AbhishekGhosh1998 commited on
Commit
cfc1548
·
verified ·
1 Parent(s): ab46f35

granularity results tiera-rerun-fixed-moonep @ 2026-08-30T07:40:52Z from sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0

Browse files
Files changed (18) hide show
  1. tiera-rerun-fixed-moonep/.completed/ultraep__a2__gbs16 +1 -0
  2. tiera-rerun-fixed-moonep/STATUS.txt +3 -2
  3. tiera-rerun-fixed-moonep/hf_watcher.log +3 -0
  4. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1/mem_samples.csv +216 -0
  5. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1/run.env +73 -0
  6. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1/train.log +917 -0
  7. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2/mem_samples.csv +216 -0
  8. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2/run.env +73 -0
  9. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2/train.log +915 -0
  10. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep3/mem_samples.csv +176 -0
  11. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep3/run.env +73 -0
  12. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep3/train.log +867 -0
  13. tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/results.tsv +3 -0
  14. tiera-rerun-fixed-moonep/moonep/a1/manifest.txt +19 -0
  15. tiera-rerun-fixed-moonep/moonep/a1/pip_freeze.txt +317 -0
  16. tiera-rerun-fixed-moonep/moonep/a1/provenance.txt +90 -0
  17. tiera-rerun-fixed-moonep/steps.tsv +1 -0
  18. tiera-rerun-fixed-moonep/study.log +64 -0
tiera-rerun-fixed-moonep/.completed/ultraep__a2__gbs16 ADDED
@@ -0,0 +1 @@
 
 
1
+ 2026-08-30T07:37:07+00:00
tiera-rerun-fixed-moonep/STATUS.txt CHANGED
@@ -1,13 +1,14 @@
1
  study_root: /workspace/tiera-rerun-fixed-moonep
2
  gbs: 16
3
- updated: 2026-08-30T07:22:03+00:00
4
- current: ultraep__a2__gbs16
5
 
6
  completed steps:
7
  echo__a1__gbs16
8
  echo__a2__gbs16
9
  echo__ponly__gbs16
10
  ultraep__a1__gbs16
 
11
  ultraep__ponly__gbs16
12
 
13
  failed steps:
 
1
  study_root: /workspace/tiera-rerun-fixed-moonep
2
  gbs: 16
3
+ updated: 2026-08-30T07:37:09+00:00
4
+ current: moonep__a1__gbs16
5
 
6
  completed steps:
7
  echo__a1__gbs16
8
  echo__a2__gbs16
9
  echo__ponly__gbs16
10
  ultraep__a1__gbs16
11
+ ultraep__a2__gbs16
12
  ultraep__ponly__gbs16
13
 
14
  failed steps:
tiera-rerun-fixed-moonep/hf_watcher.log CHANGED
@@ -44,3 +44,6 @@ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
44
  uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
45
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
46
  [2026-08-30T07:30:49+00:00] backup pass
 
 
 
 
44
  uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
45
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
46
  [2026-08-30T07:30:49+00:00] backup pass
47
+ uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
48
+ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
49
+ [2026-08-30T07:40:52+00:00] backup pass
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1/mem_samples.csv ADDED
@@ -0,0 +1,216 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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, 4
10
+ 1, 4
11
+ 2, 1
12
+ 3, 1
13
+ 4, 1
14
+ 5, 1
15
+ 6, 1
16
+ 7, 1
17
+ 0, 4
18
+ 1, 4
19
+ 2, 4
20
+ 3, 4
21
+ 4, 4
22
+ 5, 4
23
+ 6, 4
24
+ 7, 1
25
+ 0, 1
26
+ 1, 1
27
+ 2, 1
28
+ 3, 1
29
+ 4, 1
30
+ 5, 4
31
+ 6, 4
32
+ 7, 4
33
+ 0, 0
34
+ 1, 0
35
+ 2, 0
36
+ 3, 0
37
+ 4, 0
38
+ 5, 0
39
+ 6, 0
40
+ 7, 0
41
+ 0, 1
42
+ 1, 1
43
+ 2, 1
44
+ 3, 1
45
+ 4, 1
46
+ 5, 1
47
+ 6, 1
48
+ 7, 1
49
+ 0, 4
50
+ 1, 4
51
+ 2, 4
52
+ 3, 4
53
+ 4, 1
54
+ 5, 1
55
+ 6, 1
56
+ 7, 1
57
+ 0, 4
58
+ 1, 4
59
+ 2, 4
60
+ 3, 4
61
+ 4, 4
62
+ 5, 4
63
+ 6, 4
64
+ 7, 4
65
+ 0, 4
66
+ 1, 629
67
+ 2, 629
68
+ 3, 629
69
+ 4, 629
70
+ 5, 629
71
+ 6, 629
72
+ 7, 629
73
+ 0, 917
74
+ 1, 907
75
+ 2, 891
76
+ 3, 923
77
+ 4, 935
78
+ 5, 931
79
+ 6, 921
80
+ 7, 911
81
+ 0, 1895
82
+ 1, 1895
83
+ 2, 1895
84
+ 3, 1895
85
+ 4, 1895
86
+ 5, 1895
87
+ 6, 1895
88
+ 7, 1895
89
+ 0, 37231
90
+ 1, 37247
91
+ 2, 37247
92
+ 3, 37253
93
+ 4, 37275
94
+ 5, 37245
95
+ 6, 37237
96
+ 7, 37253
97
+ 0, 46021
98
+ 1, 46011
99
+ 2, 46007
100
+ 3, 46039
101
+ 4, 46053
102
+ 5, 46043
103
+ 6, 46053
104
+ 7, 46061
105
+ 0, 46423
106
+ 1, 46409
107
+ 2, 46447
108
+ 3, 46421
109
+ 4, 46457
110
+ 5, 46427
111
+ 6, 46405
112
+ 7, 46473
113
+ 0, 46921
114
+ 1, 46917
115
+ 2, 46917
116
+ 3, 46917
117
+ 4, 46917
118
+ 5, 46919
119
+ 6, 46919
120
+ 7, 46919
121
+ 0, 47383
122
+ 1, 47381
123
+ 2, 47381
124
+ 3, 47381
125
+ 4, 47381
126
+ 5, 47381
127
+ 6, 47381
128
+ 7, 47381
129
+ 0, 65793
130
+ 1, 65791
131
+ 2, 65791
132
+ 3, 65791
133
+ 4, 65791
134
+ 5, 65791
135
+ 6, 65791
136
+ 7, 65791
137
+ 0, 66255
138
+ 1, 66253
139
+ 2, 66253
140
+ 3, 66253
141
+ 4, 66253
142
+ 5, 66253
143
+ 6, 66253
144
+ 7, 66253
145
+ 0, 67263
146
+ 1, 67285
147
+ 2, 67275
148
+ 3, 67281
149
+ 4, 67293
150
+ 5, 67301
151
+ 6, 67291
152
+ 7, 67297
153
+ 0, 68351
154
+ 1, 68343
155
+ 2, 68351
156
+ 3, 68349
157
+ 4, 68351
158
+ 5, 68351
159
+ 6, 68351
160
+ 7, 68349
161
+ 0, 82181
162
+ 1, 82149
163
+ 2, 82211
164
+ 3, 82165
165
+ 4, 82177
166
+ 5, 82221
167
+ 6, 82209
168
+ 7, 82237
169
+ 0, 96849
170
+ 1, 96843
171
+ 2, 96843
172
+ 3, 96843
173
+ 4, 96843
174
+ 5, 96843
175
+ 6, 96843
176
+ 7, 96843
177
+ 0, 96849
178
+ 1, 96843
179
+ 2, 96843
180
+ 3, 96843
181
+ 4, 96843
182
+ 5, 96843
183
+ 6, 96843
184
+ 7, 96843
185
+ 0, 96849
186
+ 1, 96843
187
+ 2, 96843
188
+ 3, 96843
189
+ 4, 96843
190
+ 5, 96843
191
+ 6, 96843
192
+ 7, 96843
193
+ 0, 96203
194
+ 1, 95903
195
+ 2, 95653
196
+ 3, 96197
197
+ 4, 95909
198
+ 5, 95675
199
+ 6, 96197
200
+ 7, 96139
201
+ 0, 94455
202
+ 1, 94449
203
+ 2, 94449
204
+ 3, 94449
205
+ 4, 94449
206
+ 5, 94449
207
+ 6, 94449
208
+ 7, 94061
209
+ 0, 4
210
+ 1, 4
211
+ 2, 4
212
+ 3, 4
213
+ 4, 91759
214
+ 5, 4
215
+ 6, 4
216
+ 7, 4
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1/run.env ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=moonep
2
+ config=G1
3
+ family=tierA
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=moonep_backend_async_grad_reduce
6
+ commit=2e68c8ad
7
+ moonep_source_dir=/workspace/MoonEP
8
+ moonep_source_branch=megatron-ddp-symmetric-params
9
+ moonep_source_commit=d8bc666
10
+ num_layers=16
11
+ hidden=4096
12
+ ffn_hidden=14336
13
+ moe_ffn_hidden=14336
14
+ num_experts=8
15
+ topk=2
16
+ experts_per_rank=1
17
+ expert_params=176160768
18
+ heads=32
19
+ kv_groups=8
20
+ kv_channels=
21
+ qk_layernorm=0
22
+ shared_ffn=0
23
+ shared_expert_overlap=0
24
+ moe_token_dispatcher=flex
25
+ mla=0
26
+ dense_first_layers=0
27
+ moe_layers=16
28
+ rotary_base=1000000
29
+ rotary_scaling=
30
+ qkv_bias=0
31
+ router_balancing=aux_loss
32
+ aux_loss_coeff=1e-2
33
+ router_pre_softmax=0
34
+ router_fusion=0
35
+ router_dtype=fp32
36
+ device_initiated_grouped_gemm=1
37
+ norm_epsilon=1e-5
38
+ seq_length=4096
39
+ micro_batch_size=1
40
+ global_batch_size=16
41
+ num_microbatches=2
42
+ assignments_per_rank=8192
43
+ mean_tokens_per_expert=8192
44
+ slot_policy=one
45
+ slots_per_rank=1
46
+ capacity_mode=none
47
+ capacity=none
48
+ cuda_graph=1
49
+ probe_mode=0
50
+ async_grad_reduce=1
51
+ branch_role=timing
52
+ memory_dump_mode=none
53
+ memory_snapshot_path=
54
+ cuda_graph_warmup_steps=3
55
+ recompute=none
56
+ force_balance=0
57
+ train_iters=12
58
+ ep_size=8
59
+ echo_dispatch_overlap=1
60
+ echo_dispatch_recompute=0
61
+ ultraep_quota_min_tokens=
62
+ ultraep_weight_sync_join=
63
+ profile_enable=0
64
+ kernel_trace=0
65
+ fp8_mode=mxfp8
66
+ fp8_recipe=mxfp8
67
+ training_seed=1234
68
+ learning_rate=1e-4
69
+ min_learning_rate=1.0e-5
70
+ lr_warmup_iters=5
71
+ state_dump_enable=0
72
+ state_dump_dir=
73
+ moonep_assert_no_pending_reduce=1
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1/train.log ADDED
@@ -0,0 +1,917 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G1 / moonep (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=none cap=none cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 2e68c8ad
8
+ ==============================================================
9
+ W0830 07:37:34.765000 784984 torch/distributed/run.py:766]
10
+ W0830 07:37:34.765000 784984 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:37:34.765000 784984 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 07:37:34.765000 784984 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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
+ 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
20
+ /workspace/Megatron-MoE-Echo/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
+ /workspace/Megatron-MoE-Echo/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-MoE-Echo/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
+ 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
27
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
28
+ Number of virtual stages per pipeline stage: None
29
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
30
+ using torch.bfloat16 for parameters ...
31
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
32
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
33
+ ------------------------ arguments ------------------------
34
+ account_for_embedding_in_pipeline_split ......... False
35
+ account_for_loss_in_pipeline_split .............. False
36
+ accumulate_allreduce_grads_in_fp32 .............. True
37
+ adam_beta1 ...................................... 0.9
38
+ adam_beta2 ...................................... 0.999
39
+ adam_eps ........................................ 1e-08
40
+ add_bias_linear ................................. False
41
+ add_position_embedding .......................... False
42
+ add_qkv_bias .................................... False
43
+ adlr_autoresume ................................. False
44
+ adlr_autoresume_interval ........................ 1000
45
+ align_grad_reduce ............................... True
46
+ align_param_gather .............................. False
47
+ app_tag_run_name ................................ None
48
+ app_tag_run_version ............................. 0.0.0
49
+ apply_layernorm_1p .............................. False
50
+ apply_query_key_layer_scaling ................... False
51
+ apply_residual_connection_post_layernorm ........ False
52
+ apply_rope_fusion ............................... True
53
+ async_save ...................................... None
54
+ async_tensor_model_parallel_allreduce ........... True
55
+ attention_backend ............................... AttnBackend.auto
56
+ attention_dropout ............................... 0.0
57
+ attention_softmax_in_fp32 ....................... False
58
+ auto_detect_ckpt_format ......................... False
59
+ barrier_with_L1_time ............................ True
60
+ bert_binary_head ................................ True
61
+ bert_embedder_type .............................. megatron
62
+ bert_load ....................................... None
63
+ bf16 ............................................ True
64
+ bias_dropout_fusion ............................. True
65
+ bias_gelu_fusion ................................ False
66
+ bias_swiglu_fusion .............................. True
67
+ biencoder_projection_dim ........................ 0
68
+ biencoder_shared_query_context_model ............ False
69
+ block_data_path ................................. None
70
+ calc_ft_timeouts ................................ False
71
+ calculate_per_token_loss ........................ False
72
+ check_for_large_grads ........................... False
73
+ check_for_nan_in_loss_and_grad .................. False
74
+ check_for_spiky_loss ............................ False
75
+ check_weight_hash_across_dp_replicas_interval ... None
76
+ ckpt_assume_constant_structure .................. False
77
+ ckpt_convert_format ............................. None
78
+ ckpt_convert_save ............................... None
79
+ ckpt_convert_update_legacy_dist_opt_format ...... False
80
+ ckpt_format ..................................... torch_dist
81
+ ckpt_fully_parallel_load ........................ False
82
+ ckpt_fully_parallel_save ........................ True
83
+ ckpt_fully_parallel_save_deprecated ............. False
84
+ ckpt_step ....................................... None
85
+ classes_fraction ................................ 1.0
86
+ clip_grad ....................................... 1.0
87
+ clone_scatter_output_in_embedding ............... True
88
+ config_logger_dir ...............................
89
+ consumed_train_samples .......................... 0
90
+ consumed_valid_samples .......................... 0
91
+ context_parallel_size ........................... 1
92
+ cp_comm_type .................................... ['p2p']
93
+ create_attention_mask_in_dataloader ............. True
94
+ cross_entropy_fusion_impl ....................... native
95
+ cross_entropy_loss_fusion ....................... False
96
+ cuda_graph_scope ................................ full_iteration
97
+ cuda_graph_warmup_steps ......................... 3
98
+ data_args_path .................................. None
99
+ data_cache_path ................................. None
100
+ data_parallel_random_init ....................... False
101
+ data_parallel_sharding_strategy ................. no_shard
102
+ data_parallel_size .............................. 8
103
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
104
+ data_per_class_fraction ......................... 1.0
105
+ data_sharding ................................... True
106
+ dataloader_type ................................. single
107
+ ddp_average_in_collective ....................... False
108
+ ddp_bucket_size ................................. None
109
+ ddp_num_buckets ................................. None
110
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
111
+ decoder_first_pipeline_num_layers ............... None
112
+ decoder_last_pipeline_num_layers ................ None
113
+ decoder_num_layers .............................. None
114
+ decoder_seq_length .............................. None
115
+ decoupled_lr .................................... None
116
+ decoupled_min_lr ................................ None
117
+ decrease_batch_size_if_needed ................... False
118
+ defer_embedding_wgrad_compute ................... False
119
+ delay_wgrad_compute ............................. False
120
+ deprecated_use_mcore_models ..................... True
121
+ deterministic_mode .............................. False
122
+ dino_bottleneck_size ............................ 256
123
+ dino_freeze_last_layer .......................... 1
124
+ dino_head_hidden_size ........................... 2048
125
+ dino_local_crops_number ......................... 10
126
+ dino_local_img_size ............................. 96
127
+ dino_norm_last_layer ............................ False
128
+ dino_teacher_temp ............................... 0.07
129
+ dino_warmup_teacher_temp ........................ 0.04
130
+ dino_warmup_teacher_temp_epochs ................. 30
131
+ disable_bf16_reduced_precision_matmul ........... False
132
+ disable_mamba_mem_eff_path ...................... False
133
+ disable_straggler_on_startup .................... False
134
+ dist_ckpt_format_deprecated ..................... None
135
+ dist_ckpt_strictness ............................ assume_ok_unexpected
136
+ distribute_saved_activations .................... False
137
+ distributed_backend ............................. nccl
138
+ distributed_timeout_minutes ..................... 10
139
+ embedding_init_method_std ....................... None
140
+ embedding_path .................................. None
141
+ empty_unused_memory_level ....................... 0
142
+ enable_cuda_graph ............................... True
143
+ enable_experimental ............................. False
144
+ enable_ft_package ............................... False
145
+ enable_gloo_process_groups ...................... True
146
+ enable_msc ...................................... True
147
+ enable_one_logger ............................... True
148
+ encoder_num_layers .............................. 16
149
+ encoder_seq_length .............................. 4096
150
+ end_weight_decay ................................ 0.1
151
+ eod_mask_loss ................................... False
152
+ error_injection_rate ............................ 0
153
+ error_injection_type ............................ transient_error
154
+ eval_interval ................................... 1000000
155
+ eval_iters ...................................... 0
156
+ evidence_data_path .............................. None
157
+ exit_duration_in_mins ........................... None
158
+ exit_interval ................................... None
159
+ exit_on_missing_checkpoint ...................... False
160
+ exit_signal_handler ............................. False
161
+ exp_avg_dtype ................................... torch.float32
162
+ exp_avg_sq_dtype ................................ torch.float32
163
+ expert_model_parallel_size ...................... 8
164
+ expert_tensor_parallel_size ..................... 1
165
+ export_force_local_attention .................... False
166
+ export_kd_cfg ................................... None
167
+ export_kd_teacher_ckpt_format ................... None
168
+ export_kd_teacher_load .......................... None
169
+ export_kv_cache_quant ........................... False
170
+ export_legacy_megatron .......................... False
171
+ export_model_type ............................... GPTModel
172
+ export_moe_apply_probs_on_input ................. False
173
+ export_qk_l2_norm ............................... False
174
+ export_quant_cfg ................................ None
175
+ export_real_quant_cfg ........................... None
176
+ export_te_mcore_model ........................... False
177
+ external_cuda_graph ............................. False
178
+ ffn_hidden_size ................................. 14336
179
+ finetune ........................................ False
180
+ finetune_data_split ............................. train
181
+ finetune_hf_dataset ............................. None
182
+ first_last_layers_bf16 .......................... False
183
+ flash_decode .................................... False
184
+ fp16 ............................................ False
185
+ fp16_lm_cross_entropy ........................... False
186
+ fp32_residual_connection ........................ False
187
+ fp8 ............................................. e4m3
188
+ fp8_amax_compute_algo ........................... most_recent
189
+ fp8_amax_history_len ............................ 1
190
+ fp8_interval .................................... 1
191
+ fp8_margin ...................................... 0
192
+ fp8_param_gather ................................ False
193
+ fp8_recipe ...................................... mxfp8
194
+ fp8_wgrad ....................................... True
195
+ fsdp_double_buffer .............................. False
196
+ global_batch_size ............................... 16
197
+ grad_reduce_in_bf16 ............................. False
198
+ gradient_accumulation_fusion .................... True
199
+ gradient_reduce_div_fusion ...................... True
200
+ group_query_attention ........................... True
201
+ head_lr_mult .................................... 1.0
202
+ heterogeneous_layers_config_encoded_json ........ None
203
+ heterogeneous_layers_config_path ................ None
204
+ hidden_dropout .................................. 0.0
205
+ hidden_size ..................................... 4096
206
+ hierarchical_context_parallel_sizes ............. None
207
+ high_priority_stream_groups ..................... []
208
+ hybrid_attention_ratio .......................... 0.0
209
+ hybrid_mlp_ratio ................................ 0.0
210
+ hybrid_override_pattern ......................... None
211
+ hysteresis ...................................... 2
212
+ ict_head_size ................................... None
213
+ ict_load ........................................ None
214
+ img_h ........................................... 224
215
+ img_w ........................................... 224
216
+ indexer_batch_size .............................. 128
217
+ indexer_log_interval ............................ 1000
218
+ inference_batch_times_seqlen_threshold .......... -1
219
+ inference_dynamic_batching ...................... False
220
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
221
+ inference_dynamic_batching_buffer_overflow_factor None
222
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
223
+ inference_dynamic_batching_chunk_size ........... 256
224
+ inference_dynamic_batching_max_requests_override None
225
+ inference_dynamic_batching_max_tokens_override .. None
226
+ inference_dynamic_batching_num_cuda_graphs ...... 16
227
+ inference_max_batch_size ........................ 8
228
+ inference_max_seq_length ........................ 2560
229
+ inference_rng_tracker ........................... False
230
+ init_method_std ................................. 0.01
231
+ init_method_xavier_uniform ...................... False
232
+ init_model_with_meta_device ..................... False
233
+ initial_loss_scale .............................. 4294967296
234
+ inprocess_active_world_size ..................... 8
235
+ inprocess_barrier_timeout ....................... 120
236
+ inprocess_completion_timeout .................... 120
237
+ inprocess_empty_cuda_cache ...................... False
238
+ inprocess_granularity ........................... node
239
+ inprocess_hard_timeout .......................... 90
240
+ inprocess_heartbeat_interval .................... 30
241
+ inprocess_heartbeat_timeout ..................... 60
242
+ inprocess_last_call_wait ........................ 1
243
+ inprocess_max_iterations ........................ None
244
+ inprocess_monitor_process_interval .............. 1.0
245
+ inprocess_monitor_thread_interval ............... 1.0
246
+ inprocess_progress_watchdog_interval ............ 1.0
247
+ inprocess_restart ............................... False
248
+ inprocess_soft_timeout .......................... 60
249
+ inprocess_termination_grace_time ................ 1
250
+ is_hybrid_model ................................. False
251
+ iter_per_epoch .................................. 1250
252
+ iterations_to_skip .............................. []
253
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
