AbhishekGhosh1998 commited on
Commit
b666176
·
verified ·
1 Parent(s): 4a968d8

granularity results granularity_next_b200/tiera_topk_trace @ 2026-09-17T02:54:53Z from hp-aiinfra-interactive-aghosh-8-xzthl-8xb200-moe-cg-0

Browse files
Files changed (12) hide show
  1. .gitattributes +1 -0
  2. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/train.log +43 -0
  3. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/verify_trace_ranks.log +6 -0
  4. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667.nsys-rep +3 -0
  5. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/run.env +102 -0
  6. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/train.log +0 -0
  7. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/verify_trace_ranks.log +6 -0
  8. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K8__echo__matched__seq4096__gbs16__cap1p109375/nsys/run.env +102 -0
  9. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K8__echo__matched__seq4096__gbs16__cap1p109375/nsys/train.log +562 -0
  10. granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/trace_results.tsv +2 -0
  11. granularity_next_b200/tiera_topk_trace/hf_watcher.log +3 -0
  12. granularity_next_b200/tiera_topk_trace/study.log +10 -0
.gitattributes CHANGED
@@ -87,3 +87,4 @@ granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G3_K2__e
87
  granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G3_K8__echo__matched__seq4096__gbs16__cap1p09375/nsys/G3_K8__echo__matched__seq4096__gbs16__cap1p09375.nsys-rep filter=lfs diff=lfs merge=lfs -text
88
  granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K2__echo__matched__seq4096__gbs16__cap1p265625/nsys/G4_K2__echo__matched__seq4096__gbs16__cap1p265625.nsys-rep filter=lfs diff=lfs merge=lfs -text
89
  granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/G4_K4__echo__matched__seq4096__gbs16__cap1p171875.nsys-rep filter=lfs diff=lfs merge=lfs -text
 
 
87
  granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G3_K8__echo__matched__seq4096__gbs16__cap1p09375/nsys/G3_K8__echo__matched__seq4096__gbs16__cap1p09375.nsys-rep filter=lfs diff=lfs merge=lfs -text
88
  granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K2__echo__matched__seq4096__gbs16__cap1p265625/nsys/G4_K2__echo__matched__seq4096__gbs16__cap1p265625.nsys-rep filter=lfs diff=lfs merge=lfs -text
89
  granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/G4_K4__echo__matched__seq4096__gbs16__cap1p171875.nsys-rep filter=lfs diff=lfs merge=lfs -text
90
+ granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667.nsys-rep filter=lfs diff=lfs merge=lfs -text
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/train.log CHANGED
@@ -1500,3 +1500,46 @@ Generating '/tmp/nsys-report-0c8b.qdstrm'
1500
 
1501
 
1502
  Collection stop is not allowed in this state.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1500
 
1501
 
1502
  Collection stop is not allowed in this state.
1503
+ W0917 02:44:59.050000 839997 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1504
+ W0917 02:44:59.050000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840111 closing signal SIGTERM
1505
+ Generated:
1506
+ /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/G4_K4__echo__matched__seq4096__gbs16__cap1p171875.nsys-rep
1507
+ W0917 02:44:59.056000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840112 closing signal SIGTERM
1508
+ W0917 02:44:59.061000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840113 closing signal SIGTERM
1509
+ W0917 02:44:59.063000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840114 closing signal SIGTERM
1510
+ W0917 02:44:59.068000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840115 closing signal SIGTERM
1511
+ W0917 02:44:59.076000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840116 closing signal SIGTERM
1512
+ W0917 02:44:59.081000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840117 closing signal SIGTERM
1513
+ W0917 02:44:59.084000 839997 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 840118 closing signal SIGTERM
1514
+ Traceback (most recent call last):
1515
+ File "/usr/local/bin/torchrun", line 8, in <module>
1516
+ sys.exit(main())
1517
+ ^^^^^^
1518
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1519
+ return f(*args, **kwargs)
1520
+ ^^^^^^^^^^^^^^^^^^
1521
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1522
+ run(args)
1523
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1524
+ elastic_launch(
1525
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1526
+ return launch_agent(self._config, self._entrypoint, list(args))
1527
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1528
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1529
+ result = agent.run()
1530
+ ^^^^^^^^^^^
1531
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1532
+ result = f(*args, **kwargs)
1533
+ ^^^^^^^^^^^^^^^^^^
1534
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1535
+ result = self._invoke_run(role)
1536
+ ^^^^^^^^^^^^^^^^^^^^^^
1537
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1538
+ time.sleep(monitor_interval)
1539
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1540
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1541
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 839997 got signal: 15
1542
+ The target application terminated. One or more process it created re-parented.
