952
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
953
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight7
954
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight5
955
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
956
+ Params for bucket 38 (44040192 elements, 44040192 padded size):
957
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
958
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight7
959
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
960
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight6
961
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
962
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight5
963
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight4
964
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
965
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
966
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
967
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
968
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
969
+ Params for bucket 41 (44040192 elements, 44040192 padded size):
970
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight6
971
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight4
972
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
973
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight7
974
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
975
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight5
976
+ Params for bucket 42 (44040192 elements, 44040192 padded size):
977
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
978
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
979
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight7
980
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight6
981
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
982
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight5
983
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight4
984
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
985
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
986
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
987
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
988
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
989
+ Params for bucket 45 (44040192 elements, 44040192 padded size):
990
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight4
991
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight6
992
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
993
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight7
994
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
995
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight5
996
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
997
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
998
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
999
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight6
1000
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight7
1001
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
1002
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight5
1003
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
1004
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight4
1005
+ Params for bucket 48 (44040192 elements, 44040192 padded size):
1006
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
1007
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
1008
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
1009
+ Params for bucket 49 (44040192 elements, 44040192 padded size):
1010
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight7
1011
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
1012
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight5
1013
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight4
1014
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight6
1015
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
1016
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
1017
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
1018
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
1019
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight6
1020
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight7
1021
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
1022
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight5
1023
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
1024
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight4
1025
+ Params for bucket 52 (44040192 elements, 44040192 padded size):
1026
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
1027
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
1028
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
1029
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
1030
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight7
1031
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight5
1032
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
1033
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight4
1034
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight6
1035
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
1036
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
1037
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
1038
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight6
1039
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
1040
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight7
1041
+ Params for bucket 55 (44040192 elements, 44040192 padded size):
1042
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight5
1043
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
1044
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight4
1045
+ Params for bucket 56 (44040192 elements, 44040192 padded size):
1046
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
1047
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
1048
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
1049
+ Params for bucket 57 (44040192 elements, 44040192 padded size):
1050
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight7
1051
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight5
1052
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
1053
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight6
1054
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight4
1055
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
1056
+ Params for bucket 58 (44040192 elements, 44040192 padded size):
1057
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
1058
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight7
1059
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight6
1060
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
1061
+ Params for bucket 59 (44040192 elements, 44040192 padded size):
1062
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight5
1063
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight4
1064
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
1065
+ Params for bucket 60 (44040192 elements, 44040192 padded size):
1066
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
1067
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
1068
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
1069
+ Params for bucket 61 (44040192 elements, 44040192 padded size):
1070
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight7
1071
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight5
1072
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
1073
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight6
1074
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
1075
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight4
1076
+ Params for bucket 62 (44040192 elements, 44040192 padded size):
1077
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
1078
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight6
1079
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight7
1080
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
1081
+ Params for bucket 63 (44040192 elements, 44040192 padded size):
1082
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight5
1083
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight4
1084
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
1085
+ Params for bucket 64 (44040192 elements, 44040192 padded size):
1086
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
1087
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
1088
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
1089
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x76900b124ec0>, config_logger_dir='')
1090
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
1091
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:08:31
1092
+ > building train, validation, and test datasets ...
1093
+ > datasets target sizes (minimum size):
1094
+ train: 192
1095
+ validation: 0
1096
+ test: 0
1097
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
1098
+ > building train, validation, and test datasets for GPT ...
1099
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x76900b2210d0>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
1100
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
1101
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
1102
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
1103
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
1104
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
1105
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
1106
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
1107
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
1108
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
1109
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
1110
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
1111
+ > finished creating GPT datasets ...
1112
+ [after dataloaders are built] datetime: 2026-08-30 07:08:31
1113
+ done with setup ...
1114
+ (min, max) time across ranks (ms):
1115
+ model-and-optimizer-setup ......................: (3661.43, 3670.90)
1116
+ train/valid/test-data-iterators-setup ..........: (73.55, 91.46)
1117
+ training ...
1118
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
1119
+ [before the start of training step] datetime: 2026-08-30 07:08:31
1120
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1121
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1122
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1123
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1124
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1125
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1126
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1127
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1128
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1129
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1130
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1131
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1132
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1133
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1134
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1135
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1136
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1137
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1138
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1139
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1140
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1141
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1142
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1143
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1144
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1145
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1146
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1147
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1148
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1149
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1150
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1151
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1152
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1153
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1154
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1155
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1156
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1157
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1158
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1159
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1160
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1161
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1162
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1163
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1164
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1165
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1166
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1167
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1168
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1169
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1170
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1171
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1172
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1173
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1174
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1175
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1176
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1177
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1178
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1179
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1180
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1181
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1182
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1183
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1184
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1185
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1186
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1187
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1188
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1189
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1190
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1191
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1192
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1193
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1194
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1195
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1196
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1197
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1198
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1199
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1200
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1201
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1202
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1203
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1204
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1205
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1206
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1207
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1208
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1209
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1210
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1211
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1212
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1213
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1214
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1215
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1216
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1217
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1218
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1219
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1220
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1221
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1222
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1223
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1224
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1225
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1226
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1227
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1228
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1229
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1230
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1231
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1232
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1233
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1234
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1235
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1236
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1237
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1238
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1239
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1240
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1241
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1242
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1243
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1244
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1245
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1246
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1247
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1248
+ [2026-08-30 07:08:57] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26549.4 | throughput per GPU (TFLOP/s/GPU): 12.6 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103531E+01 | loss scale: 1.0 | grad norm: 44.064 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |Number of parameters in transformer block in billions: 23.22
1249
+
1250
+ Number of parameters in embedding layers in billions: 0.41
1251
+ Total number of parameters in billions: 23.63
1252
+ Number of parameters in most loaded shard in billions: 23.6320
1253
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1254
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.49609375 | max allocated: 67206.4716796875 | reserved: 68904.0 | max reserved: 68904.0
1255
+ [2026-08-30 07:08:58] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 507.3 | throughput per GPU (TFLOP/s/GPU): 657.1 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.103001E+01 | loss scale: 1.0 | grad norm: 45.975 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1256
+ [2026-08-30 07:08:58] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 463.0 | throughput per GPU (TFLOP/s/GPU): 720.1 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.043814E+01 | loss scale: 1.0 | grad norm: 205.503 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1257
+ Capture CUDA graph for training!!!
1258
+ CUDA graph capture done!!!
1259
+ [2026-08-30 07:09:00] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2120.5 | throughput per GPU (TFLOP/s/GPU): 157.2 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.319265E+01 | loss scale: 1.0 | grad norm: 35.581 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1260
+ [2026-08-30 07:09:01] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 433.6 | throughput per GPU (TFLOP/s/GPU): 768.9 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.039609E+01 | loss scale: 1.0 | grad norm: 15.618 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1261
+ [2026-08-30 07:09:01] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 452.3 | throughput per GPU (TFLOP/s/GPU): 737.1 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.046226E+01 | loss scale: 1.0 | grad norm: 39.295 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1262
+ [2026-08-30 07:09:02] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 449.0 | throughput per GPU (TFLOP/s/GPU): 742.6 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.002612E+01 | loss scale: 1.0 | grad norm: 8.387 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1263
+ [2026-08-30 07:09:02] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 453.2 | throughput per GPU (TFLOP/s/GPU): 735.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.541579E+00 | loss scale: 1.0 | grad norm: 5.811 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1264
+ [2026-08-30 07:09:03] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 438.5 | throughput per GPU (TFLOP/s/GPU): 760.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.312139E+00 | loss scale: 1.0 | grad norm: 5.649 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1265
+ [2026-08-30 07:09:03] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 456.8 | throughput per GPU (TFLOP/s/GPU): 729.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.155394E+00 | loss scale: 1.0 | grad norm: 5.472 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1266
+ [2026-08-30 07:09:03] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 443.9 | throughput per GPU (TFLOP/s/GPU): 751.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.034643E+00 | loss scale: 1.0 | grad norm: 5.276 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1267
+ [2026-08-30 07:09:04] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 442.8 | throughput per GPU (TFLOP/s/GPU): 752.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.911410E+00 | loss scale: 1.0 | grad norm: 5.354 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1268
+ [after training is done] datetime: 2026-08-30 07:09:04
1269
+ [rank0]:[W830 07:09:05.865172068 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1270
+ [rank3]:[W830 07:09:06.170413075 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1271
+ [rank7]:[W830 07:09:06.175257610 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1272
+ [rank4]:[W830 07:09:06.175542703 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1273
+ [rank1]:[W830 07:09:06.175580971 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1274
+ [rank5]:[W830 07:09:06.176744581 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1275
+ [rank2]:[W830 07:09:07.234745359 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1276
+ [rank6]:[W830 07:09:07.254645769 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1277
+ W0830 07:09:12.052000 711599 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1278
+ W0830 07:09:12.053000 711599 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 711687 closing signal SIGTERM
1279
+ W0830 07:09:12.054000 711599 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 711688 closing signal SIGTERM
1280
+ W0830 07:09:12.054000 711599 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 711691 closing signal SIGTERM
1281
+ Traceback (most recent call last):
1282
+ File "/usr/local/bin/torchrun", line 8, in <module>
1283
+ sys.exit(main())
1284
+ ^^^^^^
1285
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1286
+ return f(*args, **kwargs)
1287
+ ^^^^^^^^^^^^^^^^^^
1288
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1289
+ run(args)
1290
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1291
+ elastic_launch(
1292
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1293
+ return launch_agent(self._config, self._entrypoint, list(args))
1294
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1295
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1296
+ result = agent.run()
1297
+ ^^^^^^^^^^^
1298
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1299
+ result = f(*args, **kwargs)
1300
+ ^^^^^^^^^^^^^^^^^^
1301
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1302
+ result = self._invoke_run(role)
1303
+ ^^^^^^^^^^^^^^^^^^^^^^
1304
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1305
+ time.sleep(monitor_interval)
1306
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1307
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1308
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 711599 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv ADDED
@@ -0,0 +1,224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 4
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 4
42
+ 1, 629
43
+ 2, 629
44
+ 3, 629
45
+ 4, 629
46
+ 5, 629
47
+ 6, 629
48
+ 7, 629
49
+ 0, 881
50
+ 1, 889
51
+ 2, 893
52
+ 3, 887
53
+ 4, 895
54
+ 5, 921
55
+ 6, 905
56
+ 7, 921
57
+ 0, 1895
58
+ 1, 1895
59
+ 2, 1895
60
+ 3, 1895
61
+ 4, 1895
62
+ 5, 1895
63
+ 6, 1895
64
+ 7, 1895
65
+ 0, 2357
66
+ 1, 2349
67
+ 2, 2353
68
+ 3, 2355
69
+ 4, 2369
70
+ 5, 2361
71
+ 6, 2387
72
+ 7, 2379
73
+ 0, 38759
74
+ 1, 38527
75
+ 2, 39031
76
+ 3, 38919
77
+ 4, 38863
78
+ 5, 38919
79
+ 6, 38695
80
+ 7, 39087
81
+ 0, 40937
82
+ 1, 40925
83
+ 2, 40925
84
+ 3, 40925
85
+ 4, 40925
86
+ 5, 40925
87
+ 6, 40925
88
+ 7, 40925
89
+ 0, 41977
90
+ 1, 41961
91
+ 2, 41961
92
+ 3, 41961
93
+ 4, 41961
94
+ 5, 41961
95
+ 6, 41961
96
+ 7, 41961
97
+ 0, 42251
98
+ 1, 42235
99
+ 2, 42235
100
+ 3, 42235
101
+ 4, 42235
102
+ 5, 42235
103
+ 6, 42235
104
+ 7, 42235
105
+ 0, 42801
106
+ 1, 42785
107
+ 2, 42785
108
+ 3, 42785
109
+ 4, 42785
110
+ 5, 42785
111
+ 6, 42785
112
+ 7, 42785
113
+ 0, 44637
114
+ 1, 44621
115
+ 2, 44621
116
+ 3, 44621
117
+ 4, 44621
118
+ 5, 44621
119
+ 6, 44621
120
+ 7, 44621
121
+ 0, 66717
122
+ 1, 66735
123
+ 2, 66735
124
+ 3, 66735
125
+ 4, 66239
126
+ 5, 66735
127
+ 6, 66735
128
+ 7, 66735
129
+ 0, 67631
130
+ 1, 67649
131
+ 2, 67649
132
+ 3, 67649
133
+ 4, 67703
134
+ 5, 67649
135
+ 6, 67649
136
+ 7, 67649
137
+ 0, 67631
138
+ 1, 67649
139
+ 2, 67649
140
+ 3, 67649
141
+ 4, 67703
142
+ 5, 67649
143
+ 6, 67649
144
+ 7, 67649
145
+ 0, 74731
146
+ 1, 74735
147
+ 2, 74735
148
+ 3, 74735
149
+ 4, 74789
150
+ 5, 74735
151
+ 6, 74735
152
+ 7, 74735
153
+ 0, 75671
154
+ 1, 75721
155
+ 2, 75709
156
+ 3, 75741
157
+ 4, 75823
158
+ 5, 75769
159
+ 6, 75769
160
+ 7, 75769
161
+ 0, 76651
162
+ 1, 76639
163
+ 2, 76669
164
+ 3, 76633
165
+ 4, 76725
166
+ 5, 76651
167
+ 6, 76697
168
+ 7, 76667
169
+ 0, 99955
170
+ 1, 100037
171
+ 2, 100037
172
+ 3, 100037
173
+ 4, 100005
174
+ 5, 100037
175
+ 6, 100037
176
+ 7, 100037
177
+ 0, 101939
178
+ 1, 101827
179
+ 2, 101827
180
+ 3, 101827
181
+ 4, 102043
182
+ 5, 101827
183
+ 6, 101827
184
+ 7, 101827
185
+ 0, 101939
186
+ 1, 101827
187
+ 2, 101827
188
+ 3, 101827
189
+ 4, 102043
190
+ 5, 101827
191
+ 6, 101827
192
+ 7, 101827
193
+ 0, 101939
194
+ 1, 101827
195
+ 2, 101827
196
+ 3, 101827
197
+ 4, 102043
198
+ 5, 101827
199
+ 6, 101827
200
+ 7, 101827
201
+ 0, 99481
202
+ 1, 99369
203
+ 2, 99369
204
+ 3, 99369
205
+ 4, 99585
206
+ 5, 99369
207
+ 6, 99369
208
+ 7, 99369
209
+ 0, 97413
210
+ 1, 97301
211
+ 2, 97689
212
+ 3, 97689
213
+ 4, 97905
214
+ 5, 97301
215
+ 6, 97301
216
+ 7, 97689
217
+ 0, 4
218
+ 1, 4
219
+ 2, 95869
220
+ 3, 4
221
+ 4, 96167
222
+ 5, 4
223
+ 6, 4
224
+ 7, 4
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G3
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=1792
11
+ num_experts=64
12
+ topk=16
13
+ experts_per_rank=8
14
+ expert_params=22020096
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=65536
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.072265625
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2/train.log ADDED
@@ -0,0 +1,1310 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G3 / ultraep (slots/rank=1 policy=one)
3
+ E=64 K=16 H'=1792 m=8
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.072265625 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:09:15.552000 714645 torch/distributed/run.py:766]
10
+ W0830 07:09:15.552000 714645 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:09:15.552000 714645 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:09:15.552000 714645 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
18
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
19
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
20
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
21
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
22
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
23
+ Number of virtual stages per pipeline stage: None
24
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
25
+ using torch.bfloat16 for parameters ...