254
+ kitchen_config_file ............................. None
255
+ kitchen_recipe_number ........................... None
256
+ kv_channels ..................................... 128
257
+ kv_lora_rank .................................... 32
258
+ lazy_mpu_init ................................... None
259
+ load ............................................ None
260
+ load_main_params_from_ckpt ...................... None
261
+ load_model_opt_format ........................... False
262
+ local_rank ...................................... 0
263
+ log_energy ...................................... False
264
+ log_interval .................................... 1
265
+ log_loss_scale_to_tensorboard ................... True
266
+ log_memory_to_tensorboard ....................... False
267
+ log_num_zeros_in_grad ........................... False
268
+ log_params_norm ................................. False
269
+ log_progress .................................... False
270
+ log_straggler ................................... False
271
+ log_throughput .................................. True
272
+ log_timers_to_tensorboard ....................... False
273
+ log_validation_ppl_to_tensorboard ............... False
274
+ log_world_size_to_tensorboard ................... False
275
+ logging_level ................................... None
276
+ loss_scale ...................................... None
277
+ loss_scale_window ............................... 1000
278
+ lr .............................................. 0.0001
279
+ lr_decay_iters .................................. 8
280
+ lr_decay_samples ................................ None
281
+ lr_decay_style .................................. cosine
282
+ lr_warmup_fraction .............................. None
283
+ lr_warmup_init .................................. 0.0
284
+ lr_warmup_iters ................................. 5
285
+ lr_warmup_samples ............................... 0
286
+ lr_wsd_decay_iters .............................. None
287
+ lr_wsd_decay_samples ............................ None
288
+ lr_wsd_decay_style .............................. exponential
289
+ main_grads_dtype ................................ torch.float32
290
+ main_params_dtype ............................... torch.float32
291
+ make_vocab_size_divisible_by .................... 128
292
+ mamba_head_dim .................................. 64
293
+ mamba_num_groups ................................ 8
294
+ mamba_num_heads ................................. None
295
+ mamba_state_dim ................................. 128
296
+ manual_gc ....................................... False
297
+ manual_gc_eval .................................. True
298
+ manual_gc_interval .............................. 0
299
+ mask_factor ..................................... 1.0
300
+ mask_prob ....................................... 0.15
301
+ mask_type ....................................... random
302
+ masked_softmax_fusion ........................... False
303
+ max_position_embeddings ......................... 32768
304
+ max_tokens_to_oom ............................... 12000
305
+ memory_snapshot_path ............................ snapshot.pickle
306
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
307
+ micro_batch_size ................................ 1
308
+ microbatch_group_size_per_vp_stage .............. None
309
+ mid_level_dataset_surplus ....................... 0.005
310
+ min_loss_scale .................................. 1.0
311
+ min_lr .......................................... 1e-05
312
+ mlp_chunks_for_prefill .......................... 1
313
+ mmap_bin_files .................................. True
314
+ mock_data ....................................... False
315
+ moe_apply_probs_on_input ........................ False
316
+ moe_aux_loss_coeff .............................. 0.01
317
+ moe_deepep_num_sms .............................. 20
318
+ moe_echo_enable_random_offloading ............... False
319
+ moe_echo_expert_dispatch_overlap ................ False
320
+ moe_echo_expert_dispatcher_type ................. hybridep
321
+ moe_echo_recompute_expert_dispatch .............. False
322
+ moe_enable_deepep ............................... False
323
+ moe_enable_echo ................................. False
324
+ moe_enable_hybridep ............................. False
325
+ moe_enable_moonep ............................... True
326
+ moe_expert_capacity_factor ...................... None
327
+ moe_extended_tp ................................. False
328
+ moe_ffn_hidden_size ............................. 14336
329
+ moe_grouped_gemm ................................ True
330
+ moe_hybridep_num_sms ............................ 16
331
+ moe_input_jitter_eps ............................ None
332
+ moe_layer_freq .................................. 1
333
+ moe_layer_recompute ............................. False
334
+ moe_moonep_async_grad_reduce .................... True
335
+ moe_moonep_num_sms .............................. None
336
+ moe_num_echo_experts ............................ None
337
+ moe_pad_expert_input_to_capacity ................ False
338
+ moe_per_layer_logging ........................... False
339
+ moe_permute_fusion .............................. False
340
+ moe_received_token_capacity ..................... None
341
+ moe_router_bias_update_rate ..................... 0.001
342
+ moe_router_dtype ................................ fp32
343
+ moe_router_enable_expert_bias ................... False
344
+ moe_router_force_load_balancing ................. False
345
+ moe_router_group_topk ........................... None
346
+ moe_router_load_balancing_type .................. aux_loss
347
+ moe_router_num_groups ........................... None
348
+ moe_router_padding_for_fp8 ...................... False
349
+ moe_router_pre_softmax .......................... False
350
+ moe_router_score_function ....................... softmax
351
+ moe_router_topk ................................. 2
352
+ moe_router_topk_scaling_factor .................. None
353
+ moe_shared_expert_intermediate_size ............. None
354
+ moe_shared_expert_overlap ....................... False
355
+ moe_token_dispatcher_type ....................... flex
356
+ moe_token_drop_policy ........................... probs
357
+ moe_upcycling_granularity ....................... 1
358
+ moe_use_device_initiated_grouped_gemm ........... True
359
+ moe_use_legacy_grouped_gemm ..................... False
360
+ moe_use_upcycling ............................... False
361
+ moe_z_loss_coeff ................................ None
362
+ mrope_section ................................... None
363
+ mscale .......................................... 1.0
364
+ mscale_all_dim .................................. 0.0
365
+ mtp_loss_scaling_factor ......................... 0.1
366
+ mtp_num_layers .................................. None
367
+ multi_latent_attention .......................... False
368
+ nccl_all_reduce_for_prefill ..................... False
369
+ nccl_communicator_config_path ................... None
370
+ nccl_ub ......................................... False
371
+ no_load_optim ................................... None
372
+ no_load_rng ..................................... None
373
+ no_persist_layer_norm ........................... False
374
+ no_rope_freq .................................... None
375
+ no_save_optim ................................... None
376
+ no_save_rng ..................................... None
377
+ non_persistent_ckpt_type ........................ None
378
+ non_persistent_global_ckpt_dir .................. None
379
+ non_persistent_local_ckpt_algo .................. fully_parallel
380
+ non_persistent_local_ckpt_dir ................... None
381
+ non_persistent_save_interval .................... None
382
+ norm_epsilon .................................... 1e-05
383
+ normalization ................................... RMSNorm
384
+ num_attention_heads ............................. 32
385
+ num_channels .................................... 3
386
+ num_classes ..................................... 1000
387
+ num_dataset_builder_threads ..................... 1
388
+ num_distributed_optimizer_instances ............. 1
389
+ num_experts ..................................... 8
390
+ num_layers ...................................... 16
391
+ num_layers_at_end_in_bf16 ....................... 1
392
+ num_layers_at_start_in_bf16 ..................... 1
393
+ num_layers_per_virtual_pipeline_stage ........... None
394
+ num_query_groups ................................ 8
395
+ num_virtual_stages_per_pipeline_rank ............ None
396
+ num_workers ..................................... 2
397
+ object_storage_cache_path ....................... None
398
+ one_logger_async ................................ False
399
+ one_logger_project .............................. megatron-lm
400
+ one_logger_run_name ............................. None
401
+ onnx_safe ....................................... None
402
+ openai_gelu ..................................... False
403
+ optimizer ....................................... adam
404
+ optimizer_cpu_offload ........................... False
405
+ optimizer_offload_fraction ...................... 1.0
406
+ output_bert_embeddings .......................... False
407
+ overlap_cpu_optimizer_d2h_h2d ................... False
408
+ overlap_grad_reduce ............................. True
409
+ overlap_moe_expert_parallel_comm ................ False
410
+ overlap_p2p_comm ................................ False
411
+ overlap_p2p_comm_warmup_flush ................... False
412
+ overlap_param_gather ............................ True
413
+ overlap_param_gather_with_optimizer_step ........ False
414
+ override_opt_param_scheduler .................... False
415
+ params_dtype .................................... torch.bfloat16
416
+ patch_dim ....................................... 16
417
+ per_split_data_args_path ........................ None
418
+ perform_initialization .......................... True
419
+ pin_cpu_grads ................................... True
420
+ pin_cpu_params .................................. True
421
+ pipeline_model_parallel_comm_backend ............ None
422
+ pipeline_model_parallel_layout .................. None
423
+ pipeline_model_parallel_size .................... 1
424
+ position_embedding_type ......................... rope
425
+ pretrained_checkpoint ........................... None
426
+ profile ......................................... False
427
+ profile_ranks ................................... [0]
428
+ profile_step_end ................................ 12
429
+ profile_step_start .............................. 10
430
+ q_lora_rank ..................................... None
431
+ qk_head_dim ..................................... 128
432
+ qk_l2_norm ...................................... False
433
+ qk_layernorm .................................... False
434
+ qk_pos_emb_head_dim ............................. 64
435
+ query_in_block_prob ............................. 0.1
436
+ rampup_batch_size ............................... None
437
+ rank ............................................ 0
438
+ recompute_granularity ........................... None
439
+ recompute_method ................................ None
440
+ recompute_modules ............................... None
441
+ recompute_num_layers ............................ None
442
+ record_memory_history ........................... False
443
+ relative_attention_max_distance ................. 128
444
+ relative_attention_num_buckets .................. 32
445
+ replication ..................................... False
446
+ replication_factor .............................. 2
447
+ replication_jump ................................ None
448
+ rerun_mode ...................................... validate_results
449
+ reset_attention_mask ............................ False
450
+ reset_position_ids .............................. False
451
+ result_rejected_tracker_filename ................ None
452
+ retriever_report_topk_accuracies ................ []
453
+ retriever_score_scaling ......................... False
454
+ retriever_seq_length ............................ 256
455
+ retro_add_retriever ............................. False
456
+ retro_attention_gate ............................ 1
457
+ retro_cyclic_train_iters ........................ None
458
+ retro_encoder_attention_dropout ................. 0.1
459
+ retro_encoder_hidden_dropout .................... 0.1
460
+ retro_encoder_layers ............................ 2
461
+ retro_num_neighbors ............................. 2
462
+ retro_num_retrieved_chunks ...................... 2
463
+ retro_project_dir ............................... None
464
+ retro_verify_neighbor_count ..................... True
465
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
466
+ rope_scaling_factor ............................. 8.0
467
+ rope_type ....................................... None
468
+ rotary_base ..................................... 1000000
469
+ rotary_interleaved .............................. False
470
+ rotary_percent .................................. 1.0
471
+ rotary_scaling_factor ........................... 1.0
472
+ rotary_seq_len_interpolation_factor ............. None
473
+ run_workload_inspector_server ................... False
474
+ sample_rate ..................................... 1.0
475
+ save ............................................ None
476
+ save_interval ................................... None
477
+ save_retain_interval ............................ None
478
+ scatter_gather_tensors_in_pipeline .............. True
479
+ seed ............................................ 1234
480
+ seq_length ...................................... 4096
481
+ sequence_parallel ............................... False
482
+ sft ............................................. False
483
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
484
+ sgd_momentum .................................... 0.9
485
+ short_seq_prob .................................. 0.1
486
+ skip_train ...................................... False
487
+ skipped_train_samples ........................... 0
488
+ spec ............................................ None
489
+ split ........................................... 100,0,0
490
+ squared_relu .................................... False
491
+ start_weight_decay .............................. 0.1
492
+ straggler_ctrlr_port ............................ 65535
493
+ straggler_minmax_count .......................... 1
494
+ suggested_communication_unit_size ............... None
495
+ swiglu .......................................... True
496
+ swin_backbone_type .............................. tiny
497
+ symmetric_ar_type ............................... None
498
+ te_rng_tracker .................................. True
499
+ tensor_model_parallel_size ...................... 1
500
+ tensorboard_dir ................................. None
501
+ tensorboard_log_interval ........................ 1
502
+ tensorboard_queue_size .......................... 1000
503
+ test_data_path .................................. None
504
+ test_mode ....................................... False
505
+ tiktoken_num_special_tokens ..................... 1000
506
+ tiktoken_pattern ................................ None
507
+ tiktoken_special_tokens ......................... None
508
+ timing_log_level ................................ 0
509
+ timing_log_option ............................... minmax
510
+ titles_data_path ................................ None
511
+ tokenizer_model ................................. None
512
+ tokenizer_type .................................. GPT2BPETokenizer
513
+ torch_fsdp2_reshard_after_forward ............... True
514
+ tp_comm_bootstrap_backend ....................... nccl
515
+ tp_comm_bulk_dgrad .............................. True
516
+ tp_comm_bulk_wgrad .............................. True
517
+ tp_comm_overlap ................................. False
518
+ tp_comm_overlap_ag .............................. True
519
+ tp_comm_overlap_cfg ............................. None
520
+ tp_comm_overlap_rs .............................. True
521
+ tp_comm_overlap_rs_dgrad ........................ False
522
+ tp_comm_split_ag ................................ True
523
+ tp_comm_split_rs ................................ True
524
+ train_data_path ................................. None
525
+ train_iters ..................................... 12
526
+ train_samples ................................... None
527
+ train_sync_interval ............................. None
528
+ transformer_impl ................................ transformer_engine
529
+ transformer_pipeline_model_parallel_size ........ 1
530
+ untie_embeddings_and_output_weights ............. True
531
+ use_checkpoint_args ............................. False
532
+ use_checkpoint_opt_param_scheduler .............. False
533
+ use_cpu_initialization .......................... None
534
+ use_custom_fsdp ................................. False
535
+ use_dist_ckpt ................................... True
536
+ use_dist_ckpt_deprecated ........................ False
537
+ use_distributed_optimizer ....................... True
538
+ use_flash_attn .................................. False
539
+ use_legacy_models ............................... False
540
+ use_mp_args_from_checkpoint_args ................ False
541
+ use_one_sent_docs ............................... False
542
+ use_persistent_ckpt_worker ...................... False
543
+ use_precision_aware_optimizer ................... False
544
+ use_pytorch_profiler ............................ False
545
+ use_ring_exchange_p2p ........................... False
546
+ use_rope_scaling ................................ False
547
+ use_rotary_position_embeddings .................. False
548
+ use_sharp ....................................... False
549
+ use_tokenizer_model_from_checkpoint_args ........ True
550
+ use_torch_fsdp2 ................................. False
551
+ use_torch_optimizer_for_cpu_offload ............. False
552
+ use_tp_pp_dp_mapping ............................ False
553
+ v_head_dim ...................................... 128
554
+ valid_data_path ................................. None
555
+ variable_seq_lengths ............................ False
556
+ virtual_pipeline_model_parallel_size ............ None
557
+ vision_backbone_type ............................ vit
558
+ vision_pretraining .............................. False
559
+ vision_pretraining_type ......................... classify
560
+ vocab_extra_ids ................................. 0
561
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
562
+ vocab_size ...................................... None
563
+ wandb_exp_name ..................................
564
+ wandb_project ...................................
565
+ wandb_save_dir ..................................
566
+ weight_decay .................................... 0.1
567
+ weight_decay_incr_style ......................... constant
568
+ wgrad_deferral_limit ............................ 0
569
+ world_size ...................................... 8
570
+ yaml_cfg ........................................ None
571
+ -------------------- end of arguments ---------------------
572
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
573
+ > building GPT2BPETokenizer tokenizer ...
574
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
575
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
576
+ > initializing torch distributed ...
577
+ /workspace/Megatron-MoE-Echo/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
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
580
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
582
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank [Gloo] Rank 56 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 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
596
+ > initialized tensor model parallel with size 1
597
+ > initialized pipeline model parallel with size 1
598
+ > setting random seeds to 1234 ...
599
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
600
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
601
+ [Gloo] Rank 0[Gloo] Rank is connected to 00 peer ranks. is connected to Expected number of connected peer ranks is : 00 peer ranks.
602
+ 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
+ [rank4]:[W830 07:37:52.001672560 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.
607
+ [rank2]:[W830 07:37:52.001733066 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.
608
+ [rank1]:[W830 07:37:52.001770729 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.
609
+ [rank6]:[W830 07:37:52.001785479 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
+ [rank5]:[W830 07:37:52.001794814 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.
611
+ [rank3]:[W830 07:37:52.001811850 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
+ [rank7]:[W830 07:37:52.005292510 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.
613
+ > compiling dataset index builder ...
614
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
615
+ make: Nothing to be done for 'default'.
616
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
617
+ >>> done with dataset index builder. Compilation time: 0.036 seconds
618
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
619
+ > compiling and loading fused kernels ...
620
+ [rank0]:[W830 07:37:52.205156224 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.
621
+ >>> done with compiling and loading fused kernels. Compilation time: 3.731 seconds
622
+ time to initialize megatron (seconds): 7.187
623
+ [after megatron is initialized] datetime: 2026-08-30 07:37:58
624
+ building GPT model ...
625
+ /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.
626
+ warnings.warn(
627
+ /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.
628
+ warnings.warn(
629
+ /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.
630
+ warnings.warn(
631
+ /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.
632
+ warnings.warn(
633
+ /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.
634
+ warnings.warn(
635
+ /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.
636
+ warnings.warn(
637
+ /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.
638
+ warnings.warn(
639
+ /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.