1543
+ Waiting for termination of re-parented processes.
1544
+ Use the `--wait` option to modify this behavior.
1545
+ There are no active sessions.
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/verify_trace_ranks.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ report : /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/G4_K4__echo__matched__seq4096__gbs16__cap1p171875.nsys-rep
2
+ kernels : 281286
3
+ devices : 8 of 8 -> [0, 1, 2, 3, 4, 5, 6, 7]
4
+ processes: 1
5
+ markers : 23552 echo_trace__ kernels on devices [0, 1, 2, 3, 4, 5, 6, 7]
6
+ PASS: all ranks present and backend markers emitted
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667.nsys-rep ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a774291e009dce660cf63f08fc217215f9e3bdec9da51cf58cfafb3f983a7f1
3
+ size 19852134
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/run.env ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=echo
2
+ config=G4_K6
3
+ family=tierA
4
+ source_model=
5
+ standard_num_layers=
6
+ base_rung=G4
7
+ repo=/workspace/Megatron-MoE-Echo
8
+ branch=bracing-regions-through-kernels
9
+ commit=9a236793
10
+ num_layers=16
11
+ hidden=4096
12
+ ffn_hidden=14336
13
+ moe_ffn_hidden=896
14
+ num_experts=128
15
+ topk=6
16
+ experts_per_rank=16
17
+ expert_params=11010048
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
+ q_lora_rank=
27
+ kv_lora_rank=512
28
+ qk_head_dim=128
29
+ qk_pos_emb_head_dim=64
30
+ v_head_dim=128
31
+ dense_first_layers=0
32
+ moe_layers=16
33
+ rotary_base=1000000
34
+ rotary_percent=1.0
35
+ rotary_scaling=
36
+ mscale=
37
+ mscale_all_dim=
38
+ qkv_bias=0
39
+ router_balancing=aux_loss
40
+ aux_loss_coeff=1e-2
41
+ router_pre_softmax=0
42
+ router_fusion=0
43
+ router_dtype=fp32
44
+ router_groups=
45
+ router_group_topk=
46
+ router_score_function=softmax
47
+ router_expert_bias=0
48
+ router_bias_update_rate=
49
+ router_topk_scaling_factor=
50
+ shared_expert_gate=0
51
+ apply_layernorm_1p=0
52
+ attention_output_gate=0
53
+ linear_attention_type=
54
+ linear_attention_freq=
55
+ linear_conv_kernel_dim=
56
+ linear_key_head_dim=
57
+ linear_value_head_dim=
58
+ linear_num_key_heads=
59
+ linear_num_value_heads=
60
+ mtp_num_layers=0
61
+ mtp_loss_scaling_factor=0.1
62
+ no_weight_decay_cond_type=
63
+ max_position_embeddings=32768
64
+ init_method_std=0.01
65
+ device_initiated_grouped_gemm=1
66
+ norm_epsilon=1e-5
67
+ seq_length=4096
68
+ micro_batch_size=1
69
+ global_batch_size=16
70
+ num_microbatches=2
71
+ assignments_per_rank=24576
72
+ mean_tokens_per_expert=1536
73
+ slot_policy=matched
74
+ slots_per_rank=16
75
+ capacity_mode=static
76
+ capacity=1.1354166666666667
77
+ cuda_graph=1
78
+ probe_mode=0
79
+ async_grad_reduce=0
80
+ branch_role=timing
81
+ memory_dump_mode=none
82
+ memory_snapshot_path=
83
+ cuda_graph_warmup_steps=3
84
+ recompute=none
85
+ force_balance=0
86
+ train_iters=10
87
+ ep_size=8
88
+ echo_dispatch_overlap=1
89
+ echo_dispatch_recompute=0
90
+ ultraep_quota_min_tokens=
91
+ ultraep_weight_sync_join=
92
+ profile_enable=1
93
+ kernel_trace=1
94
+ fp8_mode=mxfp8
95
+ fp8_recipe=mxfp8
96
+ training_seed=1234
97
+ learning_rate=1e-4
98
+ min_learning_rate=1.0e-5
99
+ lr_warmup_iters=5
100
+ state_dump_enable=0
101
+ state_dump_dir=
102
+ moonep_assert_no_pending_reduce=0
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/train.log ADDED
The diff for this file is too large to render. See raw diff
 
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/verify_trace_ranks.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ report : /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667.nsys-rep
2
+ kernels : 281286
3
+ devices : 8 of 8 -> [0, 1, 2, 3, 4, 5, 6, 7]
4
+ processes: 1
5
+ markers : 23552 echo_trace__ kernels on devices [0, 1, 2, 3, 4, 5, 6, 7]