26
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
27
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
28
+ ------------------------ arguments ------------------------
29
+ account_for_embedding_in_pipeline_split ......... False
30
+ account_for_loss_in_pipeline_split .............. False
31
+ accumulate_allreduce_grads_in_fp32 .............. True
32
+ adam_beta1 ...................................... 0.9
33
+ adam_beta2 ...................................... 0.999
34
+ adam_eps ........................................ 1e-08
35
+ add_bias_linear ................................. False
36
+ add_position_embedding .......................... False
37
+ add_qkv_bias .................................... False
38
+ adlr_autoresume ................................. False
39
+ adlr_autoresume_interval ........................ 1000
40
+ align_grad_reduce ............................... True
41
+ align_param_gather .............................. False
42
+ app_tag_run_name ................................ None
43
+ app_tag_run_version ............................. 0.0.0
44
+ apply_layernorm_1p .............................. False
45
+ apply_query_key_layer_scaling ................... False
46
+ apply_residual_connection_post_layernorm ........ False
47
+ apply_rope_fusion ............................... True
48
+ async_save ...................................... None
49
+ async_tensor_model_parallel_allreduce ........... True
50
+ attention_backend ............................... AttnBackend.auto
51
+ attention_dropout ............................... 0.0
52
+ attention_softmax_in_fp32 ....................... False
53
+ auto_detect_ckpt_format ......................... False
54
+ barrier_with_L1_time ............................ True
55
+ bert_binary_head ................................ True
56
+ bert_embedder_type .............................. megatron
57
+ bert_load ....................................... None
58
+ bf16 ............................................ True
59
+ bias_dropout_fusion ............................. True
60
+ bias_gelu_fusion ................................ False
61
+ bias_swiglu_fusion .............................. True
62
+ biencoder_projection_dim ........................ 0
63
+ biencoder_shared_query_context_model ............ False
64
+ block_data_path ................................. None
65
+ calc_ft_timeouts ................................ False
66
+ calculate_per_token_loss ........................ False
67
+ check_for_large_grads ........................... False
68
+ check_for_nan_in_loss_and_grad .................. False
69
+ check_for_spiky_loss ............................ False
70
+ check_weight_hash_across_dp_replicas_interval ... None
71
+ ckpt_assume_constant_structure .................. False
72
+ ckpt_convert_format ............................. None
73
+ ckpt_convert_save ............................... None
74
+ ckpt_convert_update_legacy_dist_opt_format ...... False
75
+ ckpt_format ..................................... torch_dist
76
+ ckpt_fully_parallel_load ........................ False
77
+ ckpt_fully_parallel_save ........................ True
78
+ ckpt_fully_parallel_save_deprecated ............. False
79
+ ckpt_step ....................................... None
80
+ classes_fraction ................................ 1.0
81
+ clip_grad ....................................... 1.0
82
+ clone_scatter_output_in_embedding ............... True
83
+ config_logger_dir ...............................
84
+ consumed_train_samples .......................... 0
85
+ consumed_valid_samples .......................... 0
86
+ context_parallel_size ........................... 1
87
+ cp_comm_type .................................... ['p2p']
88
+ create_attention_mask_in_dataloader ............. True
89
+ cross_entropy_fusion_impl ....................... native
90
+ cross_entropy_loss_fusion ....................... False
91
+ cuda_graph_scope ................................ full_iteration
92
+ cuda_graph_warmup_steps ......................... 3
93
+ data_args_path .................................. None
94
+ data_cache_path ................................. None
95
+ data_parallel_random_init ....................... False
96
+ data_parallel_sharding_strategy ................. no_shard
97
+ data_parallel_size .............................. 8
98
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
99
+ data_per_class_fraction ......................... 1.0
100
+ data_sharding ................................... True
101
+ dataloader_type ................................. single
102
+ ddp_average_in_collective ....................... False
103
+ ddp_bucket_size ................................. None
104
+ ddp_num_buckets ................................. None
105
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
106
+ decoder_first_pipeline_num_layers ............... None
107
+ decoder_last_pipeline_num_layers ................ None
108
+ decoder_num_layers .............................. None
109
+ decoder_seq_length .............................. None
110
+ decoupled_lr .................................... None
111
+ decoupled_min_lr ................................ None
112
+ decrease_batch_size_if_needed ................... False
113
+ defer_embedding_wgrad_compute ................... False
114
+ delay_wgrad_compute ............................. False
115
+ deprecated_use_mcore_models ..................... True
116
+ deterministic_mode .............................. False
117
+ dino_bottleneck_size ............................ 256
118
+ dino_freeze_last_layer .......................... 1
119
+ dino_head_hidden_size ........................... 2048
120
+ dino_local_crops_number ......................... 10
121
+ dino_local_img_size ............................. 96
122
+ dino_norm_last_layer ............................ False
123
+ dino_teacher_temp ............................... 0.07
124
+ dino_warmup_teacher_temp ........................ 0.04
125
+ dino_warmup_teacher_temp_epochs ................. 30
126
+ disable_bf16_reduced_precision_matmul ........... False
127
+ disable_mamba_mem_eff_path ...................... False
128
+ disable_straggler_on_startup .................... False
129
+ dist_ckpt_format_deprecated ..................... None
130
+ dist_ckpt_strictness ............................ assume_ok_unexpected
131
+ distribute_saved_activations .................... False
132
+ distributed_backend ............................. nccl
133
+ distributed_timeout_minutes ..................... 10
134
+ embedding_init_method_std ....................... None
135
+ embedding_path .................................. None
136
+ empty_unused_memory_level ....................... 0
137
+ enable_cuda_graph ............................... True
138
+ enable_experimental ............................. False
139
+ enable_ft_package ............................... False
140
+ enable_gloo_process_groups ...................... True
141
+ enable_msc ...................................... True
142
+ enable_one_logger ............................... True
143
+ encoder_num_layers .............................. 16
144
+ encoder_seq_length .............................. 4096
145
+ end_weight_decay ................................ 0.1
146
+ eod_mask_loss ................................... False
147
+ error_injection_rate ............................ 0
148
+ error_injection_type ............................ transient_error
149
+ eval_interval ................................... 1000000
150
+ eval_iters ...................................... 0
151
+ evidence_data_path .............................. None
152
+ exit_duration_in_mins ........................... None
153
+ exit_interval ................................... None
154
+ exit_on_missing_checkpoint ...................... False
155
+ exit_signal_handler ............................. False
156
+ exp_avg_dtype ................................... torch.float32
157
+ exp_avg_sq_dtype ................................ torch.float32
158
+ expert_model_parallel_size ...................... 8
159
+ expert_tensor_parallel_size ..................... 1
160
+ export_force_local_attention .................... False
161
+ export_kd_cfg ................................... None
162
+ export_kd_teacher_ckpt_format ................... None
163
+ export_kd_teacher_load .......................... None
164
+ export_kv_cache_quant ........................... False
165
+ export_legacy_megatron .......................... False
166
+ export_model_type ............................... GPTModel
167
+ export_moe_apply_probs_on_input ................. False
168
+ export_qk_l2_norm ............................... False
169
+ export_quant_cfg ................................ None
170
+ export_real_quant_cfg ........................... None
171
+ export_te_mcore_model ........................... False
172
+ external_cuda_graph ............................. False
173
+ ffn_hidden_size ................................. 14336
174
+ finetune ........................................ False
175
+ finetune_data_split ............................. train
176
+ finetune_hf_dataset ............................. None
177
+ first_last_layers_bf16 .......................... False
178
+ flash_decode .................................... False
179
+ fp16 ............................................ False
180
+ fp16_lm_cross_entropy ........................... False
181
+ fp32_residual_connection ........................ False
182
+ fp8 ............................................. e4m3
183
+ fp8_amax_compute_algo ........................... most_recent
184
+ fp8_amax_history_len ............................ 1
185
+ fp8_interval .................................... 1
186
+ fp8_margin ...................................... 0
187
+ fp8_param_gather ................................ False
188
+ fp8_recipe ...................................... mxfp8
189
+ fp8_wgrad ....................................... True
190
+ fsdp_double_buffer .............................. False
191
+ global_batch_size ............................... 16
192
+ grad_reduce_in_bf16 ............................. False
193
+ gradient_accumulation_fusion .................... True
194
+ gradient_reduce_div_fusion ...................... True
195
+ group_query_attention ........................... True
196
+ head_lr_mult .................................... 1.0
197
+ heterogeneous_layers_config_encoded_json ........ None
198
+ heterogeneous_layers_config_path ................ None
199
+ hidden_dropout .................................. 0.0
200
+ hidden_size ..................................... 4096
201
+ hierarchical_context_parallel_sizes ............. None
202
+ high_priority_stream_groups ..................... []
203
+ hybrid_attention_ratio .......................... 0.0
204
+ hybrid_mlp_ratio ................................ 0.0
205
+ hybrid_override_pattern ......................... None
206
+ hysteresis ...................................... 2
207
+ ict_head_size ................................... None
208
+ ict_load ........................................ None
209
+ img_h ........................................... 224
210
+ img_w ........................................... 224
211
+ indexer_batch_size .............................. 128
212
+ indexer_log_interval ............................ 1000
213
+ inference_batch_times_seqlen_threshold .......... -1
214
+ inference_dynamic_batching ...................... False
215
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
216
+ inference_dynamic_batching_buffer_overflow_factor None
217
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
218
+ inference_dynamic_batching_chunk_size ........... 256
219
+ inference_dynamic_batching_max_requests_override None
220
+ inference_dynamic_batching_max_tokens_override .. None
221
+ inference_dynamic_batching_num_cuda_graphs ...... 16
222
+ inference_max_batch_size ........................ 8
223
+ inference_max_seq_length ........................ 2560
224
+ inference_rng_tracker ........................... False
225
+ init_method_std ................................. 0.01
226
+ init_method_xavier_uniform ...................... False
227
+ init_model_with_meta_device ..................... False
228
+ initial_loss_scale .............................. 4294967296
229
+ inprocess_active_world_size ..................... 8
230
+ inprocess_barrier_timeout ....................... 120
231
+ inprocess_completion_timeout .................... 120
232
+ inprocess_empty_cuda_cache ...................... False
233
+ inprocess_granularity ........................... node
234
+ inprocess_hard_timeout .......................... 90
235
+ inprocess_heartbeat_interval .................... 30
236
+ inprocess_heartbeat_timeout ..................... 60
237
+ inprocess_last_call_wait ........................ 1
238
+ inprocess_max_iterations ........................ None
239
+ inprocess_monitor_process_interval .............. 1.0
240
+ inprocess_monitor_thread_interval ............... 1.0
241
+ inprocess_progress_watchdog_interval ............ 1.0
242
+ inprocess_restart ............................... False
243
+ inprocess_soft_timeout .......................... 60
244
+ inprocess_termination_grace_time ................ 1
245
+ is_hybrid_model ................................. False
246
+ iter_per_epoch .................................. 1250
247
+ iterations_to_skip .............................. []
248
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
249
+ kitchen_config_file ............................. None
250
+ kitchen_recipe_number ........................... None
251
+ kv_channels ..................................... 128
252
+ kv_lora_rank .................................... 32
253
+ lazy_mpu_init ................................... None
254
+ load ............................................ None
255
+ load_main_params_from_ckpt ...................... None
256
+ load_model_opt_format ........................... False
257
+ local_rank ...................................... 0
258
+ log_energy ...................................... False
259
+ log_interval .................................... 1
260
+ log_loss_scale_to_tensorboard ................... True
261
+ log_memory_to_tensorboard ....................... False
262
+ log_num_zeros_in_grad ........................... False
263
+ log_params_norm ................................. False
264
+ log_progress .................................... False
265
+ log_straggler ................................... False
266
+ log_throughput .................................. True
267
+ log_timers_to_tensorboard ....................... False
268
+ log_validation_ppl_to_tensorboard ............... False
269
+ log_world_size_to_tensorboard ................... False
270
+ logging_level ................................... None
271
+ loss_scale ...................................... None
272
+ loss_scale_window ............................... 1000
273
+ lr .............................................. 0.0001
274
+ lr_decay_iters .................................. 8
275
+ lr_decay_samples ................................ None
276
+ lr_decay_style .................................. cosine
277
+ lr_warmup_fraction .............................. None
278
+ lr_warmup_init .................................. 0.0
279
+ lr_warmup_iters ................................. 5
280
+ lr_warmup_samples ............................... 0
281
+ lr_wsd_decay_iters .............................. None
282
+ lr_wsd_decay_samples ............................ None
283
+ lr_wsd_decay_style .............................. exponential
284
+ main_grads_dtype ................................ torch.float32
285
+ main_params_dtype ............................... torch.float32
286
+ make_vocab_size_divisible_by .................... 128
287
+ mamba_head_dim .................................. 64
288
+ mamba_num_groups ................................ 8
289
+ mamba_num_heads ................................. None
290
+ mamba_state_dim ................................. 128
291
+ manual_gc ....................................... False
292
+ manual_gc_eval .................................. True
293
+ manual_gc_interval .............................. 0
294
+ mask_factor ..................................... 1.0
295
+ mask_prob ....................................... 0.15
296
+ mask_type ....................................... random
297
+ masked_softmax_fusion ........................... False
298
+ max_position_embeddings ......................... 32768
299
+ max_tokens_to_oom ............................... 12000
300
+ memory_snapshot_path ............................ snapshot.pickle
301
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
302
+ micro_batch_size ................................ 1
303
+ microbatch_group_size_per_vp_stage .............. None
304
+ mid_level_dataset_surplus ....................... 0.005
305
+ min_loss_scale .................................. 1.0
306
+ min_lr .......................................... 1e-05
307
+ mlp_chunks_for_prefill .......................... 1
308
+ mmap_bin_files .................................. True
309
+ mock_data ....................................... False