640
+ warnings.warn(
641
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3902410752
642
+ 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)
643
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
644
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
645
+ module.output_layer.weight
646
+ Params for bucket 2 (41988096 elements, 41988096 padded size):
647
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
648
+ module.decoder.layers.15.mlp.router.weight
649
+ module.decoder.layers.15.self_attention.linear_proj.weight
650
+ module.decoder.final_layernorm.weight
651
+ module.decoder.layers.15.self_attention.linear_qkv.weight
652
+ module.decoder.layers.15.pre_mlp_layernorm.weight
653
+ Params for bucket 3 (41984000 elements, 41984000 padded size):
654
+ module.decoder.layers.14.self_attention.linear_qkv.weight
655
+ module.decoder.layers.14.pre_mlp_layernorm.weight
656
+ module.decoder.layers.14.mlp.router.weight
657
+ module.decoder.layers.14.self_attention.linear_proj.weight
658
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
659
+ Params for bucket 4 (41984000 elements, 41984000 padded size):
660
+ module.decoder.layers.13.mlp.router.weight
661
+ module.decoder.layers.13.self_attention.linear_proj.weight
662
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
663
+ module.decoder.layers.13.pre_mlp_layernorm.weight
664
+ module.decoder.layers.13.self_attention.linear_qkv.weight
665
+ Params for bucket 5 (41984000 elements, 41984000 padded size):
666
+ module.decoder.layers.12.self_attention.linear_qkv.weight
667
+ module.decoder.layers.12.pre_mlp_layernorm.weight
668
+ module.decoder.layers.12.mlp.router.weight
669
+ module.decoder.layers.12.self_attention.linear_proj.weight
670
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
671
+ Params for bucket 6 (41984000 elements, 41984000 padded size):
672
+ module.decoder.layers.11.mlp.router.weight
673
+ module.decoder.layers.11.self_attention.linear_proj.weight
674
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
675
+ module.decoder.layers.11.pre_mlp_layernorm.weight
676
+ module.decoder.layers.11.self_attention.linear_qkv.weight
677
+ Params for bucket 7 (41984000 elements, 41984000 padded size):
678
+ module.decoder.layers.10.self_attention.linear_qkv.weight
679
+ module.decoder.layers.10.pre_mlp_layernorm.weight
680
+ module.decoder.layers.10.mlp.router.weight
681
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
682
+ module.decoder.layers.10.self_attention.linear_proj.weight
683
+ Params for bucket 8 (41984000 elements, 41984000 padded size):
684
+ module.decoder.layers.9.mlp.router.weight
685
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
686
+ module.decoder.layers.9.self_attention.linear_proj.weight
687
+ module.decoder.layers.9.self_attention.linear_qkv.weight
688
+ module.decoder.layers.9.pre_mlp_layernorm.weight
689
+ Params for bucket 9 (41984000 elements, 41984000 padded size):
690
+ module.decoder.layers.8.self_attention.linear_qkv.weight
691
+ module.decoder.layers.8.pre_mlp_layernorm.weight
692
+ module.decoder.layers.8.mlp.router.weight
693
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
694
+ module.decoder.layers.8.self_attention.linear_proj.weight
695
+ Params for bucket 10 (41984000 elements, 41984000 padded size):
696
+ module.decoder.layers.7.self_attention.linear_qkv.weight
697
+ module.decoder.layers.7.pre_mlp_layernorm.weight
698
+ module.decoder.layers.7.mlp.router.weight
699
+ module.decoder.layers.7.self_attention.linear_proj.weight
700
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
701
+ Params for bucket 11 (41984000 elements, 41984000 padded size):
702
+ module.decoder.layers.6.mlp.router.weight
703
+ module.decoder.layers.6.self_attention.linear_proj.weight
704
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
705
+ module.decoder.layers.6.pre_mlp_layernorm.weight
706
+ module.decoder.layers.6.self_attention.linear_qkv.weight
707
+ Params for bucket 12 (41984000 elements, 41984000 padded size):
708
+ module.decoder.layers.5.mlp.router.weight
709
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
710
+ module.decoder.layers.5.self_attention.linear_qkv.weight
711
+ module.decoder.layers.5.pre_mlp_layernorm.weight
712
+ module.decoder.layers.5.self_attention.linear_proj.weight
713
+ Params for bucket 13 (41984000 elements, 41984000 padded size):
714
+ module.decoder.layers.4.mlp.router.weight
715
+ module.decoder.layers.4.self_attention.linear_proj.weight
716
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
717
+ module.decoder.layers.4.pre_mlp_layernorm.weight
718
+ module.decoder.layers.4.self_attention.linear_qkv.weight
719
+ Params for bucket 14 (41984000 elements, 41984000 padded size):
720
+ module.decoder.layers.3.self_attention.linear_qkv.weight
721
+ module.decoder.layers.3.pre_mlp_layernorm.weight
722
+ module.decoder.layers.3.mlp.router.weight
723
+ module.decoder.layers.3.self_attention.linear_proj.weight
724
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
725
+ Params for bucket 15 (41984000 elements, 41984000 padded size):
726
+ module.decoder.layers.2.mlp.router.weight
727
+ module.decoder.layers.2.self_attention.linear_proj.weight
728
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
729
+ module.decoder.layers.2.self_attention.linear_qkv.weight
730
+ module.decoder.layers.2.pre_mlp_layernorm.weight
731
+ Params for bucket 16 (41984000 elements, 41984000 padded size):
732
+ module.decoder.layers.1.mlp.router.weight
733
+ module.decoder.layers.1.self_attention.linear_qkv.weight
734
+ module.decoder.layers.1.pre_mlp_layernorm.weight
735
+ module.decoder.layers.1.self_attention.linear_proj.weight
736
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
737
+ Params for bucket 17 (41984000 elements, 41984000 padded size):
738
+ module.decoder.layers.0.self_attention.linear_qkv.weight
739
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
740
+ module.decoder.layers.0.self_attention.linear_proj.weight
741
+ module.decoder.layers.0.pre_mlp_layernorm.weight
742
+ module.decoder.layers.0.mlp.router.weight
743
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
744
+ module.embedding.word_embeddings.weight
745
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 32
746
+ Params for bucket 1 (58720256 elements, 58720256 padded size):
747
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
748
+ Params for bucket 2 (117440512 elements, 117440512 padded size):
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
750
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
751
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
752
+ Params for bucket 4 (117440512 elements, 117440512 padded size):
753
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
754
+ Params for bucket 5 (58720256 elements, 58720256 padded size):
755
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
756
+ Params for bucket 6 (117440512 elements, 117440512 padded size):
757
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
758
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
759
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
760
+ Params for bucket 8 (117440512 elements, 117440512 padded size):
761
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
762
+ Params for bucket 9 (58720256 elements, 58720256 padded size):
763
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
764
+ Params for bucket 10 (117440512 elements, 117440512 padded size):
765
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
766
+ Params for bucket 11 (58720256 elements, 58720256 padded size):
767
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
768
+ Params for bucket 12 (117440512 elements, 117440512 padded size):
769
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
770
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
771
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
772
+ Params for bucket 14 (117440512 elements, 117440512 padded size):
773
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
774
+ Params for bucket 15 (58720256 elements, 58720256 padded size):
775
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
776
+ Params for bucket 16 (117440512 elements, 117440512 padded size):
777
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
778
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
779
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
780
+ Params for bucket 18 (117440512 elements, 117440512 padded size):
781
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
782
+ Params for bucket 19 (58720256 elements, 58720256 padded size):
783
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
784
+ Params for bucket 20 (117440512 elements, 117440512 padded size):
785
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
786
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
788
+ Params for bucket 22 (117440512 elements, 117440512 padded size):
789
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
790
+ Params for bucket 23 (58720256 elements, 58720256 padded size):
791
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
792
+ Params for bucket 24 (117440512 elements, 117440512 padded size):
793
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
794
+ Params for bucket 25 (58720256 elements, 58720256 padded size):
795
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
796
+ Params for bucket 26 (117440512 elements, 117440512 padded size):
797
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
798
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
799
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
800
+ Params for bucket 28 (117440512 elements, 117440512 padded size):
801
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
802
+ Params for bucket 29 (58720256 elements, 58720256 padded size):
803
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
804
+ Params for bucket 30 (117440512 elements, 117440512 padded size):
805
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
806
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
807
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
808
+ Params for bucket 32 (117440512 elements, 117440512 padded size):
809
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
810
+ 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 0x7304dcc2d250>, config_logger_dir='')
811
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
812
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:38:01
813
+ > building train, validation, and test datasets ...
814
+ > datasets target sizes (minimum size):
815
+ train: 192
816
+ validation: 0
817
+ test: 0
818
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
819
+ > building train, validation, and test datasets for GPT ...
820
+ 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 0x73023670ab40>, 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)
821
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
822
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
823
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
824
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
825
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
826
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
827
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
828
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
829
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
830
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
831
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
832
+ > finished creating GPT datasets ...
833
+ [after dataloaders are built] datetime: 2026-08-30 07:38:02
834
+ done with setup ...
835
+ (min, max) time across ranks (ms):
836
+ model-and-optimizer-setup ......................: (3050.02, 3053.61)
837
+ train/valid/test-data-iterators-setup ..........: (67.56, 74.26)
838
+ training ...
839
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
840
+ [before the start of training step] datetime: 2026-08-30 07:38:02
841
+ INFO:moonep.api:MoonEP comm buffer sizes:
842
+ config : R=8, S=4096, H=4096, E=8, K=2, NvS=8446, NvS_padded=8448
843
+ sm_count : num_sms=32, num_sms_dedup=148
844
+ hidden_buffer_size : 66.00 MiB
845
+ meta_buffer_size : 2.00 MiB
846
+ local_temp_buffer_size : 512.66 KiB
847
+ total_size : 68.50 MiB
848
+ [MoonEP selfcheck 1/4] dropped groups empty OK | covered rows [0, 8] sizes [768, 7552] | duplicated experts [[1], [-1], [4], [1], [1], [1], [4], [4]]
849
+ [MoonEP selfcheck 2/4] dropped groups empty OK | covered rows [0, 8] sizes [7808, 512] | duplicated experts [[4], [6], [4], [-1], [3], [7], [0], [4]]
850
+ [MoonEP selfcheck 3/4] dropped groups empty OK | covered rows [0, 8] sizes [8064, 256] | duplicated experts [[3], [3], [-1], [2], [0], [2], [2], [5]]
851
+ [MoonEP selfcheck 4/4] dropped groups empty OK | covered rows [0, 8] sizes [8192, 0] | duplicated experts [[-1], [0], [1], [1], [0], [1], [1], [1]]
852
+ Number of parameters in transformer block in billions: 23.22
853
+ Number of parameters in embedding layers in billions: 0.41
854
+ Total number of parameters in billions: 23.63
855
+ Number of parameters in most loaded shard in billions: 23.6320
856
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
857
+ [2026-08-30 07:38:22] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 20584.9 | throughput per GPU (TFLOP/s/GPU): 16.2 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.104376E+01 | loss scale: 1.0 | grad norm: 35.706 | number of skipped iterations: 0 | number of nan iterations: 0 |
858
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 53350.6591796875 | max allocated: 53350.65966796875 | reserved: 58996.0 | max reserved: 58996.0
859
+ [2026-08-30 07:38:23] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 545.5 | throughput per GPU (TFLOP/s/GPU): 611.2 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.104935E+01 | loss scale: 1.0 | grad norm: 37.620 | number of skipped iterations: 0 | number of nan iterations: 0 |
860
+ [2026-08-30 07:38:23] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 527.0 | throughput per GPU (TFLOP/s/GPU): 632.6 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.680228E+00 | loss scale: 1.0 | grad norm: 42.235 | number of skipped iterations: 0 | number of nan iterations: 0 |
861
+ Capture CUDA graph for training!!!
862
+ CUDA graph capture done!!!
863
+ [2026-08-30 07:38:24] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1199.8 | throughput per GPU (TFLOP/s/GPU): 277.9 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.120710E+01 | loss scale: 1.0 | grad norm: 89.513 | number of skipped iterations: 0 | number of nan iterations: 0 |
864
+ [2026-08-30 07:38:25] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 492.5 | throughput per GPU (TFLOP/s/GPU): 677.0 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.015630E+01 | loss scale: 1.0 | grad norm: 15.498 | number of skipped iterations: 0 | number of nan iterations: 0 |
865
+ [2026-08-30 07:38:25] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 492.9 | throughput per GPU (TFLOP/s/GPU): 676.3 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.015372E+01 | loss scale: 1.0 | grad norm: 43.851 | number of skipped iterations: 0 | number of nan iterations: 0 |
866
+ [2026-08-30 07:38:26] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 493.7 | throughput per GPU (TFLOP/s/GPU): 675.3 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.880402E+00 | loss scale: 1.0 | grad norm: 8.163 | number of skipped iterations: 0 | number of nan iterations: 0 |
867
+ [2026-08-30 07:38:26] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 475.2 | throughput per GPU (TFLOP/s/GPU): 701.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.046215E+00 | loss scale: 1.0 | grad norm: 6.777 | number of skipped iterations: 0 | number of nan iterations: 0 |
868
+ [2026-08-30 07:38:27] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 462.7 | throughput per GPU (TFLOP/s/GPU): 720.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.707321E+00 | loss scale: 1.0 | grad norm: 6.987 | number of skipped iterations: 0 | number of nan iterations: 0 |
869
+ [2026-08-30 07:38:27] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 460.3 | throughput per GPU (TFLOP/s/GPU): 724.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.472165E+00 | loss scale: 1.0 | grad norm: 6.356 | number of skipped iterations: 0 | number of nan iterations: 0 |
870
+ [2026-08-30 07:38:28] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 472.7 | throughput per GPU (TFLOP/s/GPU): 705.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.377650E+00 | loss scale: 1.0 | grad norm: 5.160 | number of skipped iterations: 0 | number of nan iterations: 0 |
871
+ [2026-08-30 07:38:28] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 485.4 | throughput per GPU (TFLOP/s/GPU): 686.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.271275E+00 | loss scale: 1.0 | grad norm: 4.835 | number of skipped iterations: 0 | number of nan iterations: 0 |
872
+ [after training is done] datetime: 2026-08-30 07:38:28
873
+ [rank0]:[W830 07:38:30.397503435 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())
874
+ [rank1]:[W830 07:38:31.553407306 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())
875
+ [rank2]:[W830 07:38:31.554170131 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())
876
+ [rank5]:[W830 07:38:31.557263918 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())
877
+ [rank7]:[W830 07:38:31.565285189 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())
878
+ [rank3]:[W830 07:38:31.880926549 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())
879
+ [rank6]:[W830 07:38:31.916216084 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())
880
+ [rank4]:[W830 07:38:31.927116863 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())
881
+ W0830 07:38:34.669000 784984 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
882
+ W0830 07:38:34.670000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785165 closing signal SIGTERM
883
+ W0830 07:38:34.670000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785166 closing signal SIGTERM
884
+ W0830 07:38:34.672000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785167 closing signal SIGTERM
885
+ W0830 07:38:34.673000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785168 closing signal SIGTERM
886
+ W0830 07:38:34.674000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785169 closing signal SIGTERM
887
+ W0830 07:38:34.675000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785170 closing signal SIGTERM
888
+ W0830 07:38:34.677000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785171 closing signal SIGTERM
889
+ W0830 07:38:34.678000 784984 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 785172 closing signal SIGTERM
890
+ Traceback (most recent call last):
891
+ File "/usr/local/bin/torchrun", line 8, in <module>
892
+ sys.exit(main())
893
+ ^^^^^^
894
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
895
+ return f(*args, **kwargs)
896
+ ^^^^^^^^^^^^^^^^^^
897
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
898
+ run(args)
899
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
900
+ elastic_launch(
901
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
902
+ return launch_agent(self._config, self._entrypoint, list(args))
903
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
904
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
905
+ result = agent.run()
906
+ ^^^^^^^^^^^
907
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
908
+ result = f(*args, **kwargs)
909
+ ^^^^^^^^^^^^^^^^^^
910
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
911
+ result = self._invoke_run(role)
912
+ ^^^^^^^^^^^^^^^^^^^^^^
913
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
914
+ time.sleep(monitor_interval)
915
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
916
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
917
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 784984 got signal: 15
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2/mem_samples.csv ADDED
@@ -0,0 +1,216 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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, 4
10
+ 1, 4
11
+ 2, 1
12
+ 3, 1
13
+ 4, 1
14
+ 5, 1
15
+ 6, 1
16
+ 7, 1
17
+ 0, 4
18
+ 1, 4
19
+ 2, 4
20
+ 3, 4
21
+ 4, 4
22
+ 5, 4
23
+ 6, 4
24
+ 7, 1
25
+ 0, 1
26
+ 1, 1
27
+ 2, 1
28
+ 3, 1
29
+ 4, 3
30
+ 5, 4
31
+ 6, 4
32
+ 7, 4
33
+ 0, 0
34
+ 1, 0
35
+ 2, 0
36
+ 3, 0
37
+ 4, 0
38
+ 5, 0
39
+ 6, 0
40
+ 7, 0
41
+ 0, 1
42
+ 1, 1
43
+ 2, 1
44
+ 3, 1
45
+ 4, 1
46
+ 5, 1
47
+ 6, 1
48
+ 7, 1
49
+ 0, 4
50
+ 1, 4
51
+ 2, 4
52
+ 3, 4
53
+ 4, 1
54
+ 5, 1
55
+ 6, 1
56
+ 7, 1
57
+ 0, 4
58
+ 1, 4
59
+ 2, 4
60
+ 3, 4
61
+ 4, 4
62
+ 5, 4
63
+ 6, 4
64
+ 7, 4
65
+ 0, 4
66
+ 1, 4
67
+ 2, 4
68
+ 3, 4
69
+ 4, 4
70
+ 5, 4
71
+ 6, 4
72
+ 7, 4
73
+ 0, 821
74
+ 1, 821
75
+ 2, 821
76
+ 3, 821
77
+ 4, 821
78
+ 5, 821
79
+ 6, 821
80
+ 7, 821
81
+ 0, 1895
82
+ 1, 1895
83
+ 2, 1895
84
+ 3, 1895
85
+ 4, 1895
86
+ 5, 1895
87
+ 6, 1895
88
+ 7, 1895
89
+ 0, 37299
90
+ 1, 37287
91
+ 2, 37305
92
+ 3, 37287
93
+ 4, 37301
94
+ 5, 37315
95
+ 6, 37293
96
+ 7, 37315
97
+ 0, 46031
98
+ 1, 46001
99
+ 2, 46077
100
+ 3, 46091
101
+ 4, 46063
102
+ 5, 46065
103
+ 6, 46091
104
+ 7, 46115
105
+ 0, 46473
106
+ 1, 46473
107
+ 2, 46457
108
+ 3, 46451
109
+ 4, 46457
110
+ 5, 46473
111
+ 6, 46495
112
+ 7, 46513
113
+ 0, 46921
114
+ 1, 46917
115
+ 2, 46917
116
+ 3, 46917
117
+ 4, 46917
118
+ 5, 46919
119
+ 6, 46919
120
+ 7, 46917
121
+ 0, 47383
122
+ 1, 47381
123
+ 2, 47381
124
+ 3, 47381
125
+ 4, 47381
126
+ 5, 47381
127
+ 6, 47381
128
+ 7, 47381
129
+ 0, 66255
130
+ 1, 66253
131
+ 2, 66253
132
+ 3, 66253
133
+ 4, 66253
134
+ 5, 66253
135
+ 6, 66253
136
+ 7, 66253
137
+ 0, 67129
138
+ 1, 67159
139
+ 2, 67131
140
+ 3, 67167
141
+ 4, 67143
142
+ 5, 67131
143
+ 6, 67143
144
+ 7, 67151
145
+ 0, 68211
146
+ 1, 68201
147
+ 2, 68207
148
+ 3, 68197
149
+ 4, 68211
150
+ 5, 68229
151
+ 6, 68219
152
+ 7, 68205
153
+ 0, 82117
154
+ 1, 82089
155
+ 2, 82141
156
+ 3, 82097
157
+ 4, 82135
158
+ 5, 82219
159
+ 6, 82189
160
+ 7, 82121
161
+ 0, 96849
162
+ 1, 96845
163
+ 2, 96843
164
+ 3, 96843
165
+ 4, 96845
166
+ 5, 96843
167
+ 6, 96843
168
+ 7, 96843
169
+ 0, 96849
170
+ 1, 96845
171
+ 2, 96843
172
+ 3, 96843
173
+ 4, 96845
174
+ 5, 96843
175
+ 6, 96843
176
+ 7, 96843
177
+ 0, 96849
178
+ 1, 96845
179
+ 2, 96843
180
+ 3, 96843
181
+ 4, 96845
182
+ 5, 96843
183
+ 6, 96843
184
+ 7, 96843
185
+ 0, 96203
186
+ 1, 96199
187
+ 2, 96197
188
+ 3, 96197
189
+ 4, 96199
190
+ 5, 96197
191
+ 6, 96197
192
+ 7, 96197
193
+ 0, 94775
194
+ 1, 94451
195
+ 2, 94449
196
+ 3, 94449
197
+ 4, 94451
198
+ 5, 94061
199
+ 6, 94513
200
+ 7, 95095
201
+ 0, 76259
202
+ 1, 92771
203
+ 2, 76253
204
+ 3, 92769
205
+ 4, 75247
206
+ 5, 4
207
+ 6, 92769
208
+ 7, 92769
209
+ 0, 1
210
+ 1, 1
211
+ 2, 1
212
+ 3, 1
213
+ 4, 1
214
+ 5, 1
215
+ 6, 1
216
+ 7, 91756
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2/run.env ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=moonep
2
+ config=G1
3
+ family=tierA
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=moonep_backend_async_grad_reduce
6
+ commit=2e68c8ad
7
+ moonep_source_dir=/workspace/MoonEP
8
+ moonep_source_branch=megatron-ddp-symmetric-params
9
+ moonep_source_commit=d8bc666
10
+ num_layers=16
11
+ hidden=4096
12
+ ffn_hidden=14336
13
+ moe_ffn_hidden=14336
14
+ num_experts=8
15
+ topk=2
16
+ experts_per_rank=1
17
+ expert_params=176160768
18
+ heads=32
19
+ kv_groups=8
20
+ kv_channels=
21
+ qk_layernorm=0
22
+ shared_ffn=0
23
+ shared_expert_overlap=0
24
+ moe_token_dispatcher=flex
25
+ mla=0
26
+ dense_first_layers=0
27
+ moe_layers=16
28
+ rotary_base=1000000
29
+ rotary_scaling=
30
+ qkv_bias=0
31
+ router_balancing=aux_loss
32
+ aux_loss_coeff=1e-2
33
+ router_pre_softmax=0
34
+ router_fusion=0
35
+ router_dtype=fp32
36
+ device_initiated_grouped_gemm=1
37
+ norm_epsilon=1e-5
38
+ seq_length=4096
39
+ micro_batch_size=1
40
+ global_batch_size=16
41
+ num_microbatches=2
42
+ assignments_per_rank=8192
43
+ mean_tokens_per_expert=8192
44
+ slot_policy=one
45
+ slots_per_rank=1
46
+ capacity_mode=none
47
+ capacity=none
48
+ cuda_graph=1
49
+ probe_mode=0
50
+ async_grad_reduce=1
51
+ branch_role=timing
52
+ memory_dump_mode=none
53
+ memory_snapshot_path=
54
+ cuda_graph_warmup_steps=3
55
+ recompute=none
56
+ force_balance=0
57
+ train_iters=12
58
+ ep_size=8
59
+ echo_dispatch_overlap=1
60
+ echo_dispatch_recompute=0
61
+ ultraep_quota_min_tokens=
62
+ ultraep_weight_sync_join=
63
+ profile_enable=0
64
+ kernel_trace=0
65
+ fp8_mode=mxfp8
66
+ fp8_recipe=mxfp8
67
+ training_seed=1234
68
+ learning_rate=1e-4
69
+ min_learning_rate=1.0e-5
70
+ lr_warmup_iters=5
71
+ state_dump_enable=0
72
+ state_dump_dir=
73
+ moonep_assert_no_pending_reduce=1
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2/train.log ADDED
@@ -0,0 +1,915 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G1 / moonep (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=none cap=none cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 2e68c8ad
8
+ ==============================================================
9
+ W0830 07:38:49.057000 787789 torch/distributed/run.py:766]
10
+ W0830 07:38:49.057000 787789 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:38:49.057000 787789 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 07:38:49.057000 787789 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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
+ 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
27
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
28
+ Number of virtual stages per pipeline stage: None
29
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
30
+ using torch.bfloat16 for parameters ...