6
+ PASS: all ranks present and backend markers emitted
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K8__echo__matched__seq4096__gbs16__cap1p109375/nsys/run.env ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=echo
2
+ config=G4_K8
3
+ family=tierA
4
+ source_model=
5
+ standard_num_layers=
6
+ base_rung=G4
7
+ repo=/workspace/Megatron-MoE-Echo
8
+ branch=bracing-regions-through-kernels
9
+ commit=9a236793
10
+ num_layers=16
11
+ hidden=4096
12
+ ffn_hidden=14336
13
+ moe_ffn_hidden=896
14
+ num_experts=128
15
+ topk=8
16
+ experts_per_rank=16
17
+ expert_params=11010048
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
+ q_lora_rank=
27
+ kv_lora_rank=512
28
+ qk_head_dim=128
29
+ qk_pos_emb_head_dim=64
30
+ v_head_dim=128
31
+ dense_first_layers=0
32
+ moe_layers=16
33
+ rotary_base=1000000
34
+ rotary_percent=1.0
35
+ rotary_scaling=
36
+ mscale=
37
+ mscale_all_dim=
38
+ qkv_bias=0
39
+ router_balancing=aux_loss
40
+ aux_loss_coeff=1e-2
41
+ router_pre_softmax=0
42
+ router_fusion=0
43
+ router_dtype=fp32
44
+ router_groups=
45
+ router_group_topk=
46
+ router_score_function=softmax
47
+ router_expert_bias=0
48
+ router_bias_update_rate=
49
+ router_topk_scaling_factor=
50
+ shared_expert_gate=0
51
+ apply_layernorm_1p=0
52
+ attention_output_gate=0
53
+ linear_attention_type=
54
+ linear_attention_freq=
55
+ linear_conv_kernel_dim=
56
+ linear_key_head_dim=
57
+ linear_value_head_dim=
58
+ linear_num_key_heads=
59
+ linear_num_value_heads=
60
+ mtp_num_layers=0
61
+ mtp_loss_scaling_factor=0.1
62
+ no_weight_decay_cond_type=
63
+ max_position_embeddings=32768
64
+ init_method_std=0.01
65
+ device_initiated_grouped_gemm=1
66
+ norm_epsilon=1e-5
67
+ seq_length=4096
68
+ micro_batch_size=1
69
+ global_batch_size=16
70
+ num_microbatches=2
71
+ assignments_per_rank=32768
72
+ mean_tokens_per_expert=2048
73
+ slot_policy=matched
74
+ slots_per_rank=16
75
+ capacity_mode=static
76
+ capacity=1.109375
77
+ cuda_graph=1
78
+ probe_mode=0
79
+ async_grad_reduce=0
80
+ branch_role=timing
81
+ memory_dump_mode=none
82
+ memory_snapshot_path=
83
+ cuda_graph_warmup_steps=3
84
+ recompute=none
85
+ force_balance=0
86
+ train_iters=10
87
+ ep_size=8
88
+ echo_dispatch_overlap=1
89
+ echo_dispatch_recompute=0
90
+ ultraep_quota_min_tokens=
91
+ ultraep_weight_sync_join=
92
+ profile_enable=1
93
+ kernel_trace=1
94
+ fp8_mode=mxfp8
95
+ fp8_recipe=mxfp8
96
+ training_seed=1234
97
+ learning_rate=1e-4
98
+ min_learning_rate=1.0e-5
99
+ lr_warmup_iters=5
100
+ state_dump_enable=0
101
+ state_dump_dir=
102
+ moonep_assert_no_pending_reduce=0
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K8__echo__matched__seq4096__gbs16__cap1p109375/nsys/train.log ADDED
@@ -0,0 +1,562 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G4_K8 / echo (slots/rank=16 policy=matched)
3
+ E=128 K=8 H'=896 m=16
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.109375 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K8__echo__matched__seq4096__gbs16__cap1p109375/nsys/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 9a236793
8
+ ==============================================================
9
+ W0917 02:54:44.399000 880794 torch/distributed/run.py:766]
10
+ W0917 02:54:44.399000 880794 torch/distributed/run.py:766] *****************************************
11
+ W0917 02:54:44.399000 880794 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
+ W0917 02:54:44.399000 880794 torch/distributed/run.py:766] *****************************************
13
+ 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
14
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
15
+ Number of virtual stages per pipeline stage: None
16
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
17
+ using torch.bfloat16 for parameters ...