310
+ moe_apply_probs_on_input ........................ False
311
+ moe_aux_loss_coeff .............................. 0.01
312
+ moe_capacity_monitor ............................ True
313
+ moe_deepep_num_sms .............................. 20
314
+ moe_echo_enable_random_offloading ............... False
315
+ moe_echo_expert_dispatch_overlap ................ False
316
+ moe_echo_expert_dispatcher_type ................. hybridep
317
+ moe_echo_recompute_expert_dispatch .............. False
318
+ moe_enable_deepep ............................... False
319
+ moe_enable_echo ................................. False
320
+ moe_enable_hybridep ............................. True
321
+ moe_enable_ultraep .............................. True
322
+ moe_expert_capacity_factor ...................... None
323
+ moe_extended_tp ................................. False
324
+ moe_ffn_hidden_size ............................. 1792
325
+ moe_grouped_gemm ................................ True
326
+ moe_hybridep_num_sms ............................ 16
327
+ moe_input_jitter_eps ............................ None
328
+ moe_layer_freq .................................. 1
329
+ moe_layer_recompute ............................. False
330
+ moe_num_echo_experts ............................ None
331
+ moe_num_redundant_experts_per_rank .............. 1
332
+ moe_pad_expert_input_to_capacity ................ False
333
+ moe_per_layer_logging ........................... False
334
+ moe_permute_fusion .............................. False
335
+ moe_received_token_capacity ..................... 1.072265625
336
+ moe_router_bias_update_rate ..................... 0.001
337
+ moe_router_dtype ................................ fp32
338
+ moe_router_enable_expert_bias ................... False
339
+ moe_router_force_load_balancing ................. False
340
+ moe_router_group_topk ........................... None
341
+ moe_router_load_balancing_type .................. aux_loss
342
+ moe_router_num_groups ........................... None
343
+ moe_router_padding_for_fp8 ...................... False
344
+ moe_router_pre_softmax .......................... False
345
+ moe_router_score_function ....................... softmax
346
+ moe_router_topk ................................. 16
347
+ moe_router_topk_scaling_factor .................. None
348
+ moe_shared_expert_intermediate_size ............. None
349
+ moe_shared_expert_overlap ....................... False
350
+ moe_token_dispatcher_type ....................... flex
351
+ moe_token_drop_policy ........................... probs
352
+ moe_ultraep_max_microbatches .................... None
353
+ moe_ultraep_overlap_weight_sync ................. True
354
+ moe_ultraep_weight_sync_join .................... before_dispatch
355
+ moe_upcycling_granularity ....................... 1
356
+ moe_use_device_initiated_grouped_gemm ........... True
357
+ moe_use_legacy_grouped_gemm ..................... False
358
+ moe_use_upcycling ............................... False
359
+ moe_z_loss_coeff ................................ None
360
+ mrope_section ................................... None
361
+ mscale .......................................... 1.0
362
+ mscale_all_dim .................................. 0.0
363
+ mtp_loss_scaling_factor ......................... 0.1
364
+ mtp_num_layers .................................. None
365
+ multi_latent_attention .......................... False
366
+ nccl_all_reduce_for_prefill ..................... False
367
+ nccl_communicator_config_path ................... None
368
+ nccl_ub ......................................... False
369
+ no_load_optim ................................... None
370
+ no_load_rng ..................................... None
371
+ no_persist_layer_norm ........................... False
372
+ no_rope_freq .................................... None
373
+ no_save_optim ................................... None
374
+ no_save_rng ..................................... None
375
+ non_persistent_ckpt_type ........................ None
376
+ non_persistent_global_ckpt_dir .................. None
377
+ non_persistent_local_ckpt_algo .................. fully_parallel
378
+ non_persistent_local_ckpt_dir ................... None
379
+ non_persistent_save_interval .................... None
380
+ norm_epsilon .................................... 1e-05
381
+ normalization ................................... RMSNorm
382
+ num_attention_heads ............................. 32
383
+ num_channels .................................... 3
384
+ num_classes ..................................... 1000
385
+ num_dataset_builder_threads ..................... 1
386
+ num_distributed_optimizer_instances ............. 1
387
+ num_experts ..................................... 64
388
+ num_layers ...................................... 16
389
+ num_layers_at_end_in_bf16 ....................... 1
390
+ num_layers_at_start_in_bf16 ..................... 1
391
+ num_layers_per_virtual_pipeline_stage ........... None
392
+ num_query_groups ................................ 8
393
+ num_virtual_stages_per_pipeline_rank ............ None
394
+ num_workers ..................................... 2
395
+ object_storage_cache_path ....................... None
396
+ one_logger_async ................................ False
397
+ one_logger_project .............................. megatron-lm
398
+ one_logger_run_name ............................. None
399
+ onnx_safe ....................................... None
400
+ openai_gelu ..................................... False
401
+ optimizer ....................................... adam
402
+ optimizer_cpu_offload ........................... False
403
+ optimizer_offload_fraction ...................... 1.0
404
+ output_bert_embeddings .......................... False
405
+ overlap_cpu_optimizer_d2h_h2d ................... False
406
+ overlap_grad_reduce ............................. True
407
+ overlap_moe_expert_parallel_comm ................ False
408
+ overlap_p2p_comm ................................ False
409
+ overlap_p2p_comm_warmup_flush ................... False
410
+ overlap_param_gather ............................ True
411
+ overlap_param_gather_with_optimizer_step ........ False
412
+ override_opt_param_scheduler .................... False
413
+ params_dtype .................................... torch.bfloat16
414
+ patch_dim ....................................... 16
415
+ per_split_data_args_path ........................ None
416
+ perform_initialization .......................... True
417
+ pin_cpu_grads ................................... True
418
+ pin_cpu_params .................................. True
419
+ pipeline_model_parallel_comm_backend ............ None
420
+ pipeline_model_parallel_layout .................. None
421
+ pipeline_model_parallel_size .................... 1
422
+ position_embedding_type ......................... rope
423
+ pretrained_checkpoint ........................... None
424
+ profile ......................................... False
425
+ profile_ranks ................................... [0]
426
+ profile_step_end ................................ 12
427
+ profile_step_start .............................. 10
428
+ q_lora_rank ..................................... None
429
+ qk_head_dim ..................................... 128
430
+ qk_l2_norm ...................................... False
431
+ qk_layernorm .................................... False
432
+ qk_pos_emb_head_dim ............................. 64
433
+ query_in_block_prob ............................. 0.1
434
+ rampup_batch_size ............................... None
435
+ rank ............................................ 0
436
+ recompute_granularity ........................... None
437
+ recompute_method ................................ None
438
+ recompute_modules ............................... None
439
+ recompute_num_layers ............................ None
440
+ record_memory_history ........................... False
441
+ relative_attention_max_distance ................. 128
442
+ relative_attention_num_buckets .................. 32
443
+ replication ..................................... False
444
+ replication_factor .............................. 2
445
+ replication_jump ................................ None
446
+ rerun_mode ...................................... validate_results
447
+ reset_attention_mask ............................ False
448
+ reset_position_ids .............................. False
449
+ result_rejected_tracker_filename ................ None
450
+ retriever_report_topk_accuracies ................ []
451
+ retriever_score_scaling ......................... False
452
+ retriever_seq_length ............................ 256
453
+ retro_add_retriever ............................. False
454
+ retro_attention_gate ............................ 1
455
+ retro_cyclic_train_iters ........................ None
456
+ retro_encoder_attention_dropout ................. 0.1
457
+ retro_encoder_hidden_dropout .................... 0.1
458
+ retro_encoder_layers ............................ 2
459
+ retro_num_neighbors ............................. 2
460
+ retro_num_retrieved_chunks ...................... 2
461
+ retro_project_dir ............................... None
462
+ retro_verify_neighbor_count ..................... True
463
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
464
+ rope_scaling_factor ............................. 8.0
465
+ rope_type ....................................... None
466
+ rotary_base ..................................... 1000000
467
+ rotary_interleaved .............................. False
468
+ rotary_percent .................................. 1.0
469
+ rotary_scaling_factor ........................... 1.0
470
+ rotary_seq_len_interpolation_factor ............. None
471
+ run_workload_inspector_server ................... False
472
+ sample_rate ..................................... 1.0
473
+ save ............................................ None
474
+ save_interval ................................... None
475
+ save_retain_interval ............................ None
476
+ scatter_gather_tensors_in_pipeline .............. True
477
+ seed ............................................ 1234
478
+ seq_length ...................................... 4096
479
+ sequence_parallel ............................... False
480
+ sft ............................................. False
481
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
482
+ sgd_momentum .................................... 0.9
483
+ short_seq_prob .................................. 0.1
484
+ skip_train ...................................... False
485
+ skipped_train_samples ........................... 0
486
+ spec ............................................ None
487
+ split ........................................... 100,0,0
488
+ squared_relu .................................... False
489
+ start_weight_decay .............................. 0.1
490
+ straggler_ctrlr_port ............................ 65535
491
+ straggler_minmax_count .......................... 1
492
+ suggested_communication_unit_size ............... None
493
+ swiglu .......................................... True
494
+ swin_backbone_type .............................. tiny
495
+ symmetric_ar_type ............................... None
496
+ te_rng_tracker .................................. True
497
+ tensor_model_parallel_size ...................... 1
498
+ tensorboard_dir ................................. None
499
+ tensorboard_log_interval ........................ 1
500
+ tensorboard_queue_size .......................... 1000
501
+ test_data_path .................................. None
502
+ test_mode ....................................... False
503
+ tiktoken_num_special_tokens ..................... 1000
504
+ tiktoken_pattern ................................ None
505
+ tiktoken_special_tokens ......................... None
506
+ timing_log_level ................................ 0
507
+ timing_log_option ............................... minmax
508
+ titles_data_path ................................ None
509
+ tokenizer_model ................................. None
510
+ tokenizer_type .................................. GPT2BPETokenizer
511
+ torch_fsdp2_reshard_after_forward ............... True
512
+ tp_comm_bootstrap_backend ....................... nccl
513
+ tp_comm_bulk_dgrad .............................. True
514
+ tp_comm_bulk_wgrad .............................. True
515
+ tp_comm_overlap ................................. False
516
+ tp_comm_overlap_ag .............................. True
517
+ tp_comm_overlap_cfg ............................. None
518
+ tp_comm_overlap_rs .............................. True
519
+ tp_comm_overlap_rs_dgrad ........................ False
520
+ tp_comm_split_ag ................................ True
521
+ tp_comm_split_rs ................................ True
522
+ train_data_path ................................. None
523
+ train_iters ..................................... 12
524
+ train_samples ................................... None
525
+ train_sync_interval ............................. None
526
+ transformer_impl ................................ transformer_engine
527
+ transformer_pipeline_model_parallel_size ........ 1
528
+ untie_embeddings_and_output_weights ............. True
529
+ use_checkpoint_args ............................. False
530
+ use_checkpoint_opt_param_scheduler .............. False
531
+ use_cpu_initialization .......................... None
532
+ use_custom_fsdp ................................. False
533
+ use_dist_ckpt ................................... True
534
+ use_dist_ckpt_deprecated ........................ False
535
+ use_distributed_optimizer ....................... True
536
+ use_flash_attn .................................. False
537
+ use_legacy_models ............................... False
538
+ use_mp_args_from_checkpoint_args ................ False
539
+ use_one_sent_docs ............................... False
540
+ use_persistent_ckpt_worker ...................... False
541
+ use_precision_aware_optimizer ................... False
542
+ use_pytorch_profiler ............................ False
543
+ use_ring_exchange_p2p ........................... False
544
+ use_rope_scaling ................................ False
545
+ use_rotary_position_embeddings .................. False
546
+ use_sharp ....................................... False
547
+ use_tokenizer_model_from_checkpoint_args ........ True
548
+ use_torch_fsdp2 ................................. False
549
+ use_torch_optimizer_for_cpu_offload ............. False
550
+ use_tp_pp_dp_mapping ............................ False
551
+ v_head_dim ...................................... 128
552
+ valid_data_path ................................. None
553
+ variable_seq_lengths ............................ False
554
+ virtual_pipeline_model_parallel_size ............ None
555
+ vision_backbone_type ............................ vit
556
+ vision_pretraining .............................. False
557
+ vision_pretraining_type ......................... classify
558
+ vocab_extra_ids ................................. 0
559
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
560
+ vocab_size ...................................... None
561
+ wandb_exp_name ..................................
562
+ wandb_project ...................................
563
+ wandb_save_dir ..................................
564
+ weight_decay .................................... 0.1
565
+ weight_decay_incr_style ......................... constant
566
+ wgrad_deferral_limit ............................ 0
567
+ world_size ...................................... 8
568
+ yaml_cfg ........................................ None
569
+ -------------------- end of arguments ---------------------
570
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
571
+ > building GPT2BPETokenizer tokenizer ...
572
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
573
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
574
+ > initializing torch distributed ...
575
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
576
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
577
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
578
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
579
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
580
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
581
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
582
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank [Gloo] Rank 3 is connected to 72 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks.
586
+ Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 7[Gloo] Rank is connected to 7 peer ranks. 6Expected number of connected peer ranks is : is connected to 77
589
+ peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank [Gloo] Rank 6 is connected to 77 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks.
597
+ Expected number of connected peer ranks is : 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [Gloo] Rank 0[Gloo] Rank is connected to 00 peer ranks. is connected to Expected number of connected peer ranks is : 00 peer ranks.