31
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
32
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
33
+ ------------------------ arguments ------------------------
34
+ account_for_embedding_in_pipeline_split ......... False
35
+ account_for_loss_in_pipeline_split .............. False
36
+ accumulate_allreduce_grads_in_fp32 .............. True
37
+ adam_beta1 ...................................... 0.9
38
+ adam_beta2 ...................................... 0.999
39
+ adam_eps ........................................ 1e-08
40
+ add_bias_linear ................................. False
41
+ add_position_embedding .......................... False
42
+ add_qkv_bias .................................... False
43
+ adlr_autoresume ................................. False
44
+ adlr_autoresume_interval ........................ 1000
45
+ align_grad_reduce ............................... True
46
+ align_param_gather .............................. False
47
+ app_tag_run_name ................................ None
48
+ app_tag_run_version ............................. 0.0.0
49
+ apply_layernorm_1p .............................. False
50
+ apply_query_key_layer_scaling ................... False
51
+ apply_residual_connection_post_layernorm ........ False
52
+ apply_rope_fusion ............................... True
53
+ async_save ...................................... None
54
+ async_tensor_model_parallel_allreduce ........... True
55
+ attention_backend ............................... AttnBackend.auto
56
+ attention_dropout ............................... 0.0
57
+ attention_softmax_in_fp32 ....................... False
58
+ auto_detect_ckpt_format ......................... False
59
+ barrier_with_L1_time ............................ True
60
+ bert_binary_head ................................ True
61
+ bert_embedder_type .............................. megatron
62
+ bert_load ....................................... None
63
+ bf16 ............................................ True
64
+ bias_dropout_fusion ............................. True
65
+ bias_gelu_fusion ................................ False
66
+ bias_swiglu_fusion .............................. True
67
+ biencoder_projection_dim ........................ 0
68
+ biencoder_shared_query_context_model ............ False
69
+ block_data_path ................................. None
70
+ calc_ft_timeouts ................................ False
71
+ calculate_per_token_loss ........................ False
72
+ check_for_large_grads ........................... False
73
+ check_for_nan_in_loss_and_grad .................. False
74
+ check_for_spiky_loss ............................ False
75
+ check_weight_hash_across_dp_replicas_interval ... None
76
+ ckpt_assume_constant_structure .................. False
77
+ ckpt_convert_format ............................. None
78
+ ckpt_convert_save ............................... None
79
+ ckpt_convert_update_legacy_dist_opt_format ...... False
80
+ ckpt_format ..................................... torch_dist
81
+ ckpt_fully_parallel_load ........................ False
82
+ ckpt_fully_parallel_save ........................ True
83
+ ckpt_fully_parallel_save_deprecated ............. False
84
+ ckpt_step ....................................... None
85
+ classes_fraction ................................ 1.0
86
+ clip_grad ....................................... 1.0
87
+ clone_scatter_output_in_embedding ............... True
88
+ config_logger_dir ...............................
89
+ consumed_train_samples .......................... 0
90
+ consumed_valid_samples .......................... 0
91
+ context_parallel_size ........................... 1
92
+ cp_comm_type .................................... ['p2p']
93
+ create_attention_mask_in_dataloader ............. True
94
+ cross_entropy_fusion_impl ....................... native
95
+ cross_entropy_loss_fusion ....................... False
96
+ cuda_graph_scope ................................ full_iteration
97
+ cuda_graph_warmup_steps ......................... 3
98
+ data_args_path .................................. None
99
+ data_cache_path ................................. None
100
+ data_parallel_random_init ....................... False
101
+ data_parallel_sharding_strategy ................. no_shard
102
+ data_parallel_size .............................. 8
103
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
104
+ data_per_class_fraction ......................... 1.0
105
+ data_sharding ................................... True
106
+ dataloader_type ................................. single
107
+ ddp_average_in_collective ....................... False
108
+ ddp_bucket_size ................................. None
109
+ ddp_num_buckets ................................. None
110
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
111
+ decoder_first_pipeline_num_layers ............... None
112
+ decoder_last_pipeline_num_layers ................ None
113
+ decoder_num_layers .............................. None
114
+ decoder_seq_length .............................. None
115
+ decoupled_lr .................................... None
116
+ decoupled_min_lr ................................ None
117
+ decrease_batch_size_if_needed ................... False
118
+ defer_embedding_wgrad_compute ................... False
119
+ delay_wgrad_compute ............................. False
120
+ deprecated_use_mcore_models ..................... True
121
+ deterministic_mode .............................. False
122
+ dino_bottleneck_size ............................ 256
123
+ dino_freeze_last_layer .......................... 1
124
+ dino_head_hidden_size ........................... 2048
125
+ dino_local_crops_number ......................... 10
126
+ dino_local_img_size ............................. 96
127
+ dino_norm_last_layer ............................ False
128
+ dino_teacher_temp ............................... 0.07
129
+ dino_warmup_teacher_temp ........................ 0.04
130
+ dino_warmup_teacher_temp_epochs ................. 30
131
+ disable_bf16_reduced_precision_matmul ........... False
132
+ disable_mamba_mem_eff_path ...................... False
133
+ disable_straggler_on_startup .................... False
134
+ dist_ckpt_format_deprecated ..................... None
135
+ dist_ckpt_strictness ............................ assume_ok_unexpected
136
+ distribute_saved_activations .................... False
137
+ distributed_backend ............................. nccl
138
+ distributed_timeout_minutes ..................... 10
139
+ embedding_init_method_std ....................... None
140
+ embedding_path .................................. None
141
+ empty_unused_memory_level ....................... 0
142
+ enable_cuda_graph ............................... True
143
+ enable_experimental ............................. False
144
+ enable_ft_package ............................... False
145
+ enable_gloo_process_groups ...................... True
146
+ enable_msc ...................................... True
147
+ enable_one_logger ............................... True
148
+ encoder_num_layers .............................. 16
149
+ encoder_seq_length .............................. 4096
150
+ end_weight_decay ................................ 0.1
151
+ eod_mask_loss ................................... False
152
+ error_injection_rate ............................ 0
153
+ error_injection_type ............................ transient_error
154
+ eval_interval ................................... 1000000
155
+ eval_iters ...................................... 0
156
+ evidence_data_path .............................. None
157
+ exit_duration_in_mins ........................... None
158
+ exit_interval ................................... None
159
+ exit_on_missing_checkpoint ...................... False
160
+ exit_signal_handler ............................. False
161
+ exp_avg_dtype ................................... torch.float32
162
+ exp_avg_sq_dtype ................................ torch.float32
163
+ expert_model_parallel_size ...................... 8
164
+ expert_tensor_parallel_size ..................... 1
165
+ export_force_local_attention .................... False
166
+ export_kd_cfg ................................... None
167
+ export_kd_teacher_ckpt_format ................... None
168
+ export_kd_teacher_load .......................... None
169
+ export_kv_cache_quant ........................... False
170
+ export_legacy_megatron .......................... False
171
+ export_model_type ............................... GPTModel
172
+ export_moe_apply_probs_on_input ................. False
173
+ export_qk_l2_norm ............................... False
174
+ export_quant_cfg ................................ None
175
+ export_real_quant_cfg ........................... None
176
+ export_te_mcore_model ........................... False
177
+ external_cuda_graph ............................. False
178
+ ffn_hidden_size ................................. 14336
179
+ finetune ........................................ False
180
+ finetune_data_split ............................. train
181
+ finetune_hf_dataset ............................. None
182
+ first_last_layers_bf16 .......................... False
183
+ flash_decode .................................... False
184
+ fp16 ............................................ False
185
+ fp16_lm_cross_entropy ........................... False
186
+ fp32_residual_connection ........................ False
187
+ fp8 ............................................. e4m3
188
+ fp8_amax_compute_algo ........................... most_recent
189
+ fp8_amax_history_len ............................ 1
190
+ fp8_interval .................................... 1
191
+ fp8_margin ...................................... 0
192
+ fp8_param_gather ................................ False
193
+ fp8_recipe ...................................... mxfp8
194
+ fp8_wgrad ....................................... True
195
+ fsdp_double_buffer .............................. False
196
+ global_batch_size ............................... 16
197
+ grad_reduce_in_bf16 ............................. False
198
+ gradient_accumulation_fusion .................... True
199
+ gradient_reduce_div_fusion ...................... True
200
+ group_query_attention ........................... True
201
+ head_lr_mult .................................... 1.0
202
+ heterogeneous_layers_config_encoded_json ........ None
203
+ heterogeneous_layers_config_path ................ None
204
+ hidden_dropout .................................. 0.0
205
+ hidden_size ..................................... 4096
206
+ hierarchical_context_parallel_sizes ............. None
207
+ high_priority_stream_groups ..................... []
208
+ hybrid_attention_ratio .......................... 0.0
209
+ hybrid_mlp_ratio ................................ 0.0
210
+ hybrid_override_pattern ......................... None
211
+ hysteresis ...................................... 2
212
+ ict_head_size ................................... None
213
+ ict_load ........................................ None
214
+ img_h ........................................... 224
215
+ img_w ........................................... 224
216
+ indexer_batch_size .............................. 128
217
+ indexer_log_interval ............................ 1000
218
+ inference_batch_times_seqlen_threshold .......... -1
219
+ inference_dynamic_batching ...................... False
220
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
221
+ inference_dynamic_batching_buffer_overflow_factor None
222
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
223
+ inference_dynamic_batching_chunk_size ........... 256
224
+ inference_dynamic_batching_max_requests_override None
225
+ inference_dynamic_batching_max_tokens_override .. None
226
+ inference_dynamic_batching_num_cuda_graphs ...... 16
227
+ inference_max_batch_size ........................ 8
228
+ inference_max_seq_length ........................ 2560
229
+ inference_rng_tracker ........................... False
230
+ init_method_std ................................. 0.01
231
+ init_method_xavier_uniform ...................... False
232
+ init_model_with_meta_device ..................... False
233
+ initial_loss_scale .............................. 4294967296
234
+ inprocess_active_world_size ..................... 8
235
+ inprocess_barrier_timeout ....................... 120
236
+ inprocess_completion_timeout .................... 120
237
+ inprocess_empty_cuda_cache ...................... False
238
+ inprocess_granularity ........................... node
239
+ inprocess_hard_timeout .......................... 90
240
+ inprocess_heartbeat_interval .................... 30
241
+ inprocess_heartbeat_timeout ..................... 60
242
+ inprocess_last_call_wait ........................ 1
243
+ inprocess_max_iterations ........................ None
244
+ inprocess_monitor_process_interval .............. 1.0
245
+ inprocess_monitor_thread_interval ............... 1.0
246
+ inprocess_progress_watchdog_interval ............ 1.0
247
+ inprocess_restart ............................... False
248
+ inprocess_soft_timeout .......................... 60
249
+ inprocess_termination_grace_time ................ 1
250
+ is_hybrid_model ................................. False
251
+ iter_per_epoch .................................. 1250
252
+ iterations_to_skip .............................. []
253
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
254
+ kitchen_config_file ............................. None
255
+ kitchen_recipe_number ........................... None
256
+ kv_channels ..................................... 128
257
+ kv_lora_rank .................................... 32
258
+ lazy_mpu_init ................................... None
259
+ load ............................................ None
260
+ load_main_params_from_ckpt ...................... None
261
+ load_model_opt_format ........................... False
262
+ local_rank ...................................... 0
263
+ log_energy ...................................... False
264
+ log_interval .................................... 1
265
+ log_loss_scale_to_tensorboard ................... True
266
+ log_memory_to_tensorboard ....................... False
267
+ log_num_zeros_in_grad ........................... False
268
+ log_params_norm ................................. False
269
+ log_progress .................................... False
270
+ log_straggler ................................... False
271
+ log_throughput .................................. True
272
+ log_timers_to_tensorboard ....................... False
273
+ log_validation_ppl_to_tensorboard ............... False
274
+ log_world_size_to_tensorboard ................... False
275
+ logging_level ................................... None
276
+ loss_scale ...................................... None
277
+ loss_scale_window ............................... 1000
278
+ lr .............................................. 0.0001
279
+ lr_decay_iters .................................. 8
280
+ lr_decay_samples ................................ None
281
+ lr_decay_style .................................. cosine
282
+ lr_warmup_fraction .............................. None
283
+ lr_warmup_init .................................. 0.0
284
+ lr_warmup_iters ................................. 5
285
+ lr_warmup_samples ............................... 0
286
+ lr_wsd_decay_iters .............................. None
287
+ lr_wsd_decay_samples ............................ None
288
+ lr_wsd_decay_style .............................. exponential
289
+ main_grads_dtype ................................ torch.float32
290
+ main_params_dtype ............................... torch.float32
291
+ make_vocab_size_divisible_by .................... 128
292
+ mamba_head_dim .................................. 64
293
+ mamba_num_groups ................................ 8
294
+ mamba_num_heads ................................. None
295
+ mamba_state_dim ................................. 128
296
+ manual_gc ....................................... False
297
+ manual_gc_eval .................................. True
298
+ manual_gc_interval .............................. 0
299
+ mask_factor ..................................... 1.0
300
+ mask_prob ....................................... 0.15
301
+ mask_type ....................................... random
302
+ masked_softmax_fusion ........................... False
303
+ max_position_embeddings ......................... 32768
304
+ max_tokens_to_oom ............................... 12000
305
+ memory_snapshot_path ............................ snapshot.pickle
306
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
307
+ micro_batch_size ................................ 1
308
+ microbatch_group_size_per_vp_stage .............. None
309
+ mid_level_dataset_surplus ....................... 0.005
310
+ min_loss_scale .................................. 1.0
311
+ min_lr .......................................... 1e-05
312
+ mlp_chunks_for_prefill .......................... 1
313
+ mmap_bin_files .................................. True
314
+ mock_data ....................................... False
315
+ moe_apply_probs_on_input ........................ False
316
+ moe_aux_loss_coeff .............................. 0.01
317
+ moe_deepep_num_sms .............................. 20
318
+ moe_echo_enable_random_offloading ............... False
319
+ moe_echo_expert_dispatch_overlap ................ False
320
+ moe_echo_expert_dispatcher_type ................. hybridep
321
+ moe_echo_recompute_expert_dispatch .............. False
322
+ moe_enable_deepep ............................... False
323
+ moe_enable_echo ................................. False
324
+ moe_enable_hybridep ............................. False
325
+ moe_enable_moonep ............................... True
326
+ moe_expert_capacity_factor ...................... None
327
+ moe_extended_tp ................................. False
328
+ moe_ffn_hidden_size ............................. 14336
329
+ moe_grouped_gemm ................................ True
330
+ moe_hybridep_num_sms ............................ 16
331
+ moe_input_jitter_eps ............................ None
332
+ moe_layer_freq .................................. 1
333
+ moe_layer_recompute ............................. False
334
+ moe_moonep_async_grad_reduce .................... True
335
+ moe_moonep_num_sms .............................. None
336
+ moe_num_echo_experts ............................ None
337
+ moe_pad_expert_input_to_capacity ................ False
338
+ moe_per_layer_logging ........................... False
339
+ moe_permute_fusion .............................. False
340
+ moe_received_token_capacity ..................... None
341
+ moe_router_bias_update_rate ..................... 0.001
342
+ moe_router_dtype ................................ fp32
343
+ moe_router_enable_expert_bias ................... False
344
+ moe_router_force_load_balancing ................. False
345
+ moe_router_group_topk ........................... None
346
+ moe_router_load_balancing_type .................. aux_loss
347
+ moe_router_num_groups ........................... None
348
+ moe_router_padding_for_fp8 ...................... False
349
+ moe_router_pre_softmax .......................... False
350
+ moe_router_score_function ....................... softmax
351
+ moe_router_topk ................................. 2
352
+ moe_router_topk_scaling_factor .................. None
353
+ moe_shared_expert_intermediate_size ............. None
354
+ moe_shared_expert_overlap ....................... False
355
+ moe_token_dispatcher_type ....................... flex
356
+ moe_token_drop_policy ........................... probs
357
+ moe_upcycling_granularity ....................... 1
358
+ moe_use_device_initiated_grouped_gemm ........... True
359
+ moe_use_legacy_grouped_gemm ..................... False
360
+ moe_use_upcycling ............................... False
361
+ moe_z_loss_coeff ................................ None
362
+ mrope_section ................................... None
363
+ mscale .......................................... 1.0
364
+ mscale_all_dim .................................. 0.0
365
+ mtp_loss_scaling_factor ......................... 0.1
366
+ mtp_num_layers .................................. None
367
+ multi_latent_attention .......................... False
368
+ nccl_all_reduce_for_prefill ..................... False
369
+ nccl_communicator_config_path ................... None
370
+ nccl_ub ......................................... False
371
+ no_load_optim ................................... None
372
+ no_load_rng ..................................... None
373
+ no_persist_layer_norm ........................... False
374
+ no_rope_freq .................................... None
375
+ no_save_optim ................................... None
376
+ no_save_rng ..................................... None
377
+ non_persistent_ckpt_type ........................ None
378
+ non_persistent_global_ckpt_dir .................. None
379
+ non_persistent_local_ckpt_algo .................. fully_parallel
380
+ non_persistent_local_ckpt_dir ................... None
381
+ non_persistent_save_interval .................... None
382
+ norm_epsilon .................................... 1e-05
383
+ normalization ................................... RMSNorm
384
+ num_attention_heads ............................. 32
385
+ num_channels .................................... 3
386
+ num_classes ..................................... 1000
387
+ num_dataset_builder_threads ..................... 1
388
+ num_distributed_optimizer_instances ............. 1
389
+ num_experts ..................................... 8
390
+ num_layers ...................................... 16
391
+ num_layers_at_end_in_bf16 ....................... 1
392
+ num_layers_at_start_in_bf16 ..................... 1
393
+ num_layers_per_virtual_pipeline_stage ........... None
394
+ num_query_groups ................................ 8
395
+ num_virtual_stages_per_pipeline_rank ............ None
396
+ num_workers ..................................... 2
397
+ object_storage_cache_path ....................... None
398
+ one_logger_async ................................ False
399
+ one_logger_project .............................. megatron-lm
400
+ one_logger_run_name ............................. None
401
+ onnx_safe ....................................... None
402
+ openai_gelu ..................................... False
403
+ optimizer ....................................... adam
404
+ optimizer_cpu_offload ........................... False
405
+ optimizer_offload_fraction ...................... 1.0
406
+ output_bert_embeddings .......................... False
407
+ overlap_cpu_optimizer_d2h_h2d ................... False
408
+ overlap_grad_reduce ............................. True
409
+ overlap_moe_expert_parallel_comm ................ False
410
+ overlap_p2p_comm ................................ False
411
+ overlap_p2p_comm_warmup_flush ................... False
412
+ overlap_param_gather ............................ True
413
+ overlap_param_gather_with_optimizer_step ........ False
414
+ override_opt_param_scheduler .................... False
415
+ params_dtype .................................... torch.bfloat16
416
+ patch_dim ....................................... 16
417
+ per_split_data_args_path ........................ None
418
+ perform_initialization .......................... True
419
+ pin_cpu_grads ................................... True
420
+ pin_cpu_params .................................. True
421
+ pipeline_model_parallel_comm_backend ............ None
422
+ pipeline_model_parallel_layout .................. None
423
+ pipeline_model_parallel_size .................... 1
424
+ position_embedding_type ......................... rope
425
+ pretrained_checkpoint ........................... None
426
+ profile ......................................... False
427
+ profile_ranks ................................... [0]
428
+ profile_step_end ................................ 12
429
+ profile_step_start .............................. 10
430
+ q_lora_rank ..................................... None
431
+ qk_head_dim ..................................... 128
432
+ qk_l2_norm ...................................... False
433
+ qk_layernorm .................................... False
434
+ qk_pos_emb_head_dim ............................. 64
435
+ query_in_block_prob ............................. 0.1
436
+ rampup_batch_size ............................... None
437
+ rank ............................................ 0
438
+ recompute_granularity ........................... None
439
+ recompute_method ................................ None
440
+ recompute_modules ............................... None
441
+ recompute_num_layers ............................ None
442
+ record_memory_history ........................... False
443
+ relative_attention_max_distance ................. 128
444
+ relative_attention_num_buckets .................. 32
445
+ replication ..................................... False
446
+ replication_factor .............................. 2
447
+ replication_jump ................................ None
448
+ rerun_mode ...................................... validate_results
449
+ reset_attention_mask ............................ False
450
+ reset_position_ids .............................. False
451
+ result_rejected_tracker_filename ................ None
452
+ retriever_report_topk_accuracies ................ []
453
+ retriever_score_scaling ......................... False
454
+ retriever_seq_length ............................ 256
455
+ retro_add_retriever ............................. False
456
+ retro_attention_gate ............................ 1
457
+ retro_cyclic_train_iters ........................ None
458
+ retro_encoder_attention_dropout ................. 0.1
459
+ retro_encoder_hidden_dropout .................... 0.1
460
+ retro_encoder_layers ............................ 2
461
+ retro_num_neighbors ............................. 2
462
+ retro_num_retrieved_chunks ...................... 2
463
+ retro_project_dir ............................... None
464
+ retro_verify_neighbor_count ..................... True
465
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
466
+ rope_scaling_factor ............................. 8.0
467
+ rope_type ....................................... None
468
+ rotary_base ..................................... 1000000
469
+ rotary_interleaved .............................. False
470
+ rotary_percent .................................. 1.0
471
+ rotary_scaling_factor ........................... 1.0
472
+ rotary_seq_len_interpolation_factor ............. None
473
+ run_workload_inspector_server ................... False
474
+ sample_rate ..................................... 1.0
475
+ save ............................................ None
476
+ save_interval ................................... None
477
+ save_retain_interval ............................ None
478
+ scatter_gather_tensors_in_pipeline .............. True
479
+ seed ............................................ 1234
480
+ seq_length ...................................... 4096
481
+ sequence_parallel ............................... False
482
+ sft ............................................. False
483
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
484
+ sgd_momentum .................................... 0.9
485
+ short_seq_prob .................................. 0.1
486
+ skip_train ...................................... False
487
+ skipped_train_samples ........................... 0
488
+ spec ............................................ None
489
+ split ........................................... 100,0,0
490
+ squared_relu .................................... False
491
+ start_weight_decay .............................. 0.1
492
+ straggler_ctrlr_port ............................ 65535
493
+ straggler_minmax_count .......................... 1
494
+ suggested_communication_unit_size ............... None
495
+ swiglu .......................................... True
496
+ swin_backbone_type .............................. tiny
497
+ symmetric_ar_type ............................... None
498
+ te_rng_tracker .................................. True
499
+ tensor_model_parallel_size ...................... 1
500
+ tensorboard_dir ................................. None
501
+ tensorboard_log_interval ........................ 1
502
+ tensorboard_queue_size .......................... 1000
503
+ test_data_path .................................. None
504
+ test_mode ....................................... False
505
+ tiktoken_num_special_tokens ..................... 1000
506
+ tiktoken_pattern ................................ None
507
+ tiktoken_special_tokens ......................... None
508
+ timing_log_level ................................ 0
509
+ timing_log_option ............................... minmax
510
+ titles_data_path ................................ None
511
+ tokenizer_model ................................. None
512
+ tokenizer_type .................................. GPT2BPETokenizer
513
+ torch_fsdp2_reshard_after_forward ............... True
514
+ tp_comm_bootstrap_backend ....................... nccl
515
+ tp_comm_bulk_dgrad .............................. True
516
+ tp_comm_bulk_wgrad .............................. True
517
+ tp_comm_overlap ................................. False
518
+ tp_comm_overlap_ag .............................. True
519
+ tp_comm_overlap_cfg ............................. None
520
+ tp_comm_overlap_rs .............................. True
521
+ tp_comm_overlap_rs_dgrad ........................ False
522
+ tp_comm_split_ag ................................ True
523
+ tp_comm_split_rs ................................ True
524
+ train_data_path ................................. None
525
+ train_iters ..................................... 12
526
+ train_samples ................................... None
527
+ train_sync_interval ............................. None
528
+ transformer_impl ................................ transformer_engine
529
+ transformer_pipeline_model_parallel_size ........ 1
530
+ untie_embeddings_and_output_weights ............. True
531
+ use_checkpoint_args ............................. False
532
+ use_checkpoint_opt_param_scheduler .............. False
533
+ use_cpu_initialization .......................... None
534
+ use_custom_fsdp ................................. False
535
+ use_dist_ckpt ................................... True
536
+ use_dist_ckpt_deprecated ........................ False
537
+ use_distributed_optimizer ....................... True
538
+ use_flash_attn .................................. False
539
+ use_legacy_models ............................... False
540
+ use_mp_args_from_checkpoint_args ................ False
541
+ use_one_sent_docs ............................... False
542
+ use_persistent_ckpt_worker ...................... False
543
+ use_precision_aware_optimizer ................... False
544
+ use_pytorch_profiler ............................ False
545
+ use_ring_exchange_p2p ........................... False
546
+ use_rope_scaling ................................ False
547
+ use_rotary_position_embeddings .................. False
548
+ use_sharp ....................................... False
549
+ use_tokenizer_model_from_checkpoint_args ........ True
550
+ use_torch_fsdp2 ................................. False
551
+ use_torch_optimizer_for_cpu_offload ............. False
552
+ use_tp_pp_dp_mapping ............................ False
553
+ v_head_dim ...................................... 128
554
+ valid_data_path ................................. None
555
+ variable_seq_lengths ............................ False
556
+ virtual_pipeline_model_parallel_size ............ None
557
+ vision_backbone_type ............................ vit
558
+ vision_pretraining .............................. False
559
+ vision_pretraining_type ......................... classify
560
+ vocab_extra_ids ................................. 0
561
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
562
+ vocab_size ...................................... None
563
+ wandb_exp_name ..................................