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
+ ------------------------ arguments ------------------------
21
+ account_for_embedding_in_pipeline_split ......... False
22
+ account_for_loss_in_pipeline_split .............. False
23
+ accumulate_allreduce_grads_in_fp32 .............. True
24
+ adam_beta1 ...................................... 0.9
25
+ adam_beta2 ...................................... 0.999
26
+ adam_eps ........................................ 1e-08
27
+ add_bias_linear ................................. False
28
+ add_position_embedding .......................... False
29
+ add_qkv_bias .................................... False
30
+ adlr_autoresume ................................. False
31
+ adlr_autoresume_interval ........................ 1000
32
+ align_grad_reduce ............................... True
33
+ align_param_gather .............................. False
34
+ app_tag_run_name ................................ None
35
+ app_tag_run_version ............................. 0.0.0
36
+ apply_layernorm_1p .............................. False
37
+ apply_query_key_layer_scaling ................... False
38
+ apply_residual_connection_post_layernorm ........ False
39
+ apply_rope_fusion ............................... True
40
+ async_save ...................................... None
41
+ async_tensor_model_parallel_allreduce ........... True
42
+ attention_backend ............................... AttnBackend.auto
43
+ attention_dropout ............................... 0.0
44
+ attention_softmax_in_fp32 ....................... False
45
+ auto_detect_ckpt_format ......................... False
46
+ barrier_with_L1_time ............................ True
47
+ bert_binary_head ................................ True
48
+ bert_embedder_type .............................. megatron
49
+ bert_load ....................................... None
50
+ bf16 ............................................ True
51
+ bias_dropout_fusion ............................. True
52
+ bias_gelu_fusion ................................ False
53
+ bias_swiglu_fusion .............................. True
54
+ biencoder_projection_dim ........................ 0
55
+ biencoder_shared_query_context_model ............ False
56
+ block_data_path ................................. None
57
+ calc_ft_timeouts ................................ False
58
+ calculate_per_token_loss ........................ False
59
+ check_for_large_grads ........................... False
60
+ check_for_nan_in_loss_and_grad .................. False
61
+ check_for_spiky_loss ............................ False
62
+ check_weight_hash_across_dp_replicas_interval ... None
63
+ ckpt_assume_constant_structure .................. False
64
+ ckpt_convert_format ............................. None
65
+ ckpt_convert_save ............................... None
66
+ ckpt_convert_update_legacy_dist_opt_format ...... False
67
+ ckpt_format ..................................... torch_dist
68
+ ckpt_fully_parallel_load ........................ False
69
+ ckpt_fully_parallel_save ........................ True
70
+ ckpt_fully_parallel_save_deprecated ............. False
71
+ ckpt_step ....................................... None
72
+ classes_fraction ................................ 1.0
73
+ clip_grad ....................................... 1.0
74
+ clone_scatter_output_in_embedding ............... True
75
+ config_logger_dir ...............................