604
+ Expected number of connected peer ranks is : 0
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
607
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank5]:[W830 07:09:33.453758563 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank4]:[W830 07:09:33.454049433 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank3]:[W830 07:09:33.455848346 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank1]:[W830 07:09:33.455905666 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
613
+ [rank2]:[W830 07:09:33.455971468 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
614
+ [rank7]:[W830 07:09:33.455995755 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank6]:[W830 07:09:33.456097760 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.034 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W830 07:09:33.624893529 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.787 seconds
625
+ time to initialize megatron (seconds): 6.854
626
+ [after megatron is initialized] datetime: 2026-08-30 07:09:39
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 4258402304
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
667
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (42217472 elements, 42217472 padded size):
670
+ module.decoder.final_layernorm.weight
671
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
672
+ module.decoder.layers.15.mlp.router.weight
673
+ module.decoder.layers.15.self_attention.linear_proj.weight
674
+ module.decoder.layers.15.pre_mlp_layernorm.weight
675
+ module.decoder.layers.15.self_attention.linear_qkv.weight
676
+ Params for bucket 3 (42213376 elements, 42213376 padded size):
677
+ module.decoder.layers.14.mlp.router.weight
678
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
679
+ module.decoder.layers.14.self_attention.linear_proj.weight
680
+ module.decoder.layers.14.self_attention.linear_qkv.weight
681
+ module.decoder.layers.14.pre_mlp_layernorm.weight
682
+ Params for bucket 4 (42213376 elements, 42213376 padded size):
683
+ module.decoder.layers.13.mlp.router.weight
684
+ module.decoder.layers.13.self_attention.linear_proj.weight
685
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
686
+ module.decoder.layers.13.self_attention.linear_qkv.weight
687
+ module.decoder.layers.13.pre_mlp_layernorm.weight
688
+ Params for bucket 5 (42213376 elements, 42213376 padded size):
689
+ module.decoder.layers.12.mlp.router.weight
690
+ module.decoder.layers.12.self_attention.linear_proj.weight
691
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
692
+ module.decoder.layers.12.pre_mlp_layernorm.weight
693
+ module.decoder.layers.12.self_attention.linear_qkv.weight
694
+ Params for bucket 6 (42213376 elements, 42213376 padded size):
695
+ module.decoder.layers.11.mlp.router.weight
696
+ module.decoder.layers.11.self_attention.linear_proj.weight
697
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.11.pre_mlp_layernorm.weight
699
+ module.decoder.layers.11.self_attention.linear_qkv.weight
700
+ Params for bucket 7 (42213376 elements, 42213376 padded size):
701
+ module.decoder.layers.10.mlp.router.weight
702
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
703
+ module.decoder.layers.10.self_attention.linear_proj.weight
704
+ module.decoder.layers.10.self_attention.linear_qkv.weight
705
+ module.decoder.layers.10.pre_mlp_layernorm.weight
706
+ Params for bucket 8 (42213376 elements, 42213376 padded size):
707
+ module.decoder.layers.9.pre_mlp_layernorm.weight
708
+ module.decoder.layers.9.self_attention.linear_qkv.weight
709
+ module.decoder.layers.9.mlp.router.weight
710
+ module.decoder.layers.9.self_attention.linear_proj.weight
711
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
712
+ Params for bucket 9 (42213376 elements, 42213376 padded size):
713
+ module.decoder.layers.8.self_attention.linear_qkv.weight
714
+ module.decoder.layers.8.pre_mlp_layernorm.weight
715
+ module.decoder.layers.8.mlp.router.weight
716
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
717
+ module.decoder.layers.8.self_attention.linear_proj.weight
718
+ Params for bucket 10 (42213376 elements, 42213376 padded size):
719
+ module.decoder.layers.7.self_attention.linear_qkv.weight
720
+ module.decoder.layers.7.pre_mlp_layernorm.weight
721
+ module.decoder.layers.7.mlp.router.weight
722
+ module.decoder.layers.7.self_attention.linear_proj.weight
723
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
724
+ Params for bucket 11 (42213376 elements, 42213376 padded size):
725
+ module.decoder.layers.6.self_attention.linear_qkv.weight
726
+ module.decoder.layers.6.pre_mlp_layernorm.weight
727
+ module.decoder.layers.6.mlp.router.weight
728
+ module.decoder.layers.6.self_attention.linear_proj.weight
729
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
730
+ Params for bucket 12 (42213376 elements, 42213376 padded size):
731
+ module.decoder.layers.5.mlp.router.weight
732
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
733
+ module.decoder.layers.5.self_attention.linear_qkv.weight
734
+ module.decoder.layers.5.self_attention.linear_proj.weight
735
+ module.decoder.layers.5.pre_mlp_layernorm.weight
736
+ Params for bucket 13 (42213376 elements, 42213376 padded size):
737
+ module.decoder.layers.4.pre_mlp_layernorm.weight
738
+ module.decoder.layers.4.self_attention.linear_qkv.weight
739
+ module.decoder.layers.4.mlp.router.weight
740
+ module.decoder.layers.4.self_attention.linear_proj.weight
741
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
742
+ Params for bucket 14 (42213376 elements, 42213376 padded size):
743
+ module.decoder.layers.3.mlp.router.weight
744
+ module.decoder.layers.3.self_attention.linear_proj.weight
745
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
746
+ module.decoder.layers.3.self_attention.linear_qkv.weight
747
+ module.decoder.layers.3.pre_mlp_layernorm.weight
748
+ Params for bucket 15 (42213376 elements, 42213376 padded size):
749
+ module.decoder.layers.2.self_attention.linear_qkv.weight
750
+ module.decoder.layers.2.pre_mlp_layernorm.weight
751
+ module.decoder.layers.2.mlp.router.weight
752
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
753
+ module.decoder.layers.2.self_attention.linear_proj.weight
754
+ Params for bucket 16 (42213376 elements, 42213376 padded size):
755
+ module.decoder.layers.1.pre_mlp_layernorm.weight
756
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
757
+ module.decoder.layers.1.self_attention.linear_proj.weight
758
+ module.decoder.layers.1.mlp.router.weight
759
+ module.decoder.layers.1.self_attention.linear_qkv.weight
760
+ Params for bucket 17 (42213376 elements, 42213376 padded size):
761
+ module.decoder.layers.0.mlp.router.weight
762
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
763
+ module.decoder.layers.0.self_attention.linear_proj.weight
764
+ module.decoder.layers.0.pre_mlp_layernorm.weight
765
+ module.decoder.layers.0.self_attention.linear_qkv.weight
766
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
767
+ module.embedding.word_embeddings.weight
768
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 64
769
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
770
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight7
771
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
772
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight5
773
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight6
774
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight4
775
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
776
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
777
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
778
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
779
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight6
780
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight7
781
+ Params for bucket 3 (44040192 elements, 44040192 padded size):
782
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight5
783
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
784
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight4
785
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
786
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
787
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
788
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
789
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
790
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight7
791
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
792
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight5
793
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight4
794
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight6
795
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
796
+ Params for bucket 6 (44040192 elements, 44040192 padded size):
797
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
798
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
799
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight6
800
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight7
801
+ Params for bucket 7 (44040192 elements, 44040192 padded size):
802
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight5
803
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
804
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight4
805
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
806
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
807
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
808
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
809
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
810
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight7
811
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight5
812
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
813
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight4
814
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight6
815
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
816
+ Params for bucket 10 (44040192 elements, 44040192 padded size):
817
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
818
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight6
819
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
820
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight7
821
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
822
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight5
823
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
824
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight4
825
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
827
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
828
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
829
+ Params for bucket 13 (44040192 elements, 44040192 padded size):
830
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight7
831
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight5
832
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
833
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight6
834
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight4
835
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
836
+ Params for bucket 14 (44040192 elements, 44040192 padded size):
837
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
838
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight7
839
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight6
840
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
841
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
842
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight5
843
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight4
844
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
845
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
846
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
847
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
848
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
849
+ Params for bucket 17 (44040192 elements, 44040192 padded size):
850
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight7
851
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight5
852
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
853
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight6
854
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight4
855
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
856
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
857
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
858
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight7
859
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
860
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight6
861
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
862
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight5
863
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight4
864
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
865
+ Params for bucket 20 (44040192 elements, 44040192 padded size):
866
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
867
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
868
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
869
+ Params for bucket 21 (44040192 elements, 44040192 padded size):
870
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight7
871
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight5
872
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
873
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight6
874
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight4
875
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
876
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
877
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
878
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight7
879
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
880
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight6
881
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
882
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight5
883
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight4
884
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
885
+ Params for bucket 24 (44040192 elements, 44040192 padded size):
886
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
887
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
888
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
889
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
890
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight6
891
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight4
892
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
893
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight7
894
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
895
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight5
896
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
897
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
898
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
899
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight7
900
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight6
901
+ Params for bucket 27 (44040192 elements, 44040192 padded size):
902
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight5
903
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight4
904
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
905
+ Params for bucket 28 (44040192 elements, 44040192 padded size):
906
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
907
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
908
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
909
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
910
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight6
911
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight4
912
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
913
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight7
914
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
915
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight5
916
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
917
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
918
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
919
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight6
920
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight7
921
+ Params for bucket 31 (44040192 elements, 44040192 padded size):
922
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight5
923
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
924
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight4
925
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
926
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
927
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
928
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
929
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
930
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight6
931
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight4
932
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
933
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight7
934
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight5
935
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
936
+ Params for bucket 34 (44040192 elements, 44040192 padded size):
937
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
938
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
939
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight6
940
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight7
941
+ Params for bucket 35 (44040192 elements, 44040192 padded size):
942
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight5
943
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
944
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight4
945
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
946
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
947
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
948
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
949
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
950
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight4
951
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight6
952
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
953
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight7
954
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight5
955
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
956
+ Params for bucket 38 (44040192 elements, 44040192 padded size):
957
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
958
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight7
959
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight6
960
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
961
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
962
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight5
963
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
964
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight4
965
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
966
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
967
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
968
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
969
+ Params for bucket 41 (44040192 elements, 44040192 padded size):
970
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight6
971
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight4
972
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
973
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight7
974
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight5
975
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
976
+ Params for bucket 42 (44040192 elements, 44040192 padded size):
977
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
978
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight7
979
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
980
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight6
981
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
982
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight5
983
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight4
984
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
985
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
986
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
987
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
988
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
989
+ Params for bucket 45 (44040192 elements, 44040192 padded size):
990
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight6
991
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight4
992
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
993
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight7
994
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight5
995
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
996
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
997
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
998
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight7
999
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
1000
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight6
1001
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
1002
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight5
1003
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight4
1004
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
1005
+ Params for bucket 48 (44040192 elements, 44040192 padded size):
1006
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
1007
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
1008
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
1009
+ Params for bucket 49 (44040192 elements, 44040192 padded size):
1010
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight6
1011
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight4
1012
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
1013
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight7
1014
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
1015
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight5
1016
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
1017
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
1018
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight6
1019
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight7
1020
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
1021
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
1022
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight5
1023
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
1024
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight4
1025
+ Params for bucket 52 (44040192 elements, 44040192 padded size):
1026
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
1027
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
1028
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
1029
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
1030
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight7
1031
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight5
1032
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
1033
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight6
1034
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight4
1035
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
1036
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
1037
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
1038
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight7
1039
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight6
1040
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
1041
+ Params for bucket 55 (44040192 elements, 44040192 padded size):
1042
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight5
1043
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight4
1044
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
1045
+ Params for bucket 56 (44040192 elements, 44040192 padded size):
1046
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
1047
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
1048
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
1049
+ Params for bucket 57 (44040192 elements, 44040192 padded size):
1050
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight6
1051
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight4
1052
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
1053
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight7
1054
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight5
1055
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
1056
+ Params for bucket 58 (44040192 elements, 44040192 padded size):
1057
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
1058
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight7
1059
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
1060
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight6
1061
+ Params for bucket 59 (44040192 elements, 44040192 padded size):
1062
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight5
1063
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight4
1064
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
1065
+ Params for bucket 60 (44040192 elements, 44040192 padded size):
1066
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
1067
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
1068
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
1069
+ Params for bucket 61 (44040192 elements, 44040192 padded size):
1070
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight4
1071
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight6
1072
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
1073
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight7
1074
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
1075
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight5
1076
+ Params for bucket 62 (44040192 elements, 44040192 padded size):
1077
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
1078
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight7
1079
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight6
1080
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
1081
+ Params for bucket 63 (44040192 elements, 44040192 padded size):
1082
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight5
1083
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
1084
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight4
1085
+ Params for bucket 64 (44040192 elements, 44040192 padded size):
1086
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
1087
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
1088
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
1089
+ INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=<megatron.core.timers.Timers object at 0x75a70448d070>, config_logger_dir='')
1090
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
1091
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:09:42
1092
+ > building train, validation, and test datasets ...
1093
+ > datasets target sizes (minimum size):
1094
+ train: 192
1095
+ validation: 0
1096
+ test: 0
1097
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
1098
+ > building train, validation, and test datasets for GPT ...
1099
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x75a7045bffe0>, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None)
1100
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
1101
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
1102
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
1103
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
1104
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
1105
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
1106
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
1107
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
1108
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
1109
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
1110
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
1111
+ > finished creating GPT datasets ...
1112
+ [after dataloaders are built] datetime: 2026-08-30 07:09:42
1113
+ done with setup ...
1114
+ training ...
1115
+ (min, max) time across ranks (ms):
1116
+ model-and-optimizer-setup ......................: (3606.74, 3611.63)
1117
+ train/valid/test-data-iterators-setup ..........: (73.10, 94.93)
1118
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
1119
+ [before the start of training step] datetime: 2026-08-30 07:09:42
1120
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1121
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1122
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1123
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1124
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1125
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1126
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1127
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1128
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1129
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1130
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1131
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1132
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1133
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1134
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1135
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1136
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1137
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1138
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1139
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1140
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1141
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1142
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1143
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1144
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1145
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1146
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1147
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1148
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1149
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1150
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1151
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1152
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1153
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1154
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1155
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1156
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1157
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1158
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1159
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1160
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1161
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1162
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1163
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1164
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1165
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1166
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1167
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1168
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1169
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1170
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1171
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1172
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1173
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1174
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1175
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1176
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1177
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1178
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1179
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1180
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1181
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1182
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1183
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1184
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1185
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1186
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1187
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1188
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1189
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1190
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1191
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1192
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1193
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1194
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1195
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1196
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1197
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1198
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1199
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1200
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1201
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1202
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1203
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1204
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1205
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1206
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1207
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1208
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1209
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1210
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1211
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1212
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1213
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1214
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1215
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1216
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1217
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1218
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1219
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1220
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1221
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1222
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1223
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1224
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1225
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1226
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1227
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1228
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1229
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1230
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1231
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1232
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1233
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1234
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1235
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1236
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1237
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1238
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1239
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1240
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1241
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1242
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1243
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1244
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1245
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1246
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1247
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1248
+ [2026-08-30 07:10:09] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26830.2 | throughput per GPU (TFLOP/s/GPU): 12.4 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103531E+01 | loss scale: 1.0 | grad norm: 44.062 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1249
+ Number of parameters in transformer block in billions: 23.22
1250
+ Number of parameters in embedding layers in billions: 0.41
1251
+ Total number of parameters in billions: 23.63
1252
+ Number of parameters in most loaded shard in billions: 23.6320
1253
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1254
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.49609375 | max allocated: 67206.4716796875 | reserved: 68904.0 | max reserved: 68904.0
1255
+ [2026-08-30 07:10:10] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 499.4 | throughput per GPU (TFLOP/s/GPU): 667.5 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.103001E+01 | loss scale: 1.0 | grad norm: 45.975 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1256
+ [2026-08-30 07:10:10] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 465.1 | throughput per GPU (TFLOP/s/GPU): 716.8 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.043762E+01 | loss scale: 1.0 | grad norm: 205.433 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1257
+ Capture CUDA graph for training!!!
1258
+ CUDA graph capture done!!!