564
+ wandb_project ...................................
565
+ wandb_save_dir ..................................
566
+ weight_decay .................................... 0.1
567
+ weight_decay_incr_style ......................... constant
568
+ wgrad_deferral_limit ............................ 0
569
+ world_size ...................................... 8
570
+ yaml_cfg ........................................ None
571
+ -------------------- end of arguments ---------------------
572
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
573
+ > building GPT2BPETokenizer tokenizer ...
574
+ /workspace/Megatron-MoE-Echo/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
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
577
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
578
+ > initializing torch distributed ...
579
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
580
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
582
+ [Gloo] Rank [Gloo] Rank 76 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
583
+
584
+ [Gloo] Rank [Gloo] Rank 0 is connected to 72 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks. Expected number of connected peer ranks is :
585
+ 7
586
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank [Gloo] Rank 5 is connected to 37 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
591
+ 7
592
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank 74 is connected to 6 is connected to 7 is connected to 7 peer ranks. 7 peer ranks. Expected number of connected peer ranks is : peer ranks. Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 7
593
+ 7
594
+
595
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
596
+ > initialized tensor model parallel with size 1
597
+ > initialized pipeline model parallel with size 1
598
+ > setting random seeds to 1234 ...
599
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
600
+ [Gloo] Rank [Gloo] Rank 00 is connected to [Gloo] Rank is connected to 000 peer ranks. is connected to peer ranks. Expected number of connected peer ranks is : 0Expected number of connected peer ranks is : 0 peer ranks. 0
601
+ Expected number of connected peer ranks is : [Gloo] Rank
602
+ 00
603
+ 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
+ [rank4]:[W830 07:39:06.187207158 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.
607
+ [rank1]:[W830 07:39:06.189535390 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.
608
+ [rank3]:[W830 07:39:06.189636123 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.
609
+ [rank2]:[W830 07:39:06.189669260 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.
610
+ [rank6]:[W830 07:39:06.189676243 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.
611
+ [rank5]:[W830 07:39:06.189680589 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
+ [rank7]:[W830 07:39:06.189693081 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.
613
+ > compiling dataset index builder ...
614
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
615
+ make: Nothing to be done for 'default'.
616
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
617
+ >>> done with dataset index builder. Compilation time: 0.035 seconds
618
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
619
+ > compiling and loading fused kernels ...
620
+ [rank0]:[W830 07:39:07.345533846 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.
621
+ >>> done with compiling and loading fused kernels. Compilation time: 3.897 seconds
622
+ time to initialize megatron (seconds): 7.217
623
+ [after megatron is initialized] datetime: 2026-08-30 07:39:13
624
+ building GPT model ...
625
+ /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.
626
+ warnings.warn(
627
+ /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.
628
+ warnings.warn(
629
+ /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.
630
+ warnings.warn(
631
+ /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.
632
+ warnings.warn(
633
+ /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.
634
+ warnings.warn(
635
+ /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.
636
+ warnings.warn(
637
+ /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.
638
+ warnings.warn(
639
+ /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.
640
+ warnings.warn(
641
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3902410752
642
+ 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)
643
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
644
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
645
+ module.output_layer.weight
646
+ Params for bucket 2 (41988096 elements, 41988096 padded size):
647
+ module.decoder.final_layernorm.weight
648
+ module.decoder.layers.15.self_attention.linear_qkv.weight
649
+ module.decoder.layers.15.pre_mlp_layernorm.weight
650
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
651
+ module.decoder.layers.15.mlp.router.weight
652
+ module.decoder.layers.15.self_attention.linear_proj.weight
653
+ Params for bucket 3 (41984000 elements, 41984000 padded size):
654
+ module.decoder.layers.14.mlp.router.weight
655
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
656
+ module.decoder.layers.14.self_attention.linear_proj.weight
657
+ module.decoder.layers.14.self_attention.linear_qkv.weight
658
+ module.decoder.layers.14.pre_mlp_layernorm.weight
659
+ Params for bucket 4 (41984000 elements, 41984000 padded size):
660
+ module.decoder.layers.13.self_attention.linear_qkv.weight
661
+ module.decoder.layers.13.pre_mlp_layernorm.weight
662
+ module.decoder.layers.13.mlp.router.weight
663
+ module.decoder.layers.13.self_attention.linear_proj.weight
664
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
665
+ Params for bucket 5 (41984000 elements, 41984000 padded size):
666
+ module.decoder.layers.12.self_attention.linear_qkv.weight
667
+ module.decoder.layers.12.pre_mlp_layernorm.weight
668
+ module.decoder.layers.12.mlp.router.weight
669
+ module.decoder.layers.12.self_attention.linear_proj.weight
670
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
671
+ Params for bucket 6 (41984000 elements, 41984000 padded size):
672
+ module.decoder.layers.11.self_attention.linear_qkv.weight
673
+ module.decoder.layers.11.pre_mlp_layernorm.weight
674
+ module.decoder.layers.11.mlp.router.weight
675
+ module.decoder.layers.11.self_attention.linear_proj.weight
676
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
677
+ Params for bucket 7 (41984000 elements, 41984000 padded size):
678
+ module.decoder.layers.10.self_attention.linear_qkv.weight
679
+ module.decoder.layers.10.pre_mlp_layernorm.weight
680
+ module.decoder.layers.10.mlp.router.weight
681
+ module.decoder.layers.10.self_attention.linear_proj.weight
682
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
683
+ Params for bucket 8 (41984000 elements, 41984000 padded size):
684
+ module.decoder.layers.9.mlp.router.weight
685
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
686
+ module.decoder.layers.9.self_attention.linear_proj.weight
687
+ module.decoder.layers.9.self_attention.linear_qkv.weight
688
+ module.decoder.layers.9.pre_mlp_layernorm.weight
689
+ Params for bucket 9 (41984000 elements, 41984000 padded size):
690
+ module.decoder.layers.8.pre_mlp_layernorm.weight
691
+ module.decoder.layers.8.self_attention.linear_qkv.weight
692
+ module.decoder.layers.8.mlp.router.weight
693
+ module.decoder.layers.8.self_attention.linear_proj.weight
694
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
695
+ Params for bucket 10 (41984000 elements, 41984000 padded size):
696
+ module.decoder.layers.7.mlp.router.weight
697
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.7.self_attention.linear_proj.weight
699
+ module.decoder.layers.7.self_attention.linear_qkv.weight
700
+ module.decoder.layers.7.pre_mlp_layernorm.weight
701
+ Params for bucket 11 (41984000 elements, 41984000 padded size):
702
+ module.decoder.layers.6.pre_mlp_layernorm.weight
703
+ module.decoder.layers.6.self_attention.linear_qkv.weight
704
+ module.decoder.layers.6.mlp.router.weight
705
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
706
+ module.decoder.layers.6.self_attention.linear_proj.weight
707
+ Params for bucket 12 (41984000 elements, 41984000 padded size):
708
+ module.decoder.layers.5.mlp.router.weight
709
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
710
+ module.decoder.layers.5.self_attention.linear_qkv.weight
711
+ module.decoder.layers.5.self_attention.linear_proj.weight
712
+ module.decoder.layers.5.pre_mlp_layernorm.weight
713
+ Params for bucket 13 (41984000 elements, 41984000 padded size):
714
+ module.decoder.layers.4.mlp.router.weight
715
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
716
+ module.decoder.layers.4.self_attention.linear_proj.weight
717
+ module.decoder.layers.4.self_attention.linear_qkv.weight
718
+ module.decoder.layers.4.pre_mlp_layernorm.weight
719
+ Params for bucket 14 (41984000 elements, 41984000 padded size):
720
+ module.decoder.layers.3.mlp.router.weight
721
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
722
+ module.decoder.layers.3.self_attention.linear_proj.weight
723
+ module.decoder.layers.3.self_attention.linear_qkv.weight
724
+ module.decoder.layers.3.pre_mlp_layernorm.weight
725
+ Params for bucket 15 (41984000 elements, 41984000 padded size):
726
+ module.decoder.layers.2.mlp.router.weight
727
+ module.decoder.layers.2.self_attention.linear_proj.weight
728
+ module.decoder.layers.2.self_attention.linear_qkv.weight
729
+ module.decoder.layers.2.pre_mlp_layernorm.weight
730
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
731
+ Params for bucket 16 (41984000 elements, 41984000 padded size):
732
+ module.decoder.layers.1.pre_mlp_layernorm.weight
733
+ module.decoder.layers.1.self_attention.linear_qkv.weight
734
+ module.decoder.layers.1.mlp.router.weight
735
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
736
+ module.decoder.layers.1.self_attention.linear_proj.weight
737
+ Params for bucket 17 (41984000 elements, 41984000 padded size):
738
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
739
+ module.decoder.layers.0.mlp.router.weight
740
+ module.decoder.layers.0.self_attention.linear_proj.weight
741
+ module.decoder.layers.0.self_attention.linear_qkv.weight
742
+ module.decoder.layers.0.pre_mlp_layernorm.weight
743
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
744
+ module.embedding.word_embeddings.weight
745
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 32
746
+ Params for bucket 1 (58720256 elements, 58720256 padded size):
747
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
748
+ Params for bucket 2 (117440512 elements, 117440512 padded size):
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
750
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
751
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
752
+ Params for bucket 4 (117440512 elements, 117440512 padded size):
753
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
754
+ Params for bucket 5 (58720256 elements, 58720256 padded size):
755
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
756
+ Params for bucket 6 (117440512 elements, 117440512 padded size):
757
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
758
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
759
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
760
+ Params for bucket 8 (117440512 elements, 117440512 padded size):
761
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
762
+ Params for bucket 9 (58720256 elements, 58720256 padded size):
763
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
764
+ Params for bucket 10 (117440512 elements, 117440512 padded size):
765
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
766
+ Params for bucket 11 (58720256 elements, 58720256 padded size):
767
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
768
+ Params for bucket 12 (117440512 elements, 117440512 padded size):
769
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
770
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
771
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
772
+ Params for bucket 14 (117440512 elements, 117440512 padded size):
773
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
774
+ Params for bucket 15 (58720256 elements, 58720256 padded size):
775
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
776
+ Params for bucket 16 (117440512 elements, 117440512 padded size):
777
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
778
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
779
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
780
+ Params for bucket 18 (117440512 elements, 117440512 padded size):
781
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
782
+ Params for bucket 19 (58720256 elements, 58720256 padded size):
783
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
784
+ Params for bucket 20 (117440512 elements, 117440512 padded size):
785
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
786
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
788
+ Params for bucket 22 (117440512 elements, 117440512 padded size):
789
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
790
+ Params for bucket 23 (58720256 elements, 58720256 padded size):
791
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
792
+ Params for bucket 24 (117440512 elements, 117440512 padded size):
793
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
794
+ Params for bucket 25 (58720256 elements, 58720256 padded size):
795
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
796
+ Params for bucket 26 (117440512 elements, 117440512 padded size):
797
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
798
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
799
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
800
+ Params for bucket 28 (117440512 elements, 117440512 padded size):
801
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
802
+ Params for bucket 29 (58720256 elements, 58720256 padded size):
803
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
804
+ Params for bucket 30 (117440512 elements, 117440512 padded size):
805
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
806
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
807
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
808
+ Params for bucket 32 (117440512 elements, 117440512 padded size):
809
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
810
+ 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 0x70519d511340>, config_logger_dir='')
811
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
812
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:39:16
813
+ > building train, validation, and test datasets ...
814
+ > datasets target sizes (minimum size):
815
+ train: 192
816
+ validation: 0
817
+ test: 0
818
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
819
+ > building train, validation, and test datasets for GPT ...
820
+ 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 0x704ef6e65a00>, 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)
821
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
822
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
823
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
824
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
825
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
826
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
827
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
828
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
829
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
830
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
831
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
832
+ > finished creating GPT datasets ...
833
+ [after dataloaders are built] datetime: 2026-08-30 07:39:16
834
+ done with setup ...
835
+ training ...
836
+ (min, max) time across ranks (ms):
837
+ model-and-optimizer-setup ......................: (2794.85, 2796.81)
838
+ train/valid/test-data-iterators-setup ..........: (64.79, 76.69)
839
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
840
+ [before the start of training step] datetime: 2026-08-30 07:39:16
841
+ INFO:moonep.api:MoonEP comm buffer sizes:
842
+ config : R=8, S=4096, H=4096, E=8, K=2, NvS=8446, NvS_padded=8448
843
+ sm_count : num_sms=32, num_sms_dedup=148
844
+ hidden_buffer_size : 66.00 MiB
845
+ meta_buffer_size : 2.00 MiB
846
+ local_temp_buffer_size : 512.66 KiB
847
+ total_size : 68.50 MiB
848
+ [MoonEP selfcheck 1/4] dropped groups empty OK | covered rows [0, 8] sizes [768, 7552] | duplicated experts [[1], [-1], [4], [1], [1], [1], [4], [4]]
849
+ [MoonEP selfcheck 2/4] dropped groups empty OK | covered rows [0, 8] sizes [7808, 512] | duplicated experts [[4], [6], [4], [-1], [3], [7], [0], [4]]
850
+ [MoonEP selfcheck 3/4] dropped groups empty OK | covered rows [0, 8] sizes [8064, 256] | duplicated experts [[3], [3], [-1], [2], [0], [2], [2], [5]]
851
+ [MoonEP selfcheck 4/4] dropped groups empty OK | covered rows [0, 8] sizes [8192, 0] | duplicated experts [[-1], [0], [1], [1], [0], [1], [1], [1]]
852
+ [2026-08-30 07:39:35] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 19784.3 | throughput per GPU (TFLOP/s/GPU): 16.9 | 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 |
853
+ Number of parameters in transformer block in billions: 23.22
854
+ Number of parameters in embedding layers in billions: 0.41
855
+ Total number of parameters in billions: 23.63
856
+ Number of parameters in most loaded shard in billions: 23.6320
857
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
858
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 53350.6591796875 | max allocated: 53350.65966796875 | reserved: 58996.0 | max reserved: 58996.0
859
+ [2026-08-30 07:39:36] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 546.3 | throughput per GPU (TFLOP/s/GPU): 610.2 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.104935E+01 | loss scale: 1.0 | grad norm: 37.624 | number of skipped iterations: 0 | number of nan iterations: 0 |
860
+ [2026-08-30 07:39:37] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 528.4 | throughput per GPU (TFLOP/s/GPU): 630.9 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.679940E+00 | loss scale: 1.0 | grad norm: 42.115 | number of skipped iterations: 0 | number of nan iterations: 0 |
861
+ Capture CUDA graph for training!!!
862
+ CUDA graph capture done!!!