76
+ consumed_train_samples .......................... 0
77
+ consumed_valid_samples .......................... 0
78
+ context_parallel_size ........................... 1
79
+ cp_comm_type .................................... ['p2p']
80
+ create_attention_mask_in_dataloader ............. True
81
+ cross_entropy_fusion_impl ....................... native
82
+ cross_entropy_loss_fusion ....................... False
83
+ cuda_graph_scope ................................ full_iteration
84
+ cuda_graph_warmup_steps ......................... 3
85
+ data_args_path .................................. None
86
+ data_cache_path ................................. None
87
+ data_parallel_random_init ....................... False
88
+ data_parallel_sharding_strategy ................. no_shard
89
+ data_parallel_size .............................. 8
90
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
91
+ data_per_class_fraction ......................... 1.0
92
+ data_sharding ................................... True
93
+ dataloader_type ................................. single
94
+ ddp_average_in_collective ....................... False
95
+ ddp_bucket_size ................................. None
96
+ ddp_num_buckets ................................. None
97
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
98
+ decoder_first_pipeline_num_layers ............... None
99
+ decoder_last_pipeline_num_layers ................ None
100
+ decoder_num_layers .............................. None
101
+ decoder_seq_length .............................. None
102
+ decoupled_lr .................................... None
103
+ decoupled_min_lr ................................ None
104
+ decrease_batch_size_if_needed ................... False
105
+ defer_embedding_wgrad_compute ................... False
106
+ delay_wgrad_compute ............................. False
107
+ deprecated_use_mcore_models ..................... True
108
+ deterministic_mode .............................. False
109
+ dino_bottleneck_size ............................ 256
110
+ dino_freeze_last_layer .......................... 1
111
+ dino_head_hidden_size ........................... 2048
112
+ dino_local_crops_number ......................... 10
113
+ dino_local_img_size ............................. 96
114
+ dino_norm_last_layer ............................ False
115
+ dino_teacher_temp ............................... 0.07
116
+ dino_warmup_teacher_temp ........................ 0.04
117
+ dino_warmup_teacher_temp_epochs ................. 30
118
+ disable_bf16_reduced_precision_matmul ........... False
119
+ disable_mamba_mem_eff_path ...................... False
120
+ disable_straggler_on_startup .................... False
121
+ dist_ckpt_format_deprecated ..................... None
122
+ dist_ckpt_strictness ............................ assume_ok_unexpected
123
+ distribute_saved_activations .................... False
124
+ distributed_backend ............................. nccl
125
+ distributed_timeout_minutes ..................... 10
126
+ embedding_init_method_std ....................... None
127
+ embedding_path .................................. None
128
+ empty_unused_memory_level ....................... 0
129
+ enable_cuda_graph ............................... True
130
+ enable_experimental ............................. False
131
+ enable_ft_package ............................... False
132
+ enable_gloo_process_groups ...................... True
133
+ enable_msc ...................................... True
134
+ enable_one_logger ............................... True
135
+ encoder_num_layers .............................. 16
136
+ encoder_seq_length .............................. 4096
137
+ end_weight_decay ................................ 0.1
138
+ eod_mask_loss ................................... False
139
+ error_injection_rate ............................ 0
140
+ error_injection_type ............................ transient_error
141
+ eval_interval ................................... 1000000
142
+ eval_iters ...................................... 0
143
+ evidence_data_path .............................. None
144
+ exit_duration_in_mins ........................... None
145
+ exit_interval ................................... None
146
+ exit_on_missing_checkpoint ...................... False
147
+ exit_signal_handler ............................. False
148
+ exp_avg_dtype ................................... torch.float32
149
+ exp_avg_sq_dtype ................................ torch.float32
150
+ expert_model_parallel_size ...................... 8
151
+ expert_tensor_parallel_size ..................... 1
152
+ export_force_local_attention .................... False
153
+ export_kd_cfg ................................... None
154
+ export_kd_teacher_ckpt_format ................... None
155
+ export_kd_teacher_load .......................... None
156
+ export_kv_cache_quant ........................... False
157
+ export_legacy_megatron .......................... False
158
+ export_model_type ............................... GPTModel
159
+ export_moe_apply_probs_on_input ................. False
160