1259
+ [2026-08-30 07:10:12] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2075.9 | throughput per GPU (TFLOP/s/GPU): 160.6 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.319299E+01 | loss scale: 1.0 | grad norm: 35.558 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1260
+ [2026-08-30 07:10:13] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 434.9 | throughput per GPU (TFLOP/s/GPU): 766.5 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.039594E+01 | loss scale: 1.0 | grad norm: 15.612 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1261
+ [2026-08-30 07:10:13] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 452.1 | throughput per GPU (TFLOP/s/GPU): 737.4 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.046178E+01 | loss scale: 1.0 | grad norm: 39.299 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1262
+ [2026-08-30 07:10:14] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 447.3 | throughput per GPU (TFLOP/s/GPU): 745.4 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.002575E+01 | loss scale: 1.0 | grad norm: 8.388 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1263
+ [2026-08-30 07:10:14] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 455.3 | throughput per GPU (TFLOP/s/GPU): 732.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.540670E+00 | loss scale: 1.0 | grad norm: 5.808 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1264
+ [2026-08-30 07:10:15] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 441.8 | throughput per GPU (TFLOP/s/GPU): 754.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.310684E+00 | loss scale: 1.0 | grad norm: 5.650 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1265
+ [2026-08-30 07:10:15] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 458.9 | throughput per GPU (TFLOP/s/GPU): 726.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.154215E+00 | loss scale: 1.0 | grad norm: 5.472 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1266
+ [2026-08-30 07:10:15] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 446.2 | throughput per GPU (TFLOP/s/GPU): 747.1 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.034491E+00 | loss scale: 1.0 | grad norm: 5.276 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1267
+ [2026-08-30 07:10:16] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 435.8 | throughput per GPU (TFLOP/s/GPU): 765.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.910418E+00 | loss scale: 1.0 | grad norm: 5.353 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 70272 | min safe capacity: 1.0723 |
1268
+ [after training is done] datetime: 2026-08-30 07:10:16
1269
+ [rank0]:[W830 07:10:17.720931925 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1270
+ [rank2]:[W830 07:10:18.176428673 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1271
+ [rank3]:[W830 07:10:18.176802808 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1272
+ [rank7]:[W830 07:10:18.179698894 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1273
+ [rank5]:[W830 07:10:18.185278831 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1274
+ [rank6]:[W830 07:10:19.258037997 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1275
+ [rank1]:[W830 07:10:19.343659403 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1276
+ [rank4]:[W830 07:10:19.783715306 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1277
+ W0830 07:10:24.251000 714645 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1278
+ W0830 07:10:24.252000 714645 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 714735 closing signal SIGTERM
1279
+ W0830 07:10:24.252000 714645 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 714736 closing signal SIGTERM
1280
+ W0830 07:10:24.253000 714645 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 714737 closing signal SIGTERM
1281
+ W0830 07:10:24.253000 714645 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 714738 closing signal SIGTERM
1282
+ W0830 07:10:24.254000 714645 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 714740 closing signal SIGTERM
1283
+ Traceback (most recent call last):
1284
+ File "/usr/local/bin/torchrun", line 8, in <module>
1285
+ sys.exit(main())
1286
+ ^^^^^^
1287
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1288
+ return f(*args, **kwargs)
1289
+ ^^^^^^^^^^^^^^^^^^
1290
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1291
+ run(args)
1292
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1293
+ elastic_launch(
1294
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1295
+ return launch_agent(self._config, self._entrypoint, list(args))
1296
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1297
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1298
+ result = agent.run()
1299
+ ^^^^^^^^^^^
1300
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1301
+ result = f(*args, **kwargs)
1302
+ ^^^^^^^^^^^^^^^^^^
1303
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1304
+ result = self._invoke_run(role)
1305
+ ^^^^^^^^^^^^^^^^^^^^^^
1306
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1307
+ time.sleep(monitor_interval)
1308
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1309
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1310
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 714645 got signal: 15
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 1
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep3/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=G3
3
+ family=tierA
4
+ repo=/workspace/Megatron-UltraEP
5
+ branch=ultraep-bracing-regions-through-kernels
6
+ commit=03d57ef7
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=1792
11
+ num_experts=64
12
+ topk=16
13
+ experts_per_rank=8
14
+ expert_params=22020096
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=65536
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.072265625
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=1024
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep3/train.log ADDED
@@ -0,0 +1,576 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G3 / ultraep (slots/rank=1 policy=one)
3
+ E=64 K=16 H'=1792 m=8
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.072265625 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep3/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0830 07:10:27.741000 717704 torch/distributed/run.py:766]
10
+ W0830 07:10:27.741000 717704 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:10:27.741000 717704 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 07:10:27.741000 717704 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
18
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
19
+ Number of virtual stages per pipeline stage: None
20
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
21
+ using torch.bfloat16 for parameters ...
22
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
23
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
24
+ ------------------------ arguments ------------------------
25
+ account_for_embedding_in_pipeline_split ......... False
26
+ account_for_loss_in_pipeline_split .............. False
27
+ accumulate_allreduce_grads_in_fp32 .............. True
28
+ adam_beta1 ...................................... 0.9
29
+ adam_beta2 ...................................... 0.999
30
+ adam_eps ........................................ 1e-08
31
+ add_bias_linear ................................. False
32
+ add_position_embedding .......................... False
33
+ add_qkv_bias .................................... False
34
+ adlr_autoresume ................................. False
35
+ adlr_autoresume_interval ........................ 1000
36
+ align_grad_reduce ............................... True
37
+ align_param_gather .............................. False
38
+ app_tag_run_name ................................ None
39
+ app_tag_run_version ............................. 0.0.0
40
+ apply_layernorm_1p .............................. False
41
+ apply_query_key_layer_scaling ................... False
42
+ apply_residual_connection_post_layernorm ........ False
43
+ apply_rope_fusion ............................... True
44
+ async_save ...................................... None
45
+ async_tensor_model_parallel_allreduce ........... True
46
+ attention_backend ............................... AttnBackend.auto
47
+ attention_dropout ............................... 0.0
48
+ attention_softmax_in_fp32 ....................... False
49
+ auto_detect_ckpt_format ......................... False
50
+ barrier_with_L1_time ............................ True
51
+ bert_binary_head ................................ True
52
+ bert_embedder_type .............................. megatron
53
+ bert_load ....................................... None
54
+ bf16 ............................................ True
55
+ bias_dropout_fusion ............................. True
56
+ bias_gelu_fusion ................................ False
57
+ bias_swiglu_fusion .............................. True
58
+ biencoder_projection_dim ........................ 0
59
+ biencoder_shared_query_context_model ............ False
60
+ block_data_path ................................. None
61
+ calc_ft_timeouts ................................ False
62
+ calculate_per_token_loss ........................ False
63
+ check_for_large_grads ........................... False
64
+ check_for_nan_in_loss_and_grad .................. False
65
+ check_for_spiky_loss ............................ False
66
+ check_weight_hash_across_dp_replicas_interval ... None
67
+ ckpt_assume_constant_structure .................. False
68
+ ckpt_convert_format ............................. None
69
+ ckpt_convert_save ............................... None
70
+ ckpt_convert_update_legacy_dist_opt_format ...... False
71
+ ckpt_format ..................................... torch_dist
72
+ ckpt_fully_parallel_load ........................ False
73
+ ckpt_fully_parallel_save ........................ True
74
+ ckpt_fully_parallel_save_deprecated ............. False
75
+ ckpt_step ....................................... None
76
+ classes_fraction ................................ 1.0
77
+ clip_grad ....................................... 1.0
78
+ clone_scatter_output_in_embedding ............... True
79
+ config_logger_dir ...............................
80
+ consumed_train_samples .......................... 0
81
+ consumed_valid_samples .......................... 0
82
+ context_parallel_size ........................... 1
83
+ cp_comm_type .................................... ['p2p']
84
+ create_attention_mask_in_dataloader ............. True
85
+ cross_entropy_fusion_impl ....................... native
86
+ cross_entropy_loss_fusion ....................... False
87
+ cuda_graph_scope ................................ full_iteration
88
+ cuda_graph_warmup_steps ......................... 3
89
+ data_args_path .................................. None
90
+ data_cache_path ................................. None
91
+ data_parallel_random_init ....................... False
92
+ data_parallel_sharding_strategy ................. no_shard
93
+ data_parallel_size .............................. 8
94
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
95
+ data_per_class_fraction ......................... 1.0
96
+ data_sharding ................................... True
97
+ dataloader_type ................................. single
98
+ ddp_average_in_collective ....................... False
99
+ ddp_bucket_size ................................. None
100
+ ddp_num_buckets ................................. None
101
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
102
+ decoder_first_pipeline_num_layers ............... None
103
+ decoder_last_pipeline_num_layers ................ None
104
+ decoder_num_layers .............................. None
105
+ decoder_seq_length .............................. None
106
+ decoupled_lr .................................... None
107
+ decoupled_min_lr ................................ None
108
+ decrease_batch_size_if_needed ................... False
109
+ defer_embedding_wgrad_compute ................... False
110
+ delay_wgrad_compute ............................. False
111
+ deprecated_use_mcore_models ..................... True
112
+ deterministic_mode .............................. False
113
+ dino_bottleneck_size ............................ 256
114
+ dino_freeze_last_layer .......................... 1
115
+ dino_head_hidden_size ........................... 2048
116
+ dino_local_crops_number ......................... 10
117
+ dino_local_img_size ............................. 96
118
+ dino_norm_last_layer ............................ False
119
+ dino_teacher_temp ............................... 0.07
120
+ dino_warmup_teacher_temp ........................ 0.04
121
+ dino_warmup_teacher_temp_epochs ................. 30
122
+ disable_bf16_reduced_precision_matmul ........... False
123
+ disable_mamba_mem_eff_path ...................... False
124
+ disable_straggler_on_startup .................... False
125
+ dist_ckpt_format_deprecated ..................... None
126
+ dist_ckpt_strictness ............................ assume_ok_unexpected
127
+ distribute_saved_activations .................... False
128
+ distributed_backend ............................. nccl
129
+ distributed_timeout_minutes ..................... 10
130
+ embedding_init_method_std ....................... None
131
+ embedding_path .................................. None
132
+ empty_unused_memory_level ....................... 0
133
+ enable_cuda_graph ............................... True
134
+ enable_experimental ............................. False
135
+ enable_ft_package ............................... False
136
+ enable_gloo_process_groups ...................... True
137
+ enable_msc ...................................... True
138
+ enable_one_logger ............................... True
139
+ encoder_num_layers .............................. 16
140
+ encoder_seq_length .............................. 4096
141
+ end_weight_decay ................................ 0.1
142
+ eod_mask_loss ................................... False
143
+ error_injection_rate ............................ 0
144
+ error_injection_type ............................ transient_error
145
+ eval_interval ................................... 1000000
146
+ eval_iters ...................................... 0
147
+ evidence_data_path .............................. None
148
+ exit_duration_in_mins ........................... None
149
+ exit_interval ................................... None
150
+ exit_on_missing_checkpoint ...................... False
151
+ exit_signal_handler ............................. False
152
+ exp_avg_dtype ................................... torch.float32
153
+ exp_avg_sq_dtype ................................ torch.float32
154
+ expert_model_parallel_size ...................... 8
155
+ expert_tensor_parallel_size ..................... 1
156
+ export_force_local_attention .................... False
157
+ export_kd_cfg ................................... None
158
+ export_kd_teacher_ckpt_format ................... None
159
+ export_kd_teacher_load .......................... None
160
+ export_kv_cache_quant ........................... False
161
+ export_legacy_megatron .......................... False
162
+ export_model_type ............................... GPTModel
163
+ export_moe_apply_probs_on_input ................. False
164
+ export_qk_l2_norm ............................... False
165
+ export_quant_cfg ................................ None
166
+ export_real_quant_cfg ........................... None
167
+ export_te_mcore_model ........................... False
168
+ external_cuda_graph ............................. False
169
+ ffn_hidden_size ................................. 14336
170
+ finetune ........................................ False
171
+ finetune_data_split ............................. train
172
+ finetune_hf_dataset ............................. None
173
+ first_last_layers_bf16 .......................... False
174
+ flash_decode .................................... False
175
+ fp16 ............................................ False
176
+ fp16_lm_cross_entropy ........................... False
177
+ fp32_residual_connection ........................ False
178
+ fp8 ............................................. e4m3
179
+ fp8_amax_compute_algo ........................... most_recent
180
+ fp8_amax_history_len ............................ 1
181
+ fp8_interval .................................... 1
182
+ fp8_margin ...................................... 0
183
+ fp8_param_gather ................................ False
184
+ fp8_recipe ...................................... mxfp8
185
+ fp8_wgrad ....................................... True
186
+ fsdp_double_buffer .............................. False
187
+ global_batch_size ............................... 16
188
+ grad_reduce_in_bf16 ............................. False
189
+ gradient_accumulation_fusion .................... True
190
+ gradient_reduce_div_fusion ...................... True
191
+ group_query_attention ........................... True
192
+ head_lr_mult .................................... 1.0
193
+ heterogeneous_layers_config_encoded_json ........ None
194
+ heterogeneous_layers_config_path ................ None
195
+ hidden_dropout .................................. 0.0
196
+ hidden_size ..................................... 4096
197
+ hierarchical_context_parallel_sizes ............. None
198
+ high_priority_stream_groups ..................... []
199
+ hybrid_attention_ratio .......................... 0.0
200
+ hybrid_mlp_ratio ................................ 0.0
201
+ hybrid_override_pattern ......................... None
202
+ hysteresis ...................................... 2
203
+ ict_head_size ................................... None
204
+ ict_load ........................................ None
205
+ img_h ........................................... 224
206
+ img_w ........................................... 224
207
+ indexer_batch_size .............................. 128
208
+ indexer_log_interval ............................ 1000
209
+ inference_batch_times_seqlen_threshold .......... -1
210
+ inference_dynamic_batching ...................... False
211
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
212
+ inference_dynamic_batching_buffer_overflow_factor None
213
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
214
+ inference_dynamic_batching_chunk_size ........... 256
215
+ inference_dynamic_batching_max_requests_override None
216
+ inference_dynamic_batching_max_tokens_override .. None
217
+ inference_dynamic_batching_num_cuda_graphs ...... 16
218
+ inference_max_batch_size ........................ 8
219
+ inference_max_seq_length ........................ 2560
220
+ inference_rng_tracker ........................... False
221
+ init_method_std ................................. 0.01
222
+ init_method_xavier_uniform ...................... False
223
+ init_model_with_meta_device ..................... False
224
+ initial_loss_scale .............................. 4294967296
225
+ inprocess_active_world_size ..................... 8
226
+ inprocess_barrier_timeout ....................... 120
227
+ inprocess_completion_timeout .................... 120
228
+ inprocess_empty_cuda_cache ...................... False
229
+ inprocess_granularity ........................... node
230
+ inprocess_hard_timeout .......................... 90
231
+ inprocess_heartbeat_interval .................... 30
232
+ inprocess_heartbeat_timeout ..................... 60
233
+ inprocess_last_call_wait ........................ 1
234
+ inprocess_max_iterations ........................ None
235
+ inprocess_monitor_process_interval .............. 1.0
236
+ inprocess_monitor_thread_interval ............... 1.0
237
+ inprocess_progress_watchdog_interval ............ 1.0
238
+ inprocess_restart ............................... False
239