863
+ [2026-08-30 07:39:38] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1233.9 | throughput per GPU (TFLOP/s/GPU): 270.2 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.118468E+01 | loss scale: 1.0 | grad norm: 88.416 | number of skipped iterations: 0 | number of nan iterations: 0 |
864
+ [2026-08-30 07:39:38] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 487.0 | throughput per GPU (TFLOP/s/GPU): 684.6 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.016546E+01 | loss scale: 1.0 | grad norm: 15.300 | number of skipped iterations: 0 | number of nan iterations: 0 |
865
+ [2026-08-30 07:39:39] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 495.7 | throughput per GPU (TFLOP/s/GPU): 672.5 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.020437E+01 | loss scale: 1.0 | grad norm: 47.574 | number of skipped iterations: 0 | number of nan iterations: 0 |
866
+ [2026-08-30 07:39:39] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 493.1 | throughput per GPU (TFLOP/s/GPU): 676.0 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.891619E+00 | loss scale: 1.0 | grad norm: 7.963 | number of skipped iterations: 0 | number of nan iterations: 0 |
867
+ [2026-08-30 07:39:40] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 472.9 | throughput per GPU (TFLOP/s/GPU): 705.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.018256E+00 | loss scale: 1.0 | grad norm: 7.359 | number of skipped iterations: 0 | number of nan iterations: 0 |
868
+ [2026-08-30 07:39:40] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 479.5 | throughput per GPU (TFLOP/s/GPU): 695.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.629150E+00 | loss scale: 1.0 | grad norm: 6.207 | number of skipped iterations: 0 | number of nan iterations: 0 |
869
+ [2026-08-30 07:39:41] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 464.0 | throughput per GPU (TFLOP/s/GPU): 718.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.452185E+00 | loss scale: 1.0 | grad norm: 5.524 | number of skipped iterations: 0 | number of nan iterations: 0 |
870
+ [2026-08-30 07:39:41] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 474.8 | throughput per GPU (TFLOP/s/GPU): 702.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.339577E+00 | loss scale: 1.0 | grad norm: 4.433 | number of skipped iterations: 0 | number of nan iterations: 0 |
871
+ [2026-08-30 07:39:42] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 490.6 | throughput per GPU (TFLOP/s/GPU): 679.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.233579E+00 | loss scale: 1.0 | grad norm: 4.305 | number of skipped iterations: 0 | number of nan iterations: 0 |
872
+ [after training is done] datetime: 2026-08-30 07:39:42
873
+ [rank0]:[W830 07:39:43.485050732 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())
874
+ [rank2]:[W830 07:39:44.816458950 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())
875
+ [rank6]:[W830 07:39:44.818712890 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())
876
+ [rank4]:[W830 07:39:44.855258657 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())
877
+ [rank5]:[W830 07:39:44.070415944 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())
878
+ [rank1]:[W830 07:39:44.167987760 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())
879
+ [rank7]:[W830 07:39:44.171423753 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())
880
+ [rank3]:[W830 07:39:45.291451680 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())
881
+ W0830 07:39:48.868000 787789 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
882
+ W0830 07:39:48.868000 787789 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787973 closing signal SIGTERM
883
+ W0830 07:39:48.869000 787789 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787974 closing signal SIGTERM
884
+ W0830 07:39:48.869000 787789 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787975 closing signal SIGTERM
885
+ W0830 07:39:48.870000 787789 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787976 closing signal SIGTERM
886
+ W0830 07:39:48.870000 787789 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787979 closing signal SIGTERM
887
+ W0830 07:39:48.870000 787789 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787980 closing signal SIGTERM
888
+ Traceback (most recent call last):
889
+ File "/usr/local/bin/torchrun", line 8, in <module>
890
+ sys.exit(main())
891
+ ^^^^^^
892
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
893
+ return f(*args, **kwargs)
894
+ ^^^^^^^^^^^^^^^^^^
895
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
896
+ run(args)
897
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
898
+ elastic_launch(
899
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
900
+ return launch_agent(self._config, self._entrypoint, list(args))
901
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
902
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
903
+ result = agent.run()
904
+ ^^^^^^^^^^^
905
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
906
+ result = f(*args, **kwargs)
907
+ ^^^^^^^^^^^^^^^^^^
908
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
909
+ result = self._invoke_run(role)
910
+ ^^^^^^^^^^^^^^^^^^^^^^
911
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
912
+ time.sleep(monitor_interval)
913
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
914
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
915
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 787789 got signal: 15
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep3/mem_samples.csv ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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, 4
10
+ 1, 4
11
+ 2, 1
12
+ 3, 1
13
+ 4, 1
14
+ 5, 1
15
+ 6, 1
16
+ 7, 1
17
+ 0, 4
18
+ 1, 4
19
+ 2, 4
20
+ 3, 4
21
+ 4, 4
22
+ 5, 4
23
+ 6, 4
24
+ 7, 1
25
+ 0, 1
26
+ 1, 1
27
+ 2, 1
28
+ 3, 1
29
+ 4, 1
30
+ 5, 1
31
+ 6, 3
32
+ 7, 4
33
+ 0, 0
34
+ 1, 0
35
+ 2, 0
36
+ 3, 0
37
+ 4, 0
38
+ 5, 0
39
+ 6, 0
40
+ 7, 0
41
+ 0, 1
42
+ 1, 1
43
+ 2, 1
44
+ 3, 1
45
+ 4, 1
46
+ 5, 1
47
+ 6, 1
48
+ 7, 1
49
+ 0, 4
50
+ 1, 4
51
+ 2, 4
52
+ 3, 4
53
+ 4, 1
54
+ 5, 1
55
+ 6, 1
56
+ 7, 1
57
+ 0, 4
58
+ 1, 4
59
+ 2, 4
60
+ 3, 4
61
+ 4, 4
62
+ 5, 4
63
+ 6, 4
64
+ 7, 4
65
+ 0, 4
66
+ 1, 629
67
+ 2, 629
68
+ 3, 629
69
+ 4, 629
70
+ 5, 35
71
+ 6, 629
72
+ 7, 629
73
+ 0, 901
74
+ 1, 911
75
+ 2, 913
76
+ 3, 897
77
+ 4, 917
78
+ 5, 919
79
+ 6, 911
80
+ 7, 917
81
+ 0, 1895
82
+ 1, 1895
83
+ 2, 1895
84
+ 3, 1895
85
+ 4, 1895
86
+ 5, 1895
87
+ 6, 1895
88
+ 7, 1895
89
+ 0, 37267
90
+ 1, 37259
91
+ 2, 37271
92
+ 3, 37263
93
+ 4, 37267
94
+ 5, 37257
95
+ 6, 37261
96
+ 7, 37267
97
+ 0, 46111
98
+ 1, 46097
99
+ 2, 46045
100
+ 3, 46095
101
+ 4, 46075
102
+ 5, 46115
103
+ 6, 46089
104
+ 7, 46119
105
+ 0, 46415
106
+ 1, 46419
107
+ 2, 46431
108
+ 3, 46439
109
+ 4, 46473
110
+ 5, 46451
111
+ 6, 46427
112
+ 7, 46461
113
+ 0, 46919
114
+ 1, 46917
115
+ 2, 46917
116
+ 3, 46917
117
+ 4, 46919
118
+ 5, 46917
119
+ 6, 46917
120
+ 7, 46917
121
+ 0, 47383
122
+ 1, 47381
123
+ 2, 47381
124
+ 3, 47381
125
+ 4, 47381
126
+ 5, 47381
127
+ 6, 47381
128
+ 7, 47381
129
+ 0, 66255
130
+ 1, 66253
131
+ 2, 66253
132
+ 3, 66253
133
+ 4, 66253
134
+ 5, 66253
135
+ 6, 66253
136
+ 7, 66253
137
+ 0, 67185
138
+ 1, 67179
139
+ 2, 67177
140
+ 3, 67211
141
+ 4, 67189
142
+ 5, 67207
143
+ 6, 67185
144
+ 7, 67201
145
+ 0, 68247
146
+ 1, 68263
147
+ 2, 68241
148
+ 3, 68247
149
+ 4, 68271
150
+ 5, 68267
151
+ 6, 68283
152
+ 7, 68245
153
+ 0, 82139
154
+ 1, 82181
155
+ 2, 82149
156
+ 3, 82189
157
+ 4, 82165
158
+ 5, 82171
159
+ 6, 82203
160
+ 7, 82193
161
+ 0, 96849
162
+ 1, 96843
163
+ 2, 96843
164
+ 3, 96843
165
+ 4, 96843
166
+ 5, 96843
167
+ 6, 96843
168
+ 7, 96843
169
+ 0, 96849
170
+ 1, 96843
171
+ 2, 96843
172
+ 3, 96843
173
+ 4, 96843
174
+ 5, 96843
175
+ 6, 96843
176
+ 7, 96843
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep3/run.env ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=moonep
2
+ config=G1
3
+ family=tierA
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=moonep_backend_async_grad_reduce
6
+ commit=2e68c8ad
7
+ moonep_source_dir=/workspace/MoonEP
8
+ moonep_source_branch=megatron-ddp-symmetric-params
9
+ moonep_source_commit=d8bc666
10
+ num_layers=16
11
+ hidden=4096
12
+ ffn_hidden=14336
13
+ moe_ffn_hidden=14336
14
+ num_experts=8
15
+ topk=2
16
+ experts_per_rank=1
17
+ expert_params=176160768
18
+ heads=32
19
+ kv_groups=8
20
+ kv_channels=
21
+ qk_layernorm=0
22
+ shared_ffn=0
23
+ shared_expert_overlap=0
24
+ moe_token_dispatcher=flex
25
+ mla=0
26
+ dense_first_layers=0
27
+ moe_layers=16
28
+ rotary_base=1000000
29
+ rotary_scaling=
30
+ qkv_bias=0
31
+ router_balancing=aux_loss
32
+ aux_loss_coeff=1e-2
33
+ router_pre_softmax=0
34
+ router_fusion=0
35
+ router_dtype=fp32
36
+ device_initiated_grouped_gemm=1
37
+ norm_epsilon=1e-5
38
+ seq_length=4096
39
+ micro_batch_size=1
40
+ global_batch_size=16
41
+ num_microbatches=2
42
+ assignments_per_rank=8192
43
+ mean_tokens_per_expert=8192
44
+ slot_policy=one
45
+ slots_per_rank=1
46
+ capacity_mode=none
47
+ capacity=none
48
+ cuda_graph=1
49
+ probe_mode=0
50
+ async_grad_reduce=1
51
+ branch_role=timing
52
+ memory_dump_mode=none
53
+ memory_snapshot_path=
54
+ cuda_graph_warmup_steps=3
55
+ recompute=none
56
+ force_balance=0
57
+ train_iters=12
58
+ ep_size=8
59
+ echo_dispatch_overlap=1
60
+ echo_dispatch_recompute=0
61
+ ultraep_quota_min_tokens=
62
+ ultraep_weight_sync_join=
63
+ profile_enable=0
64
+ kernel_trace=0
65
+ fp8_mode=mxfp8
66
+ fp8_recipe=mxfp8
67
+ training_seed=1234
68
+ learning_rate=1e-4
69
+ min_learning_rate=1.0e-5
70
+ lr_warmup_iters=5
71
+ state_dump_enable=0
72
+ state_dump_dir=
73
+ moonep_assert_no_pending_reduce=1
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep3/train.log ADDED
@@ -0,0 +1,867 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G1 / moonep (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=none cap=none cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep3/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 2e68c8ad
8
+ ==============================================================
9
+ W0830 07:40:02.348000 790344 torch/distributed/run.py:766]
10
+ W0830 07:40:02.348000 790344 torch/distributed/run.py:766] *****************************************
11
+ W0830 07:40:02.348000 790344 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 07:40:02.348000 790344 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-MoE-Echo/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-MoE-Echo/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
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
19
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
20
+ /workspace/Megatron-MoE-Echo/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
+ /workspace/Megatron-MoE-Echo/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-MoE-Echo/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
+ 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
27
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
28
+ Number of virtual stages per pipeline stage: None
29
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
30
+ using torch.bfloat16 for parameters ...
31
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
32
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
33
+ ------------------------ arguments ------------------------
34
+ account_for_embedding_in_pipeline_split ......... False
35
+ account_for_loss_in_pipeline_split .............. False
36
+ accumulate_allreduce_grads_in_fp32 .............. True
37
+ adam_beta1 ...................................... 0.9
38
+ adam_beta2 ...................................... 0.999
39
+ adam_eps ........................................ 1e-08
40
+ add_bias_linear ................................. False
41
+ add_position_embedding .......................... False
42
+ add_qkv_bias .................................... False
43
+ adlr_autoresume ................................. False
44
+ adlr_autoresume_interval ........................ 1000
45
+ align_grad_reduce ............................... True
46
+ align_param_gather .............................. False
47
+ app_tag_run_name ................................ None
48
+ app_tag_run_version ............................. 0.0.0
49
+ apply_layernorm_1p .............................. False
50
+ apply_query_key_layer_scaling ................... False
51
+ apply_residual_connection_post_layernorm ........ False
52
+ apply_rope_fusion ............................... True
53
+ async_save ...................................... None
54
+ async_tensor_model_parallel_allreduce ........... True
55
+ attention_backend ............................... AttnBackend.auto
56
+ attention_dropout ............................... 0.0
57
+ attention_softmax_in_fp32 ....................... False
58
+ auto_detect_ckpt_format ......................... False
59
+ barrier_with_L1_time ............................ True
60
+ bert_binary_head ................................ True
61
+ bert_embedder_type .............................. megatron
62
+ bert_load ....................................... None
63
+ bf16 ............................................ True
64
+ bias_dropout_fusion ............................. True
65
+ bias_gelu_fusion ................................ False
66
+ bias_swiglu_fusion .............................. True
67
+ biencoder_projection_dim ........................ 0
68
+ biencoder_shared_query_context_model ............ False
69
+ block_data_path ................................. None
70
+ calc_ft_timeouts ................................ False
71
+ calculate_per_token_loss ........................ False
72
+ check_for_large_grads ........................... False
73
+ check_for_nan_in_loss_and_grad .................. False
74
+ check_for_spiky_loss ............................ False
75
+ check_weight_hash_across_dp_replicas_interval ... None
76
+ ckpt_assume_constant_structure .................. False
77
+ ckpt_convert_format ............................. None
78
+ ckpt_convert_save ............................... None
79
+ ckpt_convert_update_legacy_dist_opt_format ...... False
80
+ ckpt_format ..................................... torch_dist
81
+ ckpt_fully_parallel_load ........................ False
82
+ ckpt_fully_parallel_save ........................ True
83
+ ckpt_fully_parallel_save_deprecated ............. False
84
+ ckpt_step ....................................... None
85
+ classes_fraction ................................ 1.0
86
+ clip_grad ....................................... 1.0
87
+ clone_scatter_output_in_embedding ............... True
88
+ config_logger_dir ...............................
89
+ consumed_train_samples .......................... 0
90
+ consumed_valid_samples .......................... 0
91
+ context_parallel_size ........................... 1
92
+ cp_comm_type .................................... ['p2p']
93
+ create_attention_mask_in_dataloader ............. True
94
+ cross_entropy_fusion_impl ....................... native
95
+ cross_entropy_loss_fusion ....................... False
96
+ cuda_graph_scope ................................ full_iteration
97
+ cuda_graph_warmup_steps ......................... 3
98
+ data_args_path .................................. None
99
+ data_cache_path ................................. None
100
+ data_parallel_random_init ....................... False
101
+ data_parallel_sharding_strategy ................. no_shard
102
+ data_parallel_size .............................. 8
103
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
104
+ data_per_class_fraction ......................... 1.0
105
+ data_sharding ................................... True
106
+ dataloader_type ................................. single
107
+ ddp_average_in_collective ....................... False
108
+ ddp_bucket_size ................................. None
109
+ ddp_num_buckets ................................. None
110
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
111
+ decoder_first_pipeline_num_layers ............... None
112
+ decoder_last_pipeline_num_layers ................ None
113
+ decoder_num_layers .............................. None
114
+ decoder_seq_length .............................. None
115
+ decoupled_lr .................................... None
116
+ decoupled_min_lr ................................ None
117
+ decrease_batch_size_if_needed ................... False
118
+ defer_embedding_wgrad_compute ................... False
119
+ delay_wgrad_compute ............................. False
120
+ deprecated_use_mcore_models ..................... True
121
+ deterministic_mode .............................. False
122
+ dino_bottleneck_size ............................ 256
123
+ dino_freeze_last_layer .......................... 1
124
+ dino_head_hidden_size ........................... 2048
125
+ dino_local_crops_number ......................... 10
126
+ dino_local_img_size ............................. 96
127
+ dino_norm_last_layer ............................ False
128
+ dino_teacher_temp ............................... 0.07
129
+ dino_warmup_teacher_temp ........................ 0.04
130
+ dino_warmup_teacher_temp_epochs ................. 30
131
+ disable_bf16_reduced_precision_matmul ........... False
132
+ disable_mamba_mem_eff_path ...................... False
133
+ disable_straggler_on_startup .................... False
134
+ dist_ckpt_format_deprecated ..................... None
135
+ dist_ckpt_strictness ............................ assume_ok_unexpected
136
+ distribute_saved_activations .................... False
137
+ distributed_backend ............................. nccl
138
+ distributed_timeout_minutes ..................... 10
139
+ embedding_init_method_std ....................... None
140
+ embedding_path .................................. None
141
+ empty_unused_memory_level ....................... 0
142
+ enable_cuda_graph ............................... True
143
+ enable_experimental ............................. False
144
+ enable_ft_package ............................... False
145
+ enable_gloo_process_groups ...................... True
146
+ enable_msc ...................................... True
147
+ enable_one_logger ............................... True
148
+ encoder_num_layers .............................. 16
149
+ encoder_seq_length .............................. 4096
150
+ end_weight_decay ................................ 0.1
151
+ eod_mask_loss ................................... False
152
+ error_injection_rate ............................ 0
153
+ error_injection_type ............................ transient_error
154
+ eval_interval ................................... 1000000
155
+ eval_iters ...................................... 0
156
+ evidence_data_path .............................. None
157
+ exit_duration_in_mins ........................... None
158
+ exit_interval ................................... None
159
+ exit_on_missing_checkpoint ...................... False
160
+ exit_signal_handler ............................. False
161
+ exp_avg_dtype ................................... torch.float32
162
+ exp_avg_sq_dtype ................................ torch.float32
163
+ expert_model_parallel_size ...................... 8
164
+ expert_tensor_parallel_size ..................... 1
165
+ export_force_local_attention .................... False
166
+ export_kd_cfg ................................... None
167
+ export_kd_teacher_ckpt_format ................... None
168
+ export_kd_teacher_load .......................... None
169
+ export_kv_cache_quant ........................... False
170
+ export_legacy_megatron .......................... False
171
+ export_model_type ............................... GPTModel
172
+ export_moe_apply_probs_on_input ................. False
173
+ export_qk_l2_norm ............................... False
174
+ export_quant_cfg ................................ None
175
+ export_real_quant_cfg ........................... None
176
+ export_te_mcore_model ........................... False
177
+ external_cuda_graph ............................. False
178
+ ffn_hidden_size ................................. 14336
179
+ finetune ........................................ False
180
+ finetune_data_split ............................. train
181
+ finetune_hf_dataset ............................. None
182
+ first_last_layers_bf16 .......................... False
183
+ flash_decode .................................... False
184
+ fp16 ............................................ False
185
+ fp16_lm_cross_entropy ........................... False
186
+ fp32_residual_connection ........................ False
187
+ fp8 ............................................. e4m3
188
+ fp8_amax_compute_algo ........................... most_recent
189
+ fp8_amax_history_len ............................ 1
190
+ fp8_interval .................................... 1
191
+ fp8_margin ...................................... 0
192
+ fp8_param_gather ................................ False
193
+ fp8_recipe ...................................... mxfp8
194
+ fp8_wgrad ....................................... True
195
+ fsdp_double_buffer .............................. False
196
+ global_batch_size ............................... 16
197
+ grad_reduce_in_bf16 ............................. False
198
+ gradient_accumulation_fusion .................... True
199
+ gradient_reduce_div_fusion ...................... True
200
+ group_query_attention ........................... True
201
+ head_lr_mult .................................... 1.0
202
+ heterogeneous_layers_config_encoded_json ........ None
203
+ heterogeneous_layers_config_path ................ None
204
+ hidden_dropout .................................. 0.0
205
+ hidden_size ..................................... 4096
206
+ hierarchical_context_parallel_sizes ............. None
207
+ high_priority_stream_groups ..................... []
208
+ hybrid_attention_ratio .......................... 0.0
209
+ hybrid_mlp_ratio ................................ 0.0
210
+ hybrid_override_pattern ......................... None
211
+ hysteresis ...................................... 2
212
+ ict_head_size ................................... None
213
+ ict_load ........................................ None
214
+ img_h ........................................... 224
215
+ img_w ........................................... 224
216
+ indexer_batch_size .............................. 128
217
+ indexer_log_interval ............................ 1000
218
+ inference_batch_times_seqlen_threshold .......... -1
219
+ inference_dynamic_batching ...................... False
220
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
221
+ inference_dynamic_batching_buffer_overflow_factor None
222
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
223
+ inference_dynamic_batching_chunk_size ........... 256
224
+ inference_dynamic_batching_max_requests_override None
225
+ inference_dynamic_batching_max_tokens_override .. None
226
+ inference_dynamic_batching_num_cuda_graphs ...... 16
227
+ inference_max_batch_size ........................ 8
228
+ inference_max_seq_length ........................ 2560
229
+ inference_rng_tracker ........................... False
230
+ init_method_std ................................. 0.01
231
+ init_method_xavier_uniform ...................... False
232
+ init_model_with_meta_device ..................... False
233
+ initial_loss_scale .............................. 4294967296
234
+ inprocess_active_world_size ..................... 8
235
+ inprocess_barrier_timeout ....................... 120
236
+ inprocess_completion_timeout .................... 120
237
+ inprocess_empty_cuda_cache ...................... False
238
+ inprocess_granularity ........................... node
239
+ inprocess_hard_timeout .......................... 90
240
+ inprocess_heartbeat_interval .................... 30
241
+ inprocess_heartbeat_timeout ..................... 60
242
+ inprocess_last_call_wait ........................ 1
243
+ inprocess_max_iterations ........................ None
244
+ inprocess_monitor_process_interval .............. 1.0
245
+ inprocess_monitor_thread_interval ............... 1.0
246
+ inprocess_progress_watchdog_interval ............ 1.0
247
+ inprocess_restart ............................... False
248
+ inprocess_soft_timeout .......................... 60
249
+ inprocess_termination_grace_time ................ 1
250
+ is_hybrid_model ................................. False
251
+ iter_per_epoch .................................. 1250
252
+ iterations_to_skip .............................. []
253
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
254
+ kitchen_config_file ............................. None
255
+ kitchen_recipe_number ........................... None
256
+ kv_channels ..................................... 128
257
+ kv_lora_rank .................................... 32
258
+ lazy_mpu_init ................................... None
259
+ load ............................................ None
260
+ load_main_params_from_ckpt ...................... None
261
+ load_model_opt_format ........................... False
262
+ local_rank ...................................... 0
263
+ log_energy ...................................... False
264
+ log_interval .................................... 1
265
+ log_loss_scale_to_tensorboard ................... True
266
+ log_memory_to_tensorboard ....................... False
267
+ log_num_zeros_in_grad ........................... False
268
+ log_params_norm ................................. False
269
+ log_progress .................................... False
270
+ log_straggler ................................... False
271
+ log_throughput .................................. True
272
+ log_timers_to_tensorboard ....................... False
273
+ log_validation_ppl_to_tensorboard ............... False
274
+ log_world_size_to_tensorboard ................... False
275
+ logging_level ................................... None
276
+ loss_scale ...................................... None
277
+ loss_scale_window ............................... 1000
278
+ lr .............................................. 0.0001
279
+ lr_decay_iters .................................. 8
280
+ lr_decay_samples ................................ None
281
+ lr_decay_style .................................. cosine
282
+ lr_warmup_fraction .............................. None
283
+ lr_warmup_init .................................. 0.0
284
+ lr_warmup_iters ................................. 5
285
+ lr_warmup_samples ............................... 0
286
+ lr_wsd_decay_iters .............................. None
287
+ lr_wsd_decay_samples ............................ None
288
+ lr_wsd_decay_style .............................. exponential
289
+ main_grads_dtype ................................ torch.float32
290
+ main_params_dtype ............................... torch.float32
291
+ make_vocab_size_divisible_by .................... 128
292
+ mamba_head_dim .................................. 64
293
+ mamba_num_groups ................................ 8
294
+ mamba_num_heads ................................. None
295
+ mamba_state_dim ................................. 128
296
+ manual_gc ....................................... False
297
+ manual_gc_eval .................................. True
298
+ manual_gc_interval .............................. 0
299
+ mask_factor ..................................... 1.0
300
+ mask_prob ....................................... 0.15
301
+ mask_type ....................................... random
302
+ masked_softmax_fusion ........................... False
303
+ max_position_embeddings ......................... 32768
304
+ max_tokens_to_oom ............................... 12000
305
+ memory_snapshot_path ............................ snapshot.pickle
306
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
307
+ micro_batch_size ................................ 1
308
+ microbatch_group_size_per_vp_stage .............. None
309
+ mid_level_dataset_surplus ....................... 0.005
310
+ min_loss_scale .................................. 1.0
311
+ min_lr .......................................... 1e-05
312
+ mlp_chunks_for_prefill .......................... 1
313
+ mmap_bin_files .................................. True
314
+ mock_data ....................................... False
315
+ moe_apply_probs_on_input ........................ False
316
+ moe_aux_loss_coeff .............................. 0.01
317
+ moe_deepep_num_sms .............................. 20
318
+ moe_echo_enable_random_offloading ............... False
319
+ moe_echo_expert_dispatch_overlap ................ False
320
+ moe_echo_expert_dispatcher_type ................. hybridep
321
+ moe_echo_recompute_expert_dispatch .............. False