+ export_qk_l2_norm ............................... False
161
+ export_quant_cfg ................................ None
162
+ export_real_quant_cfg ........................... None
163
+ export_te_mcore_model ........................... False
164
+ external_cuda_graph ............................. False
165
+ ffn_hidden_size ................................. 14336
166
+ finetune ........................................ False
167
+ finetune_data_split ............................. train
168
+ finetune_hf_dataset ............................. None
169
+ first_last_layers_bf16 .......................... False
170
+ flash_decode .................................... False
171
+ fp16 ............................................ False
172
+ fp16_lm_cross_entropy ........................... False
173
+ fp32_residual_connection ........................ False
174
+ fp8 ............................................. e4m3
175
+ fp8_amax_compute_algo ........................... most_recent
176
+ fp8_amax_history_len ............................ 1
177
+ fp8_interval .................................... 1
178
+ fp8_margin ...................................... 0
179
+ fp8_param_gather ................................ False
180
+ fp8_recipe ...................................... mxfp8
181
+ fp8_wgrad ....................................... True
182
+ fsdp_double_buffer .............................. False
183
+ global_batch_size ............................... 16
184
+ grad_reduce_in_bf16 ............................. False
185
+ gradient_accumulation_fusion .................... True
186
+ gradient_reduce_div_fusion ...................... True
187
+ group_query_attention ........................... True
188
+ head_lr_mult .................................... 1.0
189
+ heterogeneous_layers_config_encoded_json ........ None
190
+ heterogeneous_layers_config_path ................ None
191
+ hidden_dropout .................................. 0.0
192
+ hidden_size ..................................... 4096
193
+ hierarchical_context_parallel_sizes ............. None
194
+ high_priority_stream_groups ..................... []
195
+ hybrid_attention_ratio .......................... 0.0
196
+ hybrid_mlp_ratio ................................ 0.0
197
+ hybrid_override_pattern ......................... None
198
+ hysteresis ...................................... 2
199
+ ict_head_size ................................... None
200
+ ict_load ........................................ None
201
+ img_h ........................................... 224
202
+ img_w ........................................... 224
203
+ indexer_batch_size .............................. 128
204
+ indexer_log_interval ............................ 1000
205
+ inference_batch_times_seqlen_threshold .......... -1
206
+ inference_dynamic_batching ...................... False
207
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
208
+ inference_dynamic_batching_buffer_overflow_factor None
209
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
210
+ inference_dynamic_batching_chunk_size ........... 256
211
+ inference_dynamic_batching_max_requests_override None
212
+ inference_dynamic_batching_max_tokens_override .. None
213
+ inference_dynamic_batching_num_cuda_graphs ...... 16
214
+ inference_max_batch_size ........................ 8
215
+ inference_max_seq_length ........................ 2560
216
+ inference_rng_tracker ........................... False
217
+ init_method_std ................................. 0.01
218
+ init_method_xavier_uniform ...................... False
219
+ init_model_with_meta_device ..................... False
220
+ initial_loss_scale .............................. 4294967296
221
+ inprocess_active_world_size ..................... 8
222
+ inprocess_barrier_timeout ....................... 120
223
+ inprocess_completion_timeout .................... 120
224
+ inprocess_empty_cuda_cache ...................... False
225
+ inprocess_granularity ........................... node
226
+ inprocess_hard_timeout .......................... 90
227
+ inprocess_heartbeat_interval .................... 30
228
+ inprocess_heartbeat_timeout ..................... 60
229
+ inprocess_last_call_wait ........................ 1
230
+ inprocess_max_iterations ........................ None
231
+ inprocess_monitor_process_interval .............. 1.0
232
+ inprocess_monitor_thread_interval ............... 1.0
233
+ inprocess_progress_watchdog_interval ............ 1.0
234
+ inprocess_restart ............................... False
235
+ inprocess_soft_timeout .......................... 60
236
+ inprocess_termination_grace_time ................ 1
237
+ is_hybrid_model ................................. False
238
+ iter_per_epoch .................................. 1250
239
+ iterations_to_skip .............................. []
240
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
241
+ kitchen_config_file ............................. None
242
+ kitchen_recipe_number ........................... None
243
+ kv_channels ..................................... 128
244
+ kv_lora_rank .................................... 32
245