+ inprocess_soft_timeout .......................... 60
240
+ inprocess_termination_grace_time ................ 1
241
+ is_hybrid_model ................................. False
242
+ iter_per_epoch .................................. 1250
243
+ iterations_to_skip .............................. []
244
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
245
+ kitchen_config_file ............................. None
246
+ kitchen_recipe_number ........................... None
247
+ kv_channels ..................................... 128
248
+ kv_lora_rank .................................... 32
249
+ lazy_mpu_init ................................... None
250
+ load ............................................ None
251
+ load_main_params_from_ckpt ...................... None
252
+ load_model_opt_format ........................... False
253
+ local_rank ...................................... 0
254
+ log_energy ...................................... False
255
+ log_interval .................................... 1
256
+ log_loss_scale_to_tensorboard ................... True
257
+ log_memory_to_tensorboard ....................... False
258
+ log_num_zeros_in_grad ........................... False
259
+ log_params_norm ................................. False
260
+ log_progress .................................... False
261
+ log_straggler ................................... False
262
+ log_throughput .................................. True
263
+ log_timers_to_tensorboard ....................... False
264
+ log_validation_ppl_to_tensorboard ............... False
265
+ log_world_size_to_tensorboard ................... False
266
+ logging_level ................................... None
267
+ loss_scale ...................................... None
268
+ loss_scale_window ............................... 1000
269
+ lr .............................................. 0.0001
270
+ lr_decay_iters .................................. 8
271
+ lr_decay_samples ................................ None
272
+ lr_decay_style .................................. cosine
273
+ lr_warmup_fraction .............................. None
274
+ lr_warmup_init .................................. 0.0
275
+ lr_warmup_iters ................................. 5
276
+ lr_warmup_samples ............................... 0
277
+ lr_wsd_decay_iters .............................. None
278
+ lr_wsd_decay_samples ............................ None
279
+ lr_wsd_decay_style .............................. exponential
280
+ main_grads_dtype ................................ torch.float32
281
+ main_params_dtype ............................... torch.float32
282
+ make_vocab_size_divisible_by .................... 128
283
+ mamba_head_dim .................................. 64
284
+ mamba_num_groups ................................ 8
285
+ mamba_num_heads ................................. None
286
+ mamba_state_dim ................................. 128
287
+ manual_gc ....................................... False
288
+ manual_gc_eval .................................. True
289
+ manual_gc_interval .............................. 0
290
+ mask_factor ..................................... 1.0
291
+ mask_prob ....................................... 0.15
292
+ mask_type ....................................... random
293
+ masked_softmax_fusion ........................... False
294
+ max_position_embeddings ......................... 32768
295
+ max_tokens_to_oom ............................... 12000
296
+ memory_snapshot_path ............................ snapshot.pickle
297
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
298
+ micro_batch_size ................................ 1
299
+ microbatch_group_size_per_vp_stage .............. None
300
+ mid_level_dataset_surplus ....................... 0.005
301
+ min_loss_scale .................................. 1.0
302
+ min_lr .......................................... 1e-05
303
+ mlp_chunks_for_prefill .......................... 1
304
+ mmap_bin_files .................................. True
305
+ mock_data ....................................... False
306
+ moe_apply_probs_on_input ........................ False
307
+ moe_aux_loss_coeff .............................. 0.01
308
+ moe_capacity_monitor ............................ True
309
+ moe_deepep_num_sms .............................. 20
310
+ moe_echo_enable_random_offloading ............... False
311
+ moe_echo_expert_dispatch_overlap ................ False
312
+ moe_echo_expert_dispatcher_type ................. hybridep
313
+ moe_echo_recompute_expert_dispatch .............. False
314
+ moe_enable_deepep ............................... False
315
+ moe_enable_echo ................................. False
316
+ moe_enable_hybridep ............................. True
317
+ moe_enable_ultraep .............................. True
318
+ moe_expert_capacity_factor ...................... None
319
+ moe_extended_tp ................................. False
320
+ moe_ffn_hidden_size ............................. 1792
321
+ moe_grouped_gemm ................................ True
322
+ moe_hybridep_num_sms ............................ 16
323
+ moe_input_jitter_eps ............................ None
324
+ moe_layer_freq .................................. 1
325
+ moe_layer_recompute ............................. False
326
+ moe_num_echo_experts ............................ None
327
+ moe_num_redundant_experts_per_rank .............. 1
328
+ moe_pad_expert_input_to_capacity ................ False
329
+ moe_per_layer_logging ........................... False
330
+ moe_permute_fusion .............................. False
331
+ moe_received_token_capacity ..................... 1.072265625
332
+ moe_router_bias_update_rate ..................... 0.001
333
+ moe_router_dtype ................................ fp32
334
+ moe_router_enable_expert_bias ................... False
335
+ moe_router_force_load_balancing ................. False
336
+ moe_router_group_topk ........................... None
337
+ moe_router_load_balancing_type .................. aux_loss
338
+ moe_router_num_groups ........................... None
339
+ moe_router_padding_for_fp8 ...................... False
340
+ moe_router_pre_softmax .......................... False
341
+ moe_router_score_function ....................... softmax
342
+ moe_router_topk ................................. 16
343
+ moe_router_topk_scaling_factor .................. None
344
+ moe_shared_expert_intermediate_size ............. None
345
+ moe_shared_expert_overlap ....................... False
346
+ moe_token_dispatcher_type ....................... flex
347
+ moe_token_drop_policy ........................... probs
348
+ moe_ultraep_max_microbatches .................... None
349
+ moe_ultraep_overlap_weight_sync ................. True
350
+ moe_ultraep_weight_sync_join .................... before_dispatch
351
+ moe_upcycling_granularity ....................... 1
352
+ moe_use_device_initiated_grouped_gemm ........... True
353
+ moe_use_legacy_grouped_gemm ..................... False
354
+ moe_use_upcycling ............................... False
355
+ moe_z_loss_coeff ................................ None
356
+ mrope_section ................................... None
357
+ mscale .......................................... 1.0
358
+ mscale_all_dim .................................. 0.0
359
+ mtp_loss_scaling_factor ......................... 0.1
360
+ mtp_num_layers .................................. None
361
+ multi_latent_attention .......................... False
362
+ nccl_all_reduce_for_prefill ..................... False
363
+ nccl_communicator_config_path ................... None
364
+ nccl_ub ......................................... False
365
+ no_load_optim ................................... None
366
+ no_load_rng ..................................... None
367
+ no_persist_layer_norm ........................... False
368
+ no_rope_freq .................................... None
369
+ no_save_optim ................................... None
370
+ no_save_rng ..................................... None
371
+ non_persistent_ckpt_type ........................ None
372
+ non_persistent_global_ckpt_dir .................. None
373
+ non_persistent_local_ckpt_algo .................. fully_parallel
374
+ non_persistent_local_ckpt_dir ................... None
375
+ non_persistent_save_interval .................... None
376
+ norm_epsilon .................................... 1e-05
377
+ normalization ................................... RMSNorm
378
+ num_attention_heads ............................. 32
379
+ num_channels .................................... 3
380
+ num_classes ..................................... 1000
381
+ num_dataset_builder_threads ..................... 1
382
+ num_distributed_optimizer_instances ............. 1
383
+ num_experts ..................................... 64
384
+ num_layers ...................................... 16
385
+ num_layers_at_end_in_bf16 ....................... 1
386
+ num_layers_at_start_in_bf16 ..................... 1
387
+ num_layers_per_virtual_pipeline_stage ........... None
388
+ num_query_groups ................................ 8
389
+ num_virtual_stages_per_pipeline_rank ............ None
390
+ num_workers ..................................... 2
391
+ object_storage_cache_path ....................... None
392
+ one_logger_async ................................ False
393
+ one_logger_project .............................. megatron-lm
394
+ one_logger_run_name ............................. None
395
+ onnx_safe ....................................... None
396
+ openai_gelu ..................................... False
397
+ optimizer ....................................... adam
398
+ optimizer_cpu_offload ........................... False
399
+ optimizer_offload_fraction ...................... 1.0
400
+ output_bert_embeddings .......................... False
401
+ overlap_cpu_optimizer_d2h_h2d ................... False
402
+ overlap_grad_reduce ............................. True
403
+ overlap_moe_expert_parallel_comm ................ False
404
+ overlap_p2p_comm ................................ False
405
+ overlap_p2p_comm_warmup_flush ................... False
406
+ overlap_param_gather ............................ True
407
+ overlap_param_gather_with_optimizer_step ........ False
408
+ override_opt_param_scheduler .................... False
409
+ params_dtype .................................... torch.bfloat16
410
+ patch_dim ....................................... 16
411
+ per_split_data_args_path ........................ None
412
+ perform_initialization .......................... True
413
+ pin_cpu_grads ................................... True
414
+ pin_cpu_params .................................. True
415
+ pipeline_model_parallel_comm_backend ............ None
416
+ pipeline_model_parallel_layout .................. None
417
+ pipeline_model_parallel_size .................... 1
418
+ position_embedding_type ......................... rope
419
+ pretrained_checkpoint ........................... None
420
+ profile ......................................... False
421
+ profile_ranks ................................... [0]
422
+ profile_step_end ................................ 12
423
+ profile_step_start .............................. 10
424
+ q_lora_rank ..................................... None
425
+ qk_head_dim ..................................... 128
426
+ qk_l2_norm ...................................... False
427
+ qk_layernorm .................................... False
428
+ qk_pos_emb_head_dim ............................. 64
429
+ query_in_block_prob ............................. 0.1
430
+ rampup_batch_size ............................... None
431
+ rank ............................................ 0
432
+ recompute_granularity ........................... None
433
+ recompute_method ................................ None
434
+ recompute_modules ............................... None
435
+ recompute_num_layers ............................ None
436
+ record_memory_history ........................... False
437
+ relative_attention_max_distance ................. 128
438
+ relative_attention_num_buckets .................. 32
439
+ replication ..................................... False
440
+ replication_factor .............................. 2
441
+ replication_jump ................................ None
442
+ rerun_mode ...................................... validate_results
443
+ reset_attention_mask ............................ False
444
+ reset_position_ids .............................. False
445
+ result_rejected_tracker_filename ................ None
446
+ retriever_report_topk_accuracies ................ []
447
+ retriever_score_scaling ......................... False
448
+ retriever_seq_length ............................ 256
449
+ retro_add_retriever ............................. False
450
+ retro_attention_gate ............................ 1
451
+ retro_cyclic_train_iters ........................ None
452
+ retro_encoder_attention_dropout ................. 0.1
453
+ retro_encoder_hidden_dropout .................... 0.1
454
+ retro_encoder_layers ............................ 2
455
+ retro_num_neighbors ............................. 2
456
+ retro_num_retrieved_chunks ...................... 2
457
+ retro_project_dir ............................... None
458
+ retro_verify_neighbor_count ..................... True
459
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
460
+ rope_scaling_factor ............................. 8.0
461
+ rope_type ....................................... None
462
+ rotary_base ..................................... 1000000
463
+ rotary_interleaved .............................. False
464
+ rotary_percent .................................. 1.0
465
+ rotary_scaling_factor ........................... 1.0
466
+ rotary_seq_len_interpolation_factor ............. None
467
+ run_workload_inspector_server ................... False
468
+ sample_rate ..................................... 1.0
469
+ save ............................................ None
470
+ save_interval ................................... None
471
+ save_retain_interval ............................ None
472
+ scatter_gather_tensors_in_pipeline .............. True
473
+ seed ............................................ 1234
474
+ seq_length ...................................... 4096
475
+ sequence_parallel ............................... False
476
+ sft ............................................. False
477
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
478
+ sgd_momentum .................................... 0.9
479
+ short_seq_prob .................................. 0.1
480
+ skip_train ...................................... False
481
+ skipped_train_samples ........................... 0
482
+ spec ............................................ None
483
+ split ........................................... 100,0,0
484
+ squared_relu .................................... False
485
+ start_weight_decay .............................. 0.1
486
+ straggler_ctrlr_port ............................ 65535
487
+ straggler_minmax_count .......................... 1
488
+ suggested_communication_unit_size ............... None
489
+ swiglu .......................................... True
490
+ swin_backbone_type .............................. tiny
491
+ symmetric_ar_type ............................... None
492
+ te_rng_tracker .................................. True
493
+ tensor_model_parallel_size ...................... 1
494
+ tensorboard_dir ................................. None
495
+ tensorboard_log_interval ........................ 1
496
+ tensorboard_queue_size .......................... 1000
497
+ test_data_path .................................. None
498
+ test_mode ....................................... False
499
+ tiktoken_num_special_tokens ..................... 1000
500
+ tiktoken_pattern ................................ None
501
+ tiktoken_special_tokens ......................... None
502
+ timing_log_level ................................ 0
503
+ timing_log_option ............................... minmax
504
+ titles_data_path ................................ None
505
+ tokenizer_model ................................. None
506
+ tokenizer_type .................................. GPT2BPETokenizer
507
+ torch_fsdp2_reshard_after_forward ............... True
508
+ tp_comm_bootstrap_backend ....................... nccl
509
+ tp_comm_bulk_dgrad .............................. True
510
+ tp_comm_bulk_wgrad .............................. True
511
+ tp_comm_overlap ................................. False
512
+ tp_comm_overlap_ag .............................. True
513
+ tp_comm_overlap_cfg ............................. None
514
+ tp_comm_overlap_rs .............................. True
515
+ tp_comm_overlap_rs_dgrad ........................ False
516
+ tp_comm_split_ag ................................ True
517
+ tp_comm_split_rs ................................ True
518
+ train_data_path ................................. None
519
+ train_iters ..................................... 12
520
+ train_samples ................................... None
521
+ train_sync_interval ............................. None
522
+ transformer_impl ................................ transformer_engine
523
+ transformer_pipeline_model_parallel_size ........ 1
524
+ untie_embeddings_and_output_weights ............. True
525
+ use_checkpoint_args ............................. False
526
+ use_checkpoint_opt_param_scheduler .............. False
527
+ use_cpu_initialization .......................... None
528
+ use_custom_fsdp ................................. False
529
+ use_dist_ckpt ................................... True
530
+ use_dist_ckpt_deprecated ........................ False
531
+ use_distributed_optimizer ....................... True
532
+ use_flash_attn .................................. False
533
+ use_legacy_models ............................... False
534
+ use_mp_args_from_checkpoint_args ................ False
535
+ use_one_sent_docs ............................... False
536
+ use_persistent_ckpt_worker ...................... False
537
+ use_precision_aware_optimizer ................... False
538
+ use_pytorch_profiler ............................ False
539
+ use_ring_exchange_p2p ........................... False
540
+ use_rope_scaling ................................ False
541
+ use_rotary_position_embeddings .................. False
542
+ use_sharp ....................................... False
543
+ use_tokenizer_model_from_checkpoint_args ........ True
544
+ use_torch_fsdp2 ................................. False
545
+ use_torch_optimizer_for_cpu_offload ............. False
546
+ use_tp_pp_dp_mapping ............................ False
547
+ v_head_dim ...................................... 128
548
+ valid_data_path ................................. None
549
+ variable_seq_lengths ............................ False
550
+ virtual_pipeline_model_parallel_size ............ None
551
+ vision_backbone_type ............................ vit
552
+ vision_pretraining .............................. False
553
+ vision_pretraining_type ......................... classify
554
+ vocab_extra_ids ................................. 0
555
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
556
+ vocab_size ...................................... None
557
+ wandb_exp_name ..................................
558
+ wandb_project ...................................
559
+ wandb_save_dir ..................................
560
+ weight_decay .................................... 0.1
561
+ weight_decay_incr_style ......................... constant
562
+ wgrad_deferral_limit ............................ 0
563
+ world_size ...................................... 8
564
+ yaml_cfg ........................................ None
565
+ -------------------- end of arguments ---------------------
566
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
567
+ > building GPT2BPETokenizer tokenizer ...