322
+ moe_enable_deepep ............................... False
323
+ moe_enable_echo ................................. False
324
+ moe_enable_hybridep ............................. False
325
+ moe_enable_moonep ............................... True
326
+ moe_expert_capacity_factor ...................... None
327
+ moe_extended_tp ................................. False
328
+ moe_ffn_hidden_size ............................. 14336
329
+ moe_grouped_gemm ................................ True
330
+ moe_hybridep_num_sms ............................ 16
331
+ moe_input_jitter_eps ............................ None
332
+ moe_layer_freq .................................. 1
333
+ moe_layer_recompute ............................. False
334
+ moe_moonep_async_grad_reduce .................... True
335
+ moe_moonep_num_sms .............................. None
336
+ moe_num_echo_experts ............................ None
337
+ moe_pad_expert_input_to_capacity ................ False
338
+ moe_per_layer_logging ........................... False
339
+ moe_permute_fusion .............................. False
340
+ moe_received_token_capacity ..................... None
341
+ moe_router_bias_update_rate ..................... 0.001
342
+ moe_router_dtype ................................ fp32
343
+ moe_router_enable_expert_bias ................... False
344
+ moe_router_force_load_balancing ................. False
345
+ moe_router_group_topk ........................... None
346
+ moe_router_load_balancing_type .................. aux_loss
347
+ moe_router_num_groups ........................... None
348
+ moe_router_padding_for_fp8 ...................... False
349
+ moe_router_pre_softmax .......................... False
350
+ moe_router_score_function ....................... softmax
351
+ moe_router_topk ................................. 2
352
+ moe_router_topk_scaling_factor .................. None
353
+ moe_shared_expert_intermediate_size ............. None
354
+ moe_shared_expert_overlap ....................... False
355
+ moe_token_dispatcher_type ....................... flex
356
+ moe_token_drop_policy ........................... probs
357
+ moe_upcycling_granularity ....................... 1
358
+ moe_use_device_initiated_grouped_gemm ........... True
359
+ moe_use_legacy_grouped_gemm ..................... False
360
+ moe_use_upcycling ............................... False
361
+ moe_z_loss_coeff ................................ None
362
+ mrope_section ................................... None
363
+ mscale .......................................... 1.0
364
+ mscale_all_dim .................................. 0.0
365
+ mtp_loss_scaling_factor ......................... 0.1
366
+ mtp_num_layers .................................. None
367
+ multi_latent_attention .......................... False
368
+ nccl_all_reduce_for_prefill ..................... False
369
+ nccl_communicator_config_path ................... None
370
+ nccl_ub ......................................... False
371
+ no_load_optim ................................... None
372
+ no_load_rng ..................................... None
373
+ no_persist_layer_norm ........................... False
374
+ no_rope_freq .................................... None
375
+ no_save_optim ................................... None
376
+ no_save_rng ..................................... None
377
+ non_persistent_ckpt_type ........................ None
378
+ non_persistent_global_ckpt_dir .................. None
379
+ non_persistent_local_ckpt_algo .................. fully_parallel
380
+ non_persistent_local_ckpt_dir ................... None
381
+ non_persistent_save_interval .................... None
382
+ norm_epsilon .................................... 1e-05
383
+ normalization ................................... RMSNorm
384
+ num_attention_heads ............................. 32
385
+ num_channels .................................... 3
386
+ num_classes ..................................... 1000
387
+ num_dataset_builder_threads ..................... 1
388
+ num_distributed_optimizer_instances ............. 1
389
+ num_experts ..................................... 8
390
+ num_layers ...................................... 16
391
+ num_layers_at_end_in_bf16 ....................... 1
392
+ num_layers_at_start_in_bf16 ..................... 1
393
+ num_layers_per_virtual_pipeline_stage ........... None
394
+ num_query_groups ................................ 8
395
+ num_virtual_stages_per_pipeline_rank ............ None
396
+ num_workers ..................................... 2
397
+ object_storage_cache_path ....................... None
398
+ one_logger_async ................................ False
399
+ one_logger_project .............................. megatron-lm
400
+ one_logger_run_name ............................. None
401
+ onnx_safe ....................................... None
402
+ openai_gelu ..................................... False
403
+ optimizer ....................................... adam
404
+ optimizer_cpu_offload ........................... False
405
+ optimizer_offload_fraction ...................... 1.0
406
+ output_bert_embeddings .......................... False
407
+ overlap_cpu_optimizer_d2h_h2d ................... False
408
+ overlap_grad_reduce ............................. True
409
+ overlap_moe_expert_parallel_comm ................ False
410
+ overlap_p2p_comm ................................ False
411
+ overlap_p2p_comm_warmup_flush ................... False
412
+ overlap_param_gather ............................ True
413
+ overlap_param_gather_with_optimizer_step ........ False
414
+ override_opt_param_scheduler .................... False
415
+ params_dtype .................................... torch.bfloat16
416
+ patch_dim ....................................... 16
417
+ per_split_data_args_path ........................ None
418
+ perform_initialization .......................... True
419
+ pin_cpu_grads ................................... True
420
+ pin_cpu_params .................................. True
421
+ pipeline_model_parallel_comm_backend ............ None
422
+ pipeline_model_parallel_layout .................. None
423
+ pipeline_model_parallel_size .................... 1
424
+ position_embedding_type ......................... rope
425
+ pretrained_checkpoint ........................... None
426
+ profile ......................................... False
427
+ profile_ranks ................................... [0]
428
+ profile_step_end ................................ 12
429
+ profile_step_start .............................. 10
430
+ q_lora_rank ..................................... None
431
+ qk_head_dim ..................................... 128
432
+ qk_l2_norm ...................................... False
433
+ qk_layernorm .................................... False
434
+ qk_pos_emb_head_dim ............................. 64
435
+ query_in_block_prob ............................. 0.1
436
+ rampup_batch_size ............................... None
437
+ rank ............................................ 0
438
+ recompute_granularity ........................... None
439
+ recompute_method ................................ None
440
+ recompute_modules ............................... None
441
+ recompute_num_layers ............................ None
442
+ record_memory_history ........................... False
443
+ relative_attention_max_distance ................. 128
444
+ relative_attention_num_buckets .................. 32
445
+ replication ..................................... False
446
+ replication_factor .............................. 2
447
+ replication_jump ................................ None
448
+ rerun_mode ...................................... validate_results
449
+ reset_attention_mask ............................ False
450
+ reset_position_ids .............................. False
451
+ result_rejected_tracker_filename ................ None
452
+ retriever_report_topk_accuracies ................ []
453
+ retriever_score_scaling ......................... False
454
+ retriever_seq_length ............................ 256
455
+ retro_add_retriever ............................. False
456
+ retro_attention_gate ............................ 1
457
+ retro_cyclic_train_iters ........................ None
458
+ retro_encoder_attention_dropout ................. 0.1
459
+ retro_encoder_hidden_dropout .................... 0.1
460
+ retro_encoder_layers ............................ 2
461
+ retro_num_neighbors ............................. 2
462
+ retro_num_retrieved_chunks ...................... 2
463
+ retro_project_dir ............................... None
464
+ retro_verify_neighbor_count ..................... True
465
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
466
+ rope_scaling_factor ............................. 8.0
467
+ rope_type ....................................... None
468
+ rotary_base ..................................... 1000000
469
+ rotary_interleaved .............................. False
470
+ rotary_percent .................................. 1.0
471
+ rotary_scaling_factor ........................... 1.0
472
+ rotary_seq_len_interpolation_factor ............. None
473
+ run_workload_inspector_server ................... False
474
+ sample_rate ..................................... 1.0
475
+ save ............................................ None
476
+ save_interval ................................... None
477
+ save_retain_interval ............................ None
478
+ scatter_gather_tensors_in_pipeline .............. True
479
+ seed ............................................ 1234
480
+ seq_length ...................................... 4096
481
+ sequence_parallel ............................... False
482
+ sft ............................................. False
483
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
484
+ sgd_momentum .................................... 0.9
485
+ short_seq_prob .................................. 0.1
486
+ skip_train ...................................... False
487
+ skipped_train_samples ........................... 0
488
+ spec ............................................ None
489
+ split ........................................... 100,0,0
490
+ squared_relu .................................... False
491
+ start_weight_decay .............................. 0.1
492
+ straggler_ctrlr_port ............................ 65535
493
+ straggler_minmax_count .......................... 1
494
+ suggested_communication_unit_size ............... None
495
+ swiglu .......................................... True
496
+ swin_backbone_type .............................. tiny
497
+ symmetric_ar_type ............................... None
498
+ te_rng_tracker .................................. True
499
+ tensor_model_parallel_size ...................... 1
500
+ tensorboard_dir ................................. None
501
+ tensorboard_log_interval ........................ 1
502
+ tensorboard_queue_size .......................... 1000
503
+ test_data_path .................................. None
504
+ test_mode ....................................... False
505
+ tiktoken_num_special_tokens ..................... 1000
506
+ tiktoken_pattern ................................ None
507
+ tiktoken_special_tokens ......................... None
508
+ timing_log_level ................................ 0
509
+ timing_log_option ............................... minmax
510
+ titles_data_path ................................ None
511
+ tokenizer_model ................................. None
512
+ tokenizer_type .................................. GPT2BPETokenizer
513
+ torch_fsdp2_reshard_after_forward ............... True
514
+ tp_comm_bootstrap_backend ....................... nccl
515
+ tp_comm_bulk_dgrad .............................. True
516
+ tp_comm_bulk_wgrad .............................. True
517
+ tp_comm_overlap ................................. False
518
+ tp_comm_overlap_ag .............................. True
519
+ tp_comm_overlap_cfg ............................. None
520
+ tp_comm_overlap_rs .............................. True
521
+ tp_comm_overlap_rs_dgrad ........................ False
522
+ tp_comm_split_ag ................................ True
523
+ tp_comm_split_rs ................................ True
524
+ train_data_path ................................. None
525
+ train_iters ..................................... 12
526
+ train_samples ................................... None
527
+ train_sync_interval ............................. None
528
+ transformer_impl ................................ transformer_engine
529
+ transformer_pipeline_model_parallel_size ........ 1
530
+ untie_embeddings_and_output_weights ............. True
531
+ use_checkpoint_args ............................. False
532
+ use_checkpoint_opt_param_scheduler .............. False
533
+ use_cpu_initialization .......................... None
534
+ use_custom_fsdp ................................. False
535
+ use_dist_ckpt ................................... True
536
+ use_dist_ckpt_deprecated ........................ False
537
+ use_distributed_optimizer ....................... True
538
+ use_flash_attn .................................. False
539
+ use_legacy_models ............................... False
540
+ use_mp_args_from_checkpoint_args ................ False
541
+ use_one_sent_docs ............................... False
542
+ use_persistent_ckpt_worker ...................... False
543
+ use_precision_aware_optimizer ................... False
544
+ use_pytorch_profiler ............................ False
545
+ use_ring_exchange_p2p ........................... False
546
+ use_rope_scaling ................................ False
547
+ use_rotary_position_embeddings .................. False
548
+ use_sharp ....................................... False
549
+ use_tokenizer_model_from_checkpoint_args ........ True
550
+ use_torch_fsdp2 ................................. False
551
+ use_torch_optimizer_for_cpu_offload ............. False
552
+ use_tp_pp_dp_mapping ............................ False
553
+ v_head_dim ...................................... 128
554
+ valid_data_path ................................. None
555
+ variable_seq_lengths ............................ False
556
+ virtual_pipeline_model_parallel_size ............ None
557
+ vision_backbone_type ............................ vit
558
+ vision_pretraining .............................. False
559
+ vision_pretraining_type ......................... classify
560
+ vocab_extra_ids ................................. 0
561
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
562
+ vocab_size ...................................... None
563
+ wandb_exp_name ..................................
564
+ wandb_project ...................................
565
+ wandb_save_dir ..................................
566
+ weight_decay .................................... 0.1
567
+ weight_decay_incr_style ......................... constant
568
+ wgrad_deferral_limit ............................ 0
569
+ world_size ...................................... 8
570
+ yaml_cfg ........................................ None
571
+ -------------------- end of arguments ---------------------
572
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
573
+ > building GPT2BPETokenizer tokenizer ...
574
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
575
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
576
+ > initializing torch distributed ...
577
+ /workspace/Megatron-MoE-Echo/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
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
580
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
582
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 6 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 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 2 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 4 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 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
596
+ > initialized tensor model parallel with size 1
597
+ > initialized pipeline model parallel with size 1
598
+ > setting random seeds to 1234 ...
599
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
600
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
601
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
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
+ [rank5]:[W830 07:40:20.651538843 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.
607
+ [rank6]:[W830 07:40:20.651880956 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.
608
+ [rank4]:[W830 07:40:20.651928834 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.
609
+ [rank2]:[W830 07:40:20.651949539 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.
610
+ [rank3]:[W830 07:40:20.656766426 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
+ [rank1]:[W830 07:40:20.660258243 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
+ [rank7]:[W830 07:40:20.663608220 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.
613
+ > compiling dataset index builder ...
614
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
615
+ make: Nothing to be done for 'default'.
616
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
617
+ >>> done with dataset index builder. Compilation time: 0.034 seconds
618
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
619
+ > compiling and loading fused kernels ...
620
+ [rank0]:[W830 07:40:20.823935387 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.
621
+ >>> done with compiling and loading fused kernels. Compilation time: 3.851 seconds
622
+ time to initialize megatron (seconds): 7.263
623
+ [after megatron is initialized] datetime: 2026-08-30 07:40:26
624
+ building GPT model ...
625
+ /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.
626
+ warnings.warn(
627
+ /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.
628
+ warnings.warn(
629
+ /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.
630
+ warnings.warn(
631
+ /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.
632
+ warnings.warn(
633
+ /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.
634
+ warnings.warn(
635
+ /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.
636
+ warnings.warn(
637
+ /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.
638
+ warnings.warn(
639
+ /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.
640
+ warnings.warn(
641
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3902410752
642
+ 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)
643
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
644
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
645
+ module.output_layer.weight
646
+ Params for bucket 2 (41988096 elements, 41988096 padded size):
647
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
648
+ module.decoder.layers.15.mlp.router.weight
649
+ module.decoder.layers.15.self_attention.linear_proj.weight
650
+ module.decoder.final_layernorm.weight
651
+ module.decoder.layers.15.pre_mlp_layernorm.weight
652
+ module.decoder.layers.15.self_attention.linear_qkv.weight
653
+ Params for bucket 3 (41984000 elements, 41984000 padded size):
654
+ module.decoder.layers.14.pre_mlp_layernorm.weight
655
+ module.decoder.layers.14.self_attention.linear_qkv.weight
656
+ module.decoder.layers.14.mlp.router.weight
657
+ module.decoder.layers.14.self_attention.linear_proj.weight
658
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
659
+ Params for bucket 4 (41984000 elements, 41984000 padded size):
660
+ module.decoder.layers.13.pre_mlp_layernorm.weight
661
+ module.decoder.layers.13.self_attention.linear_qkv.weight
662
+ module.decoder.layers.13.mlp.router.weight
663
+ module.decoder.layers.13.self_attention.linear_proj.weight
664
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
665
+ Params for bucket 5 (41984000 elements, 41984000 padded size):
666
+ module.decoder.layers.12.mlp.router.weight
667
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
668
+ module.decoder.layers.12.self_attention.linear_proj.weight
669
+ module.decoder.layers.12.self_attention.linear_qkv.weight
670
+ module.decoder.layers.12.pre_mlp_layernorm.weight
671
+ Params for bucket 6 (41984000 elements, 41984000 padded size):
672
+ module.decoder.layers.11.pre_mlp_layernorm.weight
673
+ module.decoder.layers.11.self_attention.linear_qkv.weight
674
+ module.decoder.layers.11.mlp.router.weight
675
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
676
+ module.decoder.layers.11.self_attention.linear_proj.weight
677
+ Params for bucket 7 (41984000 elements, 41984000 padded size):
678
+ module.decoder.layers.10.mlp.router.weight
679
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
680
+ module.decoder.layers.10.self_attention.linear_proj.weight
681
+ module.decoder.layers.10.self_attention.linear_qkv.weight
682
+ module.decoder.layers.10.pre_mlp_layernorm.weight
683
+ Params for bucket 8 (41984000 elements, 41984000 padded size):
684
+ module.decoder.layers.9.pre_mlp_layernorm.weight
685
+ module.decoder.layers.9.self_attention.linear_qkv.weight
686
+ module.decoder.layers.9.mlp.router.weight
687
+ module.decoder.layers.9.self_attention.linear_proj.weight
688
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
689
+ Params for bucket 9 (41984000 elements, 41984000 padded size):
690
+ module.decoder.layers.8.pre_mlp_layernorm.weight
691
+ module.decoder.layers.8.self_attention.linear_qkv.weight
692
+ module.decoder.layers.8.mlp.router.weight
693
+ module.decoder.layers.8.self_attention.linear_proj.weight
694
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
695
+ Params for bucket 10 (41984000 elements, 41984000 padded size):
696
+ module.decoder.layers.7.mlp.router.weight
697
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.7.self_attention.linear_proj.weight
699
+ module.decoder.layers.7.self_attention.linear_qkv.weight
700
+ module.decoder.layers.7.pre_mlp_layernorm.weight
701
+ Params for bucket 11 (41984000 elements, 41984000 padded size):
702
+ module.decoder.layers.6.pre_mlp_layernorm.weight
703
+ module.decoder.layers.6.self_attention.linear_qkv.weight
704
+ module.decoder.layers.6.mlp.router.weight
705
+ module.decoder.layers.6.self_attention.linear_proj.weight
706
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
707
+ Params for bucket 12 (41984000 elements, 41984000 padded size):
708
+ module.decoder.layers.5.pre_mlp_layernorm.weight
709
+ module.decoder.layers.5.self_attention.linear_proj.weight
710
+ module.decoder.layers.5.self_attention.linear_qkv.weight
711
+ module.decoder.layers.5.mlp.router.weight
712
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
713
+ Params for bucket 13 (41984000 elements, 41984000 padded size):
714
+ module.decoder.layers.4.self_attention.linear_qkv.weight
715
+ module.decoder.layers.4.pre_mlp_layernorm.weight
716
+ module.decoder.layers.4.mlp.router.weight
717
+ module.decoder.layers.4.self_attention.linear_proj.weight
718
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
719
+ Params for bucket 14 (41984000 elements, 41984000 padded size):
720
+ module.decoder.layers.3.pre_mlp_layernorm.weight
721
+ module.decoder.layers.3.self_attention.linear_qkv.weight
722
+ module.decoder.layers.3.mlp.router.weight
723
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
724
+ module.decoder.layers.3.self_attention.linear_proj.weight
725
+ Params for bucket 15 (41984000 elements, 41984000 padded size):
726
+ module.decoder.layers.2.self_attention.linear_proj.weight
727
+ module.decoder.layers.2.pre_mlp_layernorm.weight
728
+ module.decoder.layers.2.self_attention.linear_qkv.weight
729
+ module.decoder.layers.2.mlp.router.weight
730
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
731
+ Params for bucket 16 (41984000 elements, 41984000 padded size):
732
+ module.decoder.layers.1.mlp.router.weight
733
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
734
+ module.decoder.layers.1.self_attention.linear_proj.weight
735
+ module.decoder.layers.1.pre_mlp_layernorm.weight
736
+ module.decoder.layers.1.self_attention.linear_qkv.weight
737
+ Params for bucket 17 (41984000 elements, 41984000 padded size):
738
+ module.decoder.layers.0.mlp.router.weight
739
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
740
+ module.decoder.layers.0.self_attention.linear_qkv.weight
741
+ module.decoder.layers.0.pre_mlp_layernorm.weight
742
+ module.decoder.layers.0.self_attention.linear_proj.weight
743
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
744
+ module.embedding.word_embeddings.weight
745
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 32
746
+ Params for bucket 1 (58720256 elements, 58720256 padded size):
747
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
748
+ Params for bucket 2 (117440512 elements, 117440512 padded size):
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
750
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
751
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
752
+ Params for bucket 4 (117440512 elements, 117440512 padded size):
753
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
754
+ Params for bucket 5 (58720256 elements, 58720256 padded size):
755
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
756
+ Params for bucket 6 (117440512 elements, 117440512 padded size):
757
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
758
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
759
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
760
+ Params for bucket 8 (117440512 elements, 117440512 padded size):
761
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
762
+ Params for bucket 9 (58720256 elements, 58720256 padded size):
763
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
764
+ Params for bucket 10 (117440512 elements, 117440512 padded size):
765
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
766
+ Params for bucket 11 (58720256 elements, 58720256 padded size):
767
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
768
+ Params for bucket 12 (117440512 elements, 117440512 padded size):
769
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
770
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
771
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
772
+ Params for bucket 14 (117440512 elements, 117440512 padded size):
773
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
774
+ Params for bucket 15 (58720256 elements, 58720256 padded size):
775
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
776
+ Params for bucket 16 (117440512 elements, 117440512 padded size):
777
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
778
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
779
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
780
+ Params for bucket 18 (117440512 elements, 117440512 padded size):
781
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
782
+ Params for bucket 19 (58720256 elements, 58720256 padded size):
783
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
784
+ Params for bucket 20 (117440512 elements, 117440512 padded size):
785
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
786
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
788
+ Params for bucket 22 (117440512 elements, 117440512 padded size):
789
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
790
+ Params for bucket 23 (58720256 elements, 58720256 padded size):
791
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
792
+ Params for bucket 24 (117440512 elements, 117440512 padded size):
793
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
794
+ Params for bucket 25 (58720256 elements, 58720256 padded size):
795
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
796
+ Params for bucket 26 (117440512 elements, 117440512 padded size):
797
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
798
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
799
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
800
+ Params for bucket 28 (117440512 elements, 117440512 padded size):
801
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
802
+ Params for bucket 29 (58720256 elements, 58720256 padded size):
803
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
804
+ Params for bucket 30 (117440512 elements, 117440512 padded size):
805
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
806
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
807
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
808
+ Params for bucket 32 (117440512 elements, 117440512 padded size):
809
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
810
+ 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 0x7cbe2aa29460>, config_logger_dir='')
811
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
812
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:40:29
813
+ > building train, validation, and test datasets ...
814
+ > datasets target sizes (minimum size):
815
+ train: 192
816
+ validation: 0
817
+ test: 0
818
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
819
+ > building train, validation, and test datasets for GPT ...
820
+ 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 0x7cbb846f99a0>, 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)
821
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
822
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
823
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
824
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
825
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
826
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
827
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
828
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
829
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
830
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
831
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
832
+ > finished creating GPT datasets ...
833
+ [after dataloaders are built] datetime: 2026-08-30 07:40:29
834
+ done with setup ...
835
+ (min, max) time across ranks (ms):
836
+ model-and-optimizer-setup ......................: (2843.67, 2846.41)
837
+ train/valid/test-data-iterators-setup ..........: (63.92, 75.55)
838
+ training ...
839
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
840
+ [before the start of training step] datetime: 2026-08-30 07:40:29
841
+ INFO:moonep.api:MoonEP comm buffer sizes:
842
+ config : R=8, S=4096, H=4096, E=8, K=2, NvS=8446, NvS_padded=8448
843
+ sm_count : num_sms=32, num_sms_dedup=148
844
+ hidden_buffer_size : 66.00 MiB
845
+ meta_buffer_size : 2.00 MiB
846
+ local_temp_buffer_size : 512.66 KiB
847
+ total_size : 68.50 MiB
848
+ [MoonEP selfcheck 1/4] dropped groups empty OK | covered rows [0, 8] sizes [768, 7552] | duplicated experts [[1], [-1], [4], [1], [1], [1], [4], [4]]
849
+ [MoonEP selfcheck 2/4] dropped groups empty OK | covered rows [0, 8] sizes [7808, 512] | duplicated experts [[4], [6], [4], [-1], [3], [7], [0], [4]]
850
+ [MoonEP selfcheck 3/4] dropped groups empty OK | covered rows [0, 8] sizes [8064, 256] | duplicated experts [[3], [3], [-1], [2], [0], [2], [2], [5]]
851
+ [MoonEP selfcheck 4/4] dropped groups empty OK | covered rows [0, 8] sizes [8192, 0] | duplicated experts [[-1], [0], [1], [1], [0], [1], [1], [1]]
852
+ Number of parameters in transformer block in billions: 23.22
853
+ Number of parameters in embedding layers in billions: 0.41
854
+ Total number of parameters in billions: 23.63
855
+ [2026-08-30 07:40:49] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 19483.5 | throughput per GPU (TFLOP/s/GPU): 17.1 | 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 |Number of parameters in most loaded shard in billions: 23.6320
856
+
857
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
858
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 53350.6591796875 | max allocated: 53350.65966796875 | reserved: 58996.0 | max reserved: 58996.0
859
+ [2026-08-30 07:40:49] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 540.2 | throughput per GPU (TFLOP/s/GPU): 617.1 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.104935E+01 | loss scale: 1.0 | grad norm: 37.624 | number of skipped iterations: 0 | number of nan iterations: 0 |
860
+ [2026-08-30 07:40:50] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 527.3 | throughput per GPU (TFLOP/s/GPU): 632.2 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.679993E+00 | loss scale: 1.0 | grad norm: 42.167 | number of skipped iterations: 0 | number of nan iterations: 0 |
861
+ Capture CUDA graph for training!!!
862
+ CUDA graph capture done!!!
863
+ [2026-08-30 07:40:51] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1193.1 | throughput per GPU (TFLOP/s/GPU): 279.4 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.119512E+01 | loss scale: 1.0 | grad norm: 89.204 | number of skipped iterations: 0 | number of nan iterations: 0 |
864
+ [2026-08-30 07:40:51] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 491.6 | throughput per GPU (TFLOP/s/GPU): 678.1 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.016800E+01 | loss scale: 1.0 | grad norm: 15.384 | number of skipped iterations: 0 | number of nan iterations: 0 |
865
+ [2026-08-30 07:40:52] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 498.2 | throughput per GPU (TFLOP/s/GPU): 669.2 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.014399E+01 | loss scale: 1.0 | grad norm: 43.370 | number of skipped iterations: 0 | number of nan iterations: 0 |
866
+ [2026-08-30 07:40:52] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 485.1 | throughput per GPU (TFLOP/s/GPU): 687.2 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.904866E+00 | loss scale: 1.0 | grad norm: 8.280 | number of skipped iterations: 0 | number of nan iterations: 0 |
867
+ [2026-08-30 07:40:53] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 487.6 | throughput per GPU (TFLOP/s/GPU): 683.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.981642E+00 | loss scale: 1.0 | grad norm: 7.258 | number of skipped iterations: 0 | number of nan iterations: 0 |
tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/results.tsv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ config backend policy slots seq gbs cap recompute rep status iter_ms_mean iter_ms_min peak_mib run_dir
2
+ G1 moonep one 1 4096 16 none none 1 OK 477.56 460.30 96849 /workspace/tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep1
3
+ G1 moonep one 1 4096 16 none none 2 OK 479.54 460.30 96849 /workspace/tiera-rerun-fixed-moonep/moonep/a1/a1_oneslot_gbs16/G1__moonep__one__seq4096__gbs16__capnone__async/rep2
tiera-rerun-fixed-moonep/moonep/a1/manifest.txt ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ timestamp: 2026-08-30T07:37:24+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-moonep
6
+ study_scripts_commit: f69d9ab
7
+ campaign: moonep
8
+ backend: moonep
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-MoE-Echo
17
+ megatron_branch_required: moonep_backend_async_grad_reduce
18
+ megatron_branch_actual: moonep_backend_async_grad_reduce
19
+ megatron_commit: 2e68c8ad
tiera-rerun-fixed-moonep/moonep/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@2e68c8ad6bfb5bac5d6eb962fa7d4fbf0a5fb095#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
tiera-rerun-fixed-moonep/moonep/a1/provenance.txt ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ PROVENANCE: moonep__a1__gbs16
3
+ ==============================================================
4
+ captured_at : 2026-08-30T07:37:09+00:00
5
+ hostname : sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0
6
+ kernel : 6.8.0-1043-nvidia
7
+ cwd : /workspace/blackwell_run_scripts
8
+ command : run_ladder.sh --phase a1 --gbs 16 --out /workspace/tiera-rerun-fixed-moonep/moonep/a1
9
+
10
+ ---- GPU / driver --------------------------------------------
11
+ 0, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
12
+ 1, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
13
+ 2, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
14
+ 3, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
15
+ 4, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
16
+ 5, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
17
+ 6, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
18
+ 7, NVIDIA B200, 10.0, 183359 MiB, 580.95.05
19
+ persistence/clocks:
20
+ 0, 142 MHz, 1965 MHz, 1000.00 W, 29
21
+ 1, 120 MHz, 1965 MHz, 1000.00 W, 34
22
+ 2, 157 MHz, 1965 MHz, 1000.00 W, 35
23
+ 3, 270 MHz, 1965 MHz, 1000.00 W, 29
24
+ 4, 150 MHz, 1965 MHz, 1000.00 W, 31
25
+ 5, 120 MHz, 1965 MHz, 1000.00 W, 34
26
+ 6, 180 MHz, 1965 MHz, 1000.00 W, 34
27
+ 7, 120 MHz, 1965 MHz, 1000.00 W, 30
28
+
29
+ NVLink topology:
30
+ GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 NIC8 CPU Affinity NUMA Affinity GPU NUMA ID
31
+ GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 PIX NODE SYS SYS SYS SYS SYS SYS NODE 0-27 0 N/A
32
+ GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 NODE PIX SYS SYS SYS SYS SYS SYS NODE 0-27 0 N/A
33
+ GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 SYS SYS PIX NODE SYS SYS SYS SYS SYS 28-55 1 N/A
34
+ GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 SYS SYS NODE PIX SYS SYS SYS SYS SYS 28-55 1 N/A
35
+ GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 SYS SYS SYS SYS PIX NODE SYS SYS SYS 56-83 2 N/A
36
+ GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 SYS SYS SYS SYS NODE PIX SYS SYS SYS 56-83 2 N/A
37
+ GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 SYS SYS SYS SYS SYS SYS PIX NODE SYS 84-111 3 N/A
38
+ GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X SYS SYS SYS SYS SYS SYS NODE PIX SYS 84-111 3 N/A
39
+ NIC0 PIX NODE SYS SYS SYS SYS SYS SYS X NODE SYS SYS SYS SYS SYS SYS NODE
40
+ NIC1 NODE PIX SYS SYS SYS SYS SYS SYS NODE X SYS SYS SYS SYS SYS SYS NODE
41
+ NIC2 SYS SYS PIX NODE SYS SYS SYS SYS SYS SYS X NODE SYS SYS SYS SYS SYS
42
+ NIC3 SYS SYS NODE PIX SYS SYS SYS SYS SYS SYS NODE X SYS SYS SYS SYS SYS
43
+ NIC4 SYS SYS SYS SYS PIX NODE SYS SYS SYS SYS SYS SYS X NODE SYS SYS SYS
44
+ unavailable
45
+
46
+ ---- toolchain -----------------------------------------------
47
+ nvcc : release 12.9, V12.9.86
48
+ gcc : gcc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
49
+ python : 3.12.3
50
+ torch : 2.8.0a0+5228986c39.nv25.06 cuda=12.9 nccl=2.27.3
51
+ torch arch : ['sm_75', 'sm_80', 'sm_86', 'sm_90', 'sm_100', 'sm_120', 'compute_120']
52
+ transformer_engine: 2.10.0.dev0+10f6128d [/usr/local/lib/python3.12/dist-packages/transformer_engine/__init__.py]
53
+ deep_ep : 1.2.1+34152ae [/workspace/DeepEP/deep_ep/__init__.py]
54
+ ultra_ep : 1.0.0 [/workspace/UltraEP/ultra_ep/__init__.py]
55
+ moonep : 0.0.1 [/workspace/MoonEP/moonep/__init__.py]
56
+ megatron : 0.14.0rc5 [None]
57
+ flash_attn : 2.7.4.post1 [/usr/local/lib/python3.12/dist-packages/flash_attn/__init__.py]
58
+ numpy : 1.26.4 [/usr/local/lib/python3.12/dist-packages/numpy/__init__.py]
59
+ huggingface_hub: 1.29.0 [/usr/local/lib/python3.12/dist-packages/huggingface_hub/__init__.py]
60
+
61
+ ---- repositories --------------------------------------------
62
+ Megatron-MoE-Echo moonep_backend_async_grad_reduce 2e68c8ad6bfb
63
+ Megatron-UltraEP ultraep-bracing-regions-through-kernels 03d57ef7d476
64
+ UltraEP main 94cab099b44f
65
+ DeepEP HEAD 34152ae28f80 (+3 untracked: build output)
66
+ MoonEP megatron-ddp-symmetric-params d8bc6667fad7
67
+ blackwell_run_scripts granularity-moonep f69d9abf3591
68
+
69
+ ---- environment ---------------------------------------------
70
+ CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0
71
+ CUDA_DRIVER_VERSION=575.57.08
72
+ CUDA_HOME=/usr/local/cuda
73
+ CUDA_MODULE_LOADING=LAZY
74
+ CUDA_VERSION=12.9.1.010
75
+ HF_REMOTE_ROOT=tiera-rerun-fixed-moonep
76
+ NCCL_VERSION=2.27.3
77
+ NVSHMEM_VERSION=3.2.5
78
+ PYTORCH_BUILD_NUMBER=0
79
+ PYTORCH_BUILD_VERSION=2.8.0a0+5228986
80
+ PYTORCH_HOME=/opt/pytorch/pytorch
81
+ PYTORCH_VERSION=2.8.0a0+5228986
82
+ TORCH_ALLOW_TF32_CUBLAS_OVERRIDE=1
83
+ TORCH_CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0+PTX
84
+ TORCH_NCCL_USE_COMM_NONBLOCKING=0
85
+
86
+ ---- shared library resolution -------------------------------
87
+ libnvshmem_host.so: not in ldconfig
88
+ libnccl.so: /lib/x86_64-linux-gnu/libnccl.so.2
89
+ lrwxrwxrwx 1 root root 20 Aug 29 16:41 /usr/local/lib/python3.12/dist-packages/nvidia/nvshmem/lib/libnvshmem_host.so -> libnvshmem_host.so.3
90
+ -rw-r--r-- 1 root root 179045744 Aug 29 16:37 /usr/local/lib/python3.12/dist-packages/nvidia/nvshmem/lib/libnvshmem_host.so.3
tiera-rerun-fixed-moonep/steps.tsv CHANGED
@@ -6,3 +6,4 @@ OK echo ponly 16 2026-08-30T06:36:31+00:00 2026-08-30T06:44:11+00:00 /workspace/
6
  OK echo a2 16 2026-08-30T06:44:13+00:00 2026-08-30T07:00:25+00:00 /workspace/tiera-rerun-fixed-moonep/echo/a2
7
  OK ultraep a1 16 2026-08-30T07:00:27+00:00 2026-08-30T07:15:17+00:00 /workspace/tiera-rerun-fixed-moonep/ultraep/a1
8
  OK ultraep ponly 16 2026-08-30T07:15:19+00:00 2026-08-30T07:22:01+00:00 /workspace/tiera-rerun-fixed-moonep/ultraep/ponly
 
 
6
  OK echo a2 16 2026-08-30T06:44:13+00:00 2026-08-30T07:00:25+00:00 /workspace/tiera-rerun-fixed-moonep/echo/a2
7
  OK ultraep a1 16 2026-08-30T07:00:27+00:00 2026-08-30T07:15:17+00:00 /workspace/tiera-rerun-fixed-moonep/ultraep/a1
8
  OK ultraep ponly 16 2026-08-30T07:15:19+00:00 2026-08-30T07:22:01+00:00 /workspace/tiera-rerun-fixed-moonep/ultraep/ponly
9
+ OK ultraep a2 16 2026-08-30T07:22:03+00:00 2026-08-30T07:37:07+00:00 /workspace/tiera-rerun-fixed-moonep/ultraep/a2
tiera-rerun-fixed-moonep/study.log CHANGED
@@ -780,3 +780,67 @@ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
780
  --------------------------------------------------------------------------
781
  [07:29:41] CELL G3__ultraep__matched__seq4096__gbs16__cap1p072265625 (slots=8)
782
  [guard] iter 3/12 (60s)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
780
  --------------------------------------------------------------------------
781
  [07:29:41] CELL G3__ultraep__matched__seq4096__gbs16__cap1p072265625 (slots=8)
782
  [guard] iter 3/12 (60s)
783
+ [guard] iter 12/12 (70s)
784
+ [guard] tearing down process group (outcome=DONE)
785
+ [guard] iter 3/12 (60s)
786
+ [guard] iter 12/12 (70s)
787
+ [guard] tearing down process group (outcome=DONE)
788
+ [guard] iter 3/12 (60s)
789
+ [guard] iter 12/12 (70s)
790
+ [guard] tearing down process group (outcome=DONE)
791
+ --------------------------------------------------------------------------
792
+ [07:33:23] CELL G4__ultraep__matched__seq4096__gbs16__cap1p095703125 (slots=16)
793
+ [guard] iter 5/12 (60s)
794
+ [guard] iter 12/12 (70s)
795
+ [guard] tearing down process group (outcome=DONE)
796
+ [guard] iter 3/12 (60s)
797
+ [guard] iter 12/12 (70s)
798
+ [guard] tearing down process group (outcome=DONE)
799
+ [guard] iter 5/12 (60s)
800
+ [guard] iter 12/12 (70s)
801
+ [guard] tearing down process group (outcome=DONE)
802
+ --------------------------------------------------------------------------
803
+
804
+ ==============================================================
805
+ SWEEP COMPLETE ok=12 failed=0
806
+ ==============================================================
807
+ config backend policy slots seq gbs cap recompute rep status iter_ms_mean iter_ms_min peak_mib run_dir
808
+ G1 ultraep matched 1 4096 16 2.40625 none 1 OK 893.80 876.90 129475 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G1__ultraep__matched__seq4096__gbs16__cap2p40625/rep1
809
+ G1 ultraep matched 1 4096 16 2.40625 none 2 OK 892.33 875.20 129475 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G1__ultraep__matched__seq4096__gbs16__cap2p40625/rep2
810
+ G1 ultraep matched 1 4096 16 2.40625 none 3 OK 891.92 874.90 129475 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G1__ultraep__matched__seq4096__gbs16__cap2p40625/rep3
811
+ G2 ultraep matched 4 4096 16 1.08203125 none 1 OK 492.27 465.20 112315 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G2__ultraep__matched__seq4096__gbs16__cap1p08203125/rep1
812
+ G2 ultraep matched 4 4096 16 1.08203125 none 2 OK 492.21 462.50 112315 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G2__ultraep__matched__seq4096__gbs16__cap1p08203125/rep2
813
+ G2 ultraep matched 4 4096 16 1.08203125 none 3 OK 492.54 462.50 112315 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G2__ultraep__matched__seq4096__gbs16__cap1p08203125/rep3
814
+ G3 ultraep matched 8 4096 16 1.072265625 none 1 OK 460.01 449.60 113951 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1
815
+ G3 ultraep matched 8 4096 16 1.072265625 none 2 OK 459.50 449.60 113951 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2
816
+ G3 ultraep matched 8 4096 16 1.072265625 none 3 OK 459.60 449.60 113951 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3
817
+ G4 ultraep matched 16 4096 16 1.095703125 none 1 OK 483.19 475.20 132437 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1
818
+ G4 ultraep matched 16 4096 16 1.095703125 none 2 OK 482.88 474.50 132437 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2
819
+ G4 ultraep matched 16 4096 16 1.095703125 none 3 OK 482.64 474.50 132437 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3
820
+
821
+ Results: /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/results.tsv
822
+ Roll up: python3 /workspace/blackwell_run_scripts/granularity/tools/collect.py /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16
823
+ ==============================================================
824
+
825
+ Campaign results: /workspace/tiera-rerun-fixed-moonep/ultraep/a2
826
+ Roll up: python3 /workspace/blackwell_run_scripts/granularity/tools/collect.py /workspace/tiera-rerun-fixed-moonep/ultraep/a2 --compare
827
+ Already logged in to Hugging Face as: AbhishekGhosh1998
828
+ uploaded: /workspace/tiera-rerun-fixed-moonep/ultraep/a2 -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep/ultraep/a2
829
+ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
830
+
831
+ === STEP moonep__a1__gbs16
832
+ ==============================================================
833
+ CAMPAIGN: moonep backend=moonep phase=a1
834
+ Megatron: /workspace/Megatron-MoE-Echo @ moonep_backend_async_grad_reduce
835
+ GBS=16 seq=4096 repeats=3 nsys=0
836
+ results -> /workspace/tiera-rerun-fixed-moonep/moonep/a1
837
+ ==============================================================
838
+ --------------------------------------------------------------------------
839
+ [07:37:24] CELL G1__moonep__one__seq4096__gbs16__capnone__async (slots=1)
840
+ [guard] iter 3/12 (60s)
841
+ [guard] iter 12/12 (70s)
842
+ [guard] tearing down process group (outcome=DONE)
843
+ [guard] iter 5/12 (60s)
844
+ [guard] iter 12/12 (70s)
845
+ [guard] tearing down process group (outcome=DONE)
846
+ [guard] iter 5/12 (60s)