568
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
569
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
570
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
571
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
572
+ > initializing torch distributed ...
573
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
574
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
575
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
576
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/results.tsv ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ config backend policy slots seq gbs cap recompute rep status iter_ms_mean iter_ms_min peak_mib run_dir
2
+ G1 ultraep one 1 4096 16 2.40625 none 1 OK 888.60 869.80 129475 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep1
3
+ G1 ultraep one 1 4096 16 2.40625 none 2 OK 891.61 869.80 129475 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep2
4
+ G1 ultraep one 1 4096 16 2.40625 none 3 OK 891.00 869.80 129475 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G1__ultraep__one__seq4096__gbs16__cap2p40625/rep3
5
+ G2 ultraep one 1 4096 16 1.08203125 none 1 OK 466.57 441.70 102733 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep1
6
+ G2 ultraep one 1 4096 16 1.08203125 none 2 OK 466.68 441.70 102733 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep2
7
+ G2 ultraep one 1 4096 16 1.08203125 none 3 OK 466.65 441.70 102733 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G2__ultraep__one__seq4096__gbs16__cap1p08203125/rep3
8
+ G3 ultraep one 1 4096 16 1.072265625 none 1 OK 448.07 438.50 102043 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep1
9
+ G3 ultraep one 1 4096 16 1.072265625 none 2 OK 448.14 435.80 102043 /workspace/tiera-rerun-fixed-moonep/ultraep/a1/a1_oneslot_gbs16/G3__ultraep__one__seq4096__gbs16__cap1p072265625/rep2
tiera-rerun-fixed-moonep/ultraep/a1/manifest.txt ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ timestamp: 2026-08-30T07:00:42+00:00
2
+ host: sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0
3
+ gpu: NVIDIA B200 x8
4
+ driver: 580.95.05
5
+ study_scripts_branch: granularity-ultraep
6
+ study_scripts_commit: 5e3993b
7
+ campaign: ultraep
8
+ backend: ultraep
9
+ phase: a1
10
+ gbs: 16
11
+ seq: 4096
12
+ configs: G1 G2 G3 G4
13
+ capacity: 2.40625
14
+ repeats: 3
15
+ nsys: 0
16
+ megatron_repo: /workspace/Megatron-UltraEP
17
+ megatron_branch_required: ultraep-bracing-regions-through-kernels
18
+ megatron_branch_actual: ultraep-bracing-regions-through-kernels
19
+ megatron_commit: 03d57ef7
tiera-rerun-fixed-moonep/ultraep/a1/pip_freeze.txt ADDED
@@ -0,0 +1,317 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ absl-py==2.3.0
2
+ aiohappyeyeballs @ file:///rapids/aiohappyeyeballs-2.6.1-py3-none-any.whl#sha256=f349ba8f4b75cb25c99c5c2d84e997e485204d2902a9597802b0371f09331fb8
3
+ aiohttp @ file:///rapids/aiohttp-3.12.7-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=94f98e0e5a49f89b252e115844f756c04fc8050f38252a32a3dd994ce8121f10
4
+ aiosignal @ file:///rapids/aiosignal-1.3.2-py2.py3-none-any.whl#sha256=45cde58e409a301715980c2b01d0c28bdde3770d8290b5eb2173759d9acb31a5
5
+ annotated-types==0.7.0
6
+ anyio==4.9.0
7
+ apex @ file:///opt/pytorch/apex/dist/apex-0.1-cp312-cp312-linux_x86_64.whl#sha256=a1508e4a386362e9c8f46c92ccada9110e421212b964c694235d0fd5a224b170
8
+ argon2-cffi==25.1.0
9
+ argon2-cffi-bindings==21.2.0
10
+ arrow==1.3.0
11
+ asciitree @ file:///rapids/asciitree-0.3.3-py3-none-any.whl#sha256=3e507ccfaeb417823158091fdb9d26b0205f6a59ba680b479a71db65c73d01d4
12
+ asttokens==3.0.0
13
+ astunparse==1.6.3
14
+ async-lru==2.0.5
15
+ attrs==25.3.0
16
+ audioread==3.0.1
17
+ babel==2.17.0
18
+ beautifulsoup4==4.13.4
19
+ black==25.1.0
20
+ bleach==6.2.0
21
+ blis==0.7.11
22
+ cachetools @ file:///rapids/cachetools-6.0.0-py3-none-any.whl#sha256=82e73ba88f7b30228b5507dce1a1f878498fc669d972aef2dde4f3a3c24f103e
23
+ catalogue==2.0.10
24
+ certifi==2025.4.26
25
+ cffi==1.17.1
26
+ charset-normalizer==3.4.2
27
+ click==8.5.0
28
+ cloudpathlib==0.21.1
29
+ cloudpickle @ file:///rapids/cloudpickle-3.1.1-py3-none-any.whl#sha256=c8c5a44295039331ee9dad40ba100a9c7297b6f988e50e87ccdf3765a668350e
30
+ cmake==3.31.6
31
+ comm==0.2.2
32
+ confection==0.1.5
33
+ contourpy==1.3.2
34
+ cuda-bindings @ file:///rapids/cuda_bindings-12.9.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=ff0e28d1e34758654b9c961e1f55e4786e49aee6a4dbceaf3cc24c46c672df7e
35
+ cuda-python @ file:///rapids/cuda_python-12.9.0-py3-none-any.whl#sha256=926acba49b2c0a0374c61b7c98f337c085199cf51cdfe4d6423c4129c20547a7
36
+ cudf @ file:///rapids/cudf-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=b779a8e8c0e8577b9b560895958496da211bfac49cdc4f11a49e8d18cb82e348
37
+ cudf-polars @ file:///rapids/cudf_polars-25.4.0-py3-none-any.whl#sha256=227b6c5147e93ea86282c8f7219118f916769fa2ca0cdd931e97a148aa66b869
38
+ cugraph @ file:///rapids/cugraph-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=b837c2dc2f7b235c702e4f8df64479b8fcbb342f848516d5faf0cdc4b5f31f29
39
+ cugraph-service-client @ file:///rapids/cugraph_service_client-25.4.0-py3-none-any.whl#sha256=7e568c3e41154b9752f5173555ffcc282ac52d72343feeb79653b2e2b1232470
40
+ cugraph-service-server @ file:///rapids/cugraph_service_server-25.4.0-py3-none-any.whl#sha256=a2830dcffafcccb16e12ebf85b98f8f247869d38c01ab4dd8664a1e7439d31be
41
+ cuml @ file:///rapids/cuml-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=1e90a464a18b094e24797f1676688c5d7696c4cf97c934504214a53d5aa09aa5
42
+ cupy-cuda12x @ file:///rapids/cupy_cuda12x-13.3.0-cp312-cp312-linux_x86_64.whl#sha256=6a2dfa0aa1e8189a10d1534276c08f1d2f73b747fd28bfbe7b9f4ef66d9af73f
43
+ cuvs @ file:///rapids/cuvs-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=b12f6b9f80e053a3b38542c2947d2516e468c02cad54a13d8b1a4a61d5aa7914
44
+ cycler==0.12.1
45
+ cymem==2.0.11
46
+ Cython==3.1.1
47
+ dask @ file:///rapids/dask-2025.2.0-py3-none-any.whl#sha256=f0fdeef6ceb0a06569d456c9e704f220f7f54e80f3a6ea42ab98cea6bc642b6e
48
+ dask-cuda @ file:///rapids/dask_cuda-25.4.0-py3-none-any.whl#sha256=442d87da0a95d00ca2a262b64c76e785a32ad905c82000fa2227932f81390e19
49
+ dask-cudf @ file:///rapids/dask_cudf-25.4.0-py3-none-any.whl#sha256=5e7cea0b26daea17da6e0bb457435e24460f76ba740358a3a18e239d6015c95c
50
+ datasets==5.0.1
51
+ debugpy==1.8.14
52
+ decorator==5.2.1
53
+ -e git+https://github.com/deepseek-ai/DeepEP.git@34152ae28f80bcc3ee38d7a12cb2ad87cfd4ea72#egg=deep_ep
54
+ defusedxml==0.7.1
55
+ dill==0.4.1
56
+ distributed @ file:///rapids/distributed-2025.2.0-py3-none-any.whl#sha256=368462084be47092d0a835f6c64c9b852da64dc88302b76d5f93bd5131228c67
57
+ distributed-ucxx @ file:///rapids/distributed_ucxx-0.43.0-py3-none-any.whl#sha256=94f60d0ccb0c94ba8dbfa1eaf455874d18aaab17e2ae218c38590e9fdc2d721d
58
+ dm-tree==0.1.9
59
+ einops==0.8.1
60
+ execnet==2.1.1
61
+ executing==2.2.0
62
+ expecttest==0.3.0
63
+ fasteners @ file:///rapids/fasteners-0.19-py3-none-any.whl#sha256=758819cb5d94cdedf4e836988b74de396ceacb8e2794d21f82d131fd9ee77237
64
+ fastjsonschema==2.21.1
65
+ fastrlock @ file:///rapids/fastrlock-0.8.3-cp312-cp312-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_28_x86_64.whl#sha256=5f13ec08f1adb1aa916c384b05ecb7dbebb8df9ea81abd045f60941c6283a670
66
+ filelock==3.18.0
67
+ flash_attn @ file:///opt/pytorch/flash_attn-2.7.4.post1-cp312-cp312-linux_x86_64.whl#sha256=a10455d6c56b387785b7413fd0bbcb392b42b3d4f4b88fda5d86cc8a9ce9182a
68
+ fonttools==4.58.1
69
+ fqdn==1.5.1
70
+ frozenlist @ file:///rapids/frozenlist-1.6.0-cp312-cp312-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=4e1be9111cb6756868ac242b3c2bd1f09d9aea09846e4f5c23715e7afb647103
71
+ fsspec==2025.5.1
72
+ gast==0.6.0
73
+ grpcio==1.62.1
74
+ h11==0.16.0
75
+ hf-xet==1.6.0
76
+ httpcore==1.0.9
77
+ httpx==0.28.1
78
+ huggingface_hub==1.29.0
79
+ hypothesis==6.130.8
80
+ idna==3.10
81
+ importlib_metadata==8.7.0
82
+ iniconfig==2.1.0
83
+ intel-openmp==2021.4.0
84
+ ipykernel==6.29.5
85
+ ipython==9.3.0
86
+ ipython_pygments_lexers==1.1.1
87
+ isoduration==20.11.0
88
+ isort==6.0.1
89
+ jedi==0.19.2
90
+ Jinja2==3.1.6
91
+ joblib==1.5.1
92
+ json5==0.12.0
93
+ jsonpointer==3.0.0
94
+ jsonschema==4.24.0
95
+ jsonschema-specifications==2025.4.1
96
+ jupyter-events==0.12.0
97
+ jupyter-lsp==2.2.5
98
+ jupyter_client==8.6.3
99
+ jupyter_core==5.8.1
100
+ jupyter_server==2.16.0
101
+ jupyter_server_terminals==0.5.3
102
+ jupyterlab==4.4.3
103
+ jupyterlab_code_formatter==3.0.2
104
+ jupyterlab_pygments==0.3.0
105
+ jupyterlab_server==2.27.3
106
+ jupyterlab_tensorboard_pro @ file:///builder/jupyterlab_tensorboard_pro-4.0.0-py2.py3-none-any.whl#sha256=6bfcb5ced3358eb72150f4b2ab91312faef2f0ca14d845e8379d8fa52df6902e
107
+ jupytext==1.17.2
108
+ kiwisolver==1.4.8
109
+ kvikio @ file:///rapids/kvikio-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=8b40fd8ea9e4a2974c04b50b7fc26b06094ac807f9a233281d0bafbaa71b1d85
110
+ langcodes==3.5.0
111
+ language_data==1.3.0
112
+ lazy_loader==0.4
113
+ libcudf @ file:///rapids/libcudf-25.4.0-py3-none-linux_x86_64.whl#sha256=238320b4958c1737bba2a609673e2539a78216b811b5498a68e00395d9b021a4
114
+ libcugraph @ file:///rapids/libcugraph-25.4.0-py3-none-linux_x86_64.whl#sha256=29085e901b2f7569ff2653bb1355fe6fad20f2fcaa585f5649da9867d8df766d
115
+ libcuml @ file:///rapids/libcuml-25.4.0-py3-none-linux_x86_64.whl#sha256=8e2e444c72b0e995e15a48da37ff12fb410a47a2baa69ba86ccc9b1a3db91e8c
116
+ libcuvs @ file:///rapids/libcuvs-25.4.0-py3-none-linux_x86_64.whl#sha256=e2f8db61c92b769c590dd752f43439ddd62b2f6592a3c337362b6c1ef220c3b9
117
+ libkvikio @ file:///rapids/libkvikio-25.4.0-py3-none-linux_x86_64.whl#sha256=da8a5713b05fc478bf40529a3e0fb61a2d605bb5a70ca4d461f96a6005dbe8b7
118
+ libraft @ file:///rapids/libraft-25.4.0-py3-none-linux_x86_64.whl#sha256=b921dec005aa4a321dd156776a18376feb47c96a5b67eec125c8a8e19edd1e24
119
+ librmm @ file:///rapids/librmm-25.4.0-py3-none-linux_x86_64.whl#sha256=ff951811bda729d4fa1109ab906225ae1f489a6ec2eeab5d1b674cc34bd4422f
120
+ librmm-cu12 @ file:///rapids/librmm_cu12-25.4.0-py3-none-any.whl#sha256=29296ef2b0e90679a0db280e86015be12d6ab9f4e62a181aa085f62dbadd06b3
121
+ librosa==0.11.0
122
+ libucxx @ file:///rapids/libucxx-0.43.0-py3-none-linux_x86_64.whl#sha256=f15849d7ac2fb68f0fd8998838a13321ddcfb0c594f7e6cbc401a3bcc4750566
123
+ lightning-thunder @ file:///opt/pytorch/lightning-thunder
124
+ lightning-utilities==0.14.3
125
+ lintrunner==0.12.7
126
+ llvmlite==0.42.0
127
+ locket @ file:///rapids/locket-1.0.0-py2.py3-none-any.whl#sha256=b6c819a722f7b6bd955b80781788e4a66a55628b858d347536b7e81325a3a5e3
128
+ looseversion==1.3.0
129
+ marisa-trie==1.2.1
130
+ Markdown==3.8
131
+ markdown-it-py==3.0.0
132
+ MarkupSafe==3.0.2
133
+ matplotlib==3.10.3
134
+ matplotlib-inline==0.1.7
135
+ mdit-py-plugins==0.4.2
136
+ mdurl==0.1.2
137
+ -e git+ssh://git@github.com/sriram-17-17/my-Megatron-MoE-Echo.git@9a23679390d7ab1fe462bc61fa31d90d393db49c#egg=megatron_core
138
+ mistune==3.1.3
139
+ mkl==2021.1.1
140
+ mkl-devel==2021.1.1
141
+ mkl-include==2021.1.1
142
+ ml_dtypes==0.6.0
143
+ mock==5.2.0
144
+ -e git+ssh://git@github.com/sriram-17-17/MoonEP.git@d8bc6667fad7177e6e09b637d773fa67f08bae93#egg=moonep
145
+ mpmath==1.3.0
146
+ msgpack==1.1.0
147
+ multidict @ file:///rapids/multidict-6.4.4-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=a920f9cf2abdf6e493c519492d892c362007f113c94da4c239ae88429835bad1
148
+ multiprocess==0.70.19
149
+ murmurhash==1.0.13
150
+ mypy_extensions==1.1.0
151
+ nbclient==0.10.2
152
+ nbconvert==7.16.6
153
+ nbformat==5.10.4
154
+ nest-asyncio==1.6.0
155
+ networkx==3.5
156
+ ninja==1.11.1.4
157
+ notebook==7.4.3
158
+ notebook_shim==0.2.4
159
+ numba==0.59.1
160
+ numba-cuda @ file:///rapids/numba_cuda-0.4.0-py3-none-any.whl#sha256=5b95a5f42068771e37142944d44fe844fa87f1b2f89051461a4cb0af91bcc28a
161
+ numcodecs @ file:///rapids/numcodecs-0.13.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=da2230484e6102e5fa3cc1a5dd37ca1f92dfbd183d91662074d6f7574e3e8f53
162
+ numpy==1.26.4
163
+ nvdlfw_inspect @ git+https://github.com/NVIDIA/nvidia-dlfw-inspect.git@6b60eb0e675606fb2fbbfcb12667ecbec75eaf87
164
+ nvfuser==0.2.27a0+9bf5aca
165
+ nvidia-cuda-cccl-cu12==12.9.27
166
+ nvidia-cudnn-frontend @ file:///opt/pytorch/lightning-thunder/tmp/cudnn_frontend
167
+ nvidia-cutlass-dsl==4.4.2
168
+ nvidia-cutlass-dsl-libs-base==4.4.2
169
+ nvidia-dali-cuda120==1.50.0
170
+ nvidia-ml-py @ file:///rapids/nvidia_ml_py-12.575.51-py3-none-any.whl#sha256=eb8641800d98ce40a22f479873f34b482e214a7e80349c63be51c3919845446e
171
+ nvidia-modelopt==0.29.0
172
+ nvidia-modelopt-core==0.29.0
173
+ nvidia-nvcomp-cu12==4.2.0.14
174
+ nvidia-nvimgcodec-cu12==0.5.0.13
175
+ nvidia-nvjpeg-cu12==12.4.0.16
176
+ nvidia-nvjpeg2k-cu12==0.8.1.40
177
+ nvidia-nvshmem-cu12==3.7.2
178
+ nvidia-nvtiff-cu12==0.5.0.67
179
+ nvidia-resiliency-ext==0.4.0
180
+ nvtx @ file:///rapids/nvtx-0.2.11-cp312-cp312-linux_x86_64.whl#sha256=73521e37cba42b4db0b3cd68fc5eb814e55c500a61eeed6edcdf6746315ac53e
181
+ nx-cugraph @ file:///rapids/nx_cugraph-25.4.0-py3-none-any.whl#sha256=e625444486cc899a77bd48e7da76a2edd95c0006dd95fbf323745feae7f33111
182
+ onnx @ file:///opt/pytorch/pytorch/third_party/onnx
183
+ onnx-ir==1.0.0
184
+ onnxscript==0.7.1
185
+ opt_einsum==3.4.0
186
+ optree==0.16.0
187
+ overrides==7.7.0
188
+ packaging==23.2
189
+ pandas @ file:///rapids/pandas-2.2.3-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=fffb8ae78d8af97f849404f21411c95062db1496aeb3e56f146f0355c9989319
190
+ pandocfilters==1.5.1
191
+ parso==0.8.4
192
+ partd @ file:///rapids/partd-1.4.2-py3-none-any.whl#sha256=978e4ac767ec4ba5b86c6eaa52e5a2a3bc748a2ca839e8cc798f1cc6ce6efb0f
193
+ pathspec==0.12.1
194
+ pexpect==4.9.0
195
+ pillow==11.2.1
196
+ platformdirs==4.3.8
197
+ pluggy==1.6.0
198
+ ply @ file:///rapids/ply-3.11-py2.py3-none-any.whl#sha256=096f9b8350b65ebd2fd1346b12452efe5b9607f7482813ffca50c22722a807ce
199
+ polars @ file:///rapids/polars-1.25.2-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=f7fcbb4f476784384ccda48757fca4e8c2e2c5a0a3aef3717aaf56aee4e30e09
200
+ polygraphy==0.49.20
201
+ pooch==1.8.2
202
+ preshed==3.0.10
203
+ prometheus_client==0.22.1
204
+ prompt_toolkit==3.0.51
205
+ propcache @ file:///rapids/propcache-0.3.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=563f9d8c03ad645597b8d010ef4e9eab359faeb11a0a2ac9f7b4bc8c28ebef25
206
+ protobuf==4.24.4
207
+ psutil==7.0.0
208
+ ptyprocess==0.7.0
209
+ PuLP==3.2.1
210
+ pure_eval==0.2.3
211
+ pyarrow==25.0.1
212
+ pybind11==2.13.6
213
+ pybind11_global==2.13.6
214
+ pycocotools @ git+https://github.com/nvidia/cocoapi.git@08350a157b41f0755a536441b5abec89ab934c5b#subdirectory=PythonAPI
215
+ pycparser==2.22
216
+ pydantic==2.11.5
217
+ pydantic_core==2.33.2
218
+ Pygments==2.19.1
219
+ pylibcudf @ file:///rapids/pylibcudf-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=401b8bfd022d06f84e120779ca7c3d01a28b68762aaa8f7873b10d6b1975cf6a
220
+ pylibcugraph @ file:///rapids/pylibcugraph-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=061c7389a88c815b54fd5d3a3584fb87bf693e5602f5be335260d0ba575ba557
221
+ pylibcugraphops @ file:///rapids/pylibcugraphops-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=a25ef4b22166c80718d5c7e5bddb755e81a2fe340fa5becd0ac36ec3edfb208a
222
+ pylibraft @ file:///rapids/pylibraft-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=f2ecd23fe51f9c2caa87bbbe85cb32fcaff4ecfb52ceb9a9991bc50f82782c2e
223
+ pylibwholegraph @ file:///rapids/pylibwholegraph-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=b82913d1e38bda76d8c272fbd8cc9072d987209d3d47cc93a464a37c2925aca2
224
+ pynvjitlink @ file:///rapids/pynvjitlink-0.3.0-cp312-cp312-linux_x86_64.whl#sha256=066caf1314730643d3b48c9847bd65789defdd6a7c78993176eb3d8bbe3f4919
225
+ pynvml @ file:///rapids/pynvml-12.0.0-py3-none-any.whl#sha256=fdff84b62a27dbe98e08e1a647eb77342bef1aebe0878bcd15e99a83fcbecb9e
226
+ pyparsing==3.2.3
227
+ pytest==8.1.1
228
+ pytest-flakefinder==1.1.0
229
+ pytest-rerunfailures==15.1
230
+ pytest-shard==0.1.2
231
+ pytest-xdist==3.7.0
232
+ python-dateutil==2.9.0.post0
233
+ python-hostlist==2.2.1
234
+ python-json-logger==3.3.0
235
+ pytorch-triton @ file:///tmp/dist/pytorch_triton-3.3.0%2Bgit96316ce52.nvinternal-cp312-cp312-linux_x86_64.whl#sha256=6d6921772b9dfd0243c055e0d054de42a113e92cabcfdd1d3e80a5ad9c08cf7d
236
+ pytz @ file:///rapids/pytz-2023.4-py2.py3-none-any.whl#sha256=f90ef520d95e7c46951105338d918664ebfd6f1d995bd7d153127ce90efafa6a
237
+ PyYAML==6.0.2
238
+ pyzmq==26.4.0
239
+ raft-dask @ file:///rapids/raft_dask-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=36cf37ba6ce35454c4729c2cb597a360f22780eefb11d514a16a07137da5eb6c
240
+ rapids-dask-dependency @ file:///rapids/rapids_dask_dependency-25.4.0a0-py3-none-any.whl#sha256=484faca626e7760112a7d8ad3426f5bdd720b609006d8dca87c2607076a348bd
241
+ rapids-logger @ file:///rapids/rapids_logger-0.1.18-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=3617b2d86291ac6cb957542895f20fc9737ace29773ca90337e6fb04c73bd120
242
+ referencing==0.36.2
243
+ regex==2024.11.6
244
+ requests==2.32.3
245
+ rfc3339-validator==0.1.4
246
+ rfc3986-validator==0.1.1
247
+ rich==14.0.0
248
+ rmm @ file:///rapids/rmm-25.4.0-cp312-cp312-linux_x86_64.whl#sha256=1824bcab3836296666980a997ddcb9c4bca24dfb873012cfa0a7ecea583ca712
249
+ rpds-py==0.25.1
250
+ safetensors==0.5.3
251
+ scikit-learn==1.6.1
252
+ scipy==1.15.3
253
+ Send2Trash==1.8.3
254
+ setuptools==78.1.1
255
+ shellingham==1.5.4
256
+ six==1.16.0
257
+ smart-open==7.1.0
258
+ sniffio==1.3.1
259
+ sortedcontainers==2.4.0
260
+ soundfile==0.13.1
261
+ soupsieve==2.7
262
+ soxr==0.5.0.post1
263
+ spacy==3.7.5
264
+ spacy-legacy==3.0.12
265
+ spacy-loggers==1.0.5
266
+ srsly==2.5.1
267
+ stack-data==0.6.3
268
+ sympy==1.14.0
269
+ tabulate==0.9.0
270
+ tbb==2021.13.1
271
+ tblib @ file:///rapids/tblib-3.1.0-py3-none-any.whl#sha256=670bb4582578134b3d81a84afa1b016128b429f3d48e6cbbaecc9d15675e984e
272
+ tensorboard==2.16.2
273
+ tensorboard-data-server==0.7.2
274
+ tensorrt @ file:///workspace/TensorRT-10.11.0.33/python/tensorrt-10.11.0.33-cp312-none-linux_x86_64.whl#sha256=c2f18f72758a3d54a71901495c8fda9e7094a8bcdd2a5b6b58967fbf05733b3a
275
+ terminado==0.18.1
276
+ thinc==8.2.5
277
+ threadpoolctl==3.6.0
278
+ thriftpy2 @ file:///rapids/thriftpy2-0.5.2-cp312-cp312-linux_x86_64.whl#sha256=c0c7d939bdc5ccf7739154959938445d5fdf72f1d6e12bd563947a36f67e2147
279
+ tinycss2==1.4.0
280
+ toolz @ file:///rapids/toolz-1.0.0-py3-none-any.whl#sha256=292c8f1c4e7516bf9086f8850935c799a874039c8bcf959d47b600e4c44a6236
281
+ torch @ file:///opt/transfer/torch-2.8.0a0%2B5228986c39.nv25.6-29946474-cp312-cp312-linux_x86_64.whl#sha256=7a3c9ff5288d8102a31296b3d62491f3f47c8820a5c9aaefc7dc0fe20e1e8ad5
282
+ torch_tensorrt @ file:///opt/pytorch/torch_tensorrt/dist/torch_tensorrt-2.8.0a0-cp312-cp312-linux_x86_64.whl#sha256=26f4d7a04e7e6e017a5b917b8b98b83720bb2bb0195f48d25f52d567d286bfe1
283
+ torchao @ file:///opt/pytorch/ao
284
+ torchprofile==0.0.4
285
+ torchvision @ file:///opt/transfer/torchvision-0.22.0a0%2B95f10a4e-29946474-cp312-cp312-manylinux_2_28_x86_64.whl#sha256=746a3c674cc61680a86f4a14b374b21af1b5c21f629c42765b40c9494f6b234d
286
+ tornado==6.5.1
287
+ tqdm==4.67.1
288
+ traitlets==5.14.3
289
+ transformer_engine @ file:///pvcdata/aiinfra/b-aghosh/wheels/transformer_engine-2.10.0.dev0%2B10f6128d-cp312-cp312-linux_x86_64.whl#sha256=35cf5f4b83b2baad864e12e512bb422e0929409ac05610812fa521d4942d0189
290
+ treelite @ file:///rapids/treelite-4.4.1-py3-none-linux_x86_64.whl#sha256=174741724205540a1a5b37da81483aeff55d89548f42b3461cf07acac7ae96eb
291
+ typer==0.16.0
292
+ types-dataclasses==0.6.6
293
+ types-python-dateutil==2.9.0.20250516
294
+ typing-inspection==0.4.1
295
+ typing_extensions==4.14.0
296
+ tzdata @ file:///rapids/tzdata-2025.2-py2.py3-none-any.whl#sha256=1a403fada01ff9221ca8044d701868fa132215d84beb92242d9acd2147f667a8
297
+ ucx-py @ file:///rapids/ucx_py-0.43.0-cp312-cp312-linux_x86_64.whl#sha256=4a452fd8d8c983283ada3d49086e3657dc2ee94fbad08efdd708d15ab6e0b73a
298
+ ucxx @ file:///rapids/ucxx-0.43.0-cp312-cp312-linux_x86_64.whl#sha256=f1b4e55cb45c66fe88a2d8f9c2b87751165f3cfec21fbdc9c47c4ddff75e8a5f
299
+ -e git+https://github.com/Dots-Infra/UltraEP.git@94cab099b44fffa99a82fea99e7c12d89cf65e4f#egg=ultra_ep
300
+ uri-template==1.3.0
301
+ urllib3==2.0.7
302
+ wasabi==1.1.3
303
+ wcwidth==0.2.13
304
+ weasel==0.4.1
305
+ webcolors==24.11.1
306
+ webencodings==0.5.1
307
+ websocket-client==1.8.0
308
+ Werkzeug==3.1.3
309
+ wheel==0.45.1
310
+ wrapt==1.17.2
311
+ xdoctest==1.0.2
312
+ xgboost @ file:///rapids/xgboost-2.1.4-py3-none-manylinux_2_39_x86_64.whl#sha256=25eeca27f4048fc491940b3a0b4b5782ee48c33cb3aec059cef71f8da542930b
313
+ xxhash==4.0.1
314
+ yarl @ file:///rapids/yarl-1.20.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl#sha256=bf099e2432131093cc611623e0b0bcc399b8cddd9a91eded8bfb50402ec35018
315
+ zarr @ file:///rapids/zarr-2.18.7-py3-none-any.whl#sha256=ac3dc4033e9ae4e9d7b5e27c97ea3eaf1003cc0a07f010bd83d5134bf8c4b223
316
+ zict @ file:///rapids/zict-3.0.0-py2.py3-none-any.whl#sha256=5796e36bd0e0cc8cf0fbc1ace6a68912611c1dbd74750a3f3026b9b9d6a327ae
317
+ zipp==3.22.0