+ lazy_mpu_init ................................... None
246
+ load ............................................ None
247
+ load_main_params_from_ckpt ...................... None
248
+ load_model_opt_format ........................... False
249
+ local_rank ...................................... 0
250
+ log_energy ...................................... False
251
+ log_interval .................................... 1
252
+ log_loss_scale_to_tensorboard ................... True
253
+ log_memory_to_tensorboard ....................... False
254
+ log_num_zeros_in_grad ........................... False
255
+ log_params_norm ................................. False
256
+ log_progress .................................... False
257
+ log_straggler ................................... False
258
+ log_throughput .................................. True
259
+ log_timers_to_tensorboard ....................... False
260
+ log_validation_ppl_to_tensorboard ............... False
261
+ log_world_size_to_tensorboard ................... False
262
+ logging_level ................................... None
263
+ loss_scale ...................................... None
264
+ loss_scale_window ............................... 1000
265
+ lr .............................................. 0.0001
266
+ lr_decay_iters .................................. 8
267
+ lr_decay_samples ................................ None
268
+ lr_decay_style .................................. cosine
269
+ lr_warmup_fraction .............................. None
270
+ lr_warmup_init .................................. 0.0
271
+ lr_warmup_iters ................................. 5
272
+ lr_warmup_samples ............................... 0
273
+ lr_wsd_decay_iters .............................. None
274
+ lr_wsd_decay_samples ............................ None
275
+ lr_wsd_decay_style .............................. exponential
276
+ main_grads_dtype ................................ torch.float32
277
+ main_params_dtype ............................... torch.float32
278
+ make_vocab_size_divisible_by .................... 128
279
+ mamba_head_dim .................................. 64
280
+ mamba_num_groups ................................ 8
281
+ mamba_num_heads ................................. None
282
+ mamba_state_dim ................................. 128
283
+ manual_gc ....................................... False
284
+ manual_gc_eval .................................. True
285
+ manual_gc_interval .............................. 0
286
+ mask_factor ..................................... 1.0
287
+ mask_prob ....................................... 0.15
288
+ mask_type ....................................... random
289
+ masked_softmax_fusion ........................... False
290
+ max_position_embeddings ......................... 32768
291
+ max_tokens_to_oom ............................... 12000
292
+ memory_snapshot_path ............................ snapshot.pickle
293
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
294
+ micro_batch_size ................................ 1
295
+ microbatch_group_size_per_vp_stage .............. None
296
+ mid_level_dataset_surplus ....................... 0.005
297
+ min_loss_scale .................................. 1.0
298
+ min_lr .......................................... 1e-05
299
+ mlp_chunks_for_prefill .......................... 1
300
+ mmap_bin_files .................................. True
301
+ mock_data ....................................... False
302
+ moe_apply_probs_on_input ........................ False
303
+ moe_aux_loss_coeff .............................. 0.01
304
+ moe_deepep_num_sms .............................. 20
305
+ moe_echo_enable_random_offloading ............... False
306
+ moe_echo_expert_dispatch_overlap ................ True
307
+ moe_echo_expert_dispatcher_type ................. hybridep
308
+ moe_echo_recompute_expert_dispatch .............. False
309
+ moe_enable_deepep ............................... False
310
+ moe_enable_echo ................................. True
311
+ moe_enable_hybridep ............................. True
312
+ moe_expert_capacity_factor ...................... None
313
+ moe_extended_tp ................................. False
314
+ moe_ffn_hidden_size ............................. 896
315
+ moe_grouped_gemm ................................ True
316
+ moe_hybridep_num_sms ............................ 16
317
+ moe_input_jitter_eps ............................ None
318
+ moe_layer_freq .................................. 1
319
+ moe_layer_recompute ............................. False
320
+ moe_num_echo_experts ............................ 128
321
+ moe_pad_expert_input_to_capacity ................ False
322
+ moe_per_layer_logging ........................... False
323
+ moe_permute_fusion .............................. False
324
+ moe_received_token_capacity ..................... 1.109375
325
+ moe_router_bias_update_rate ..................... 0.001
326
+ moe_router_dtype ................................ fp32
327
+ moe_router_enable_expert_bias ................... False
328
+ moe_router_force_load_balancing ................. False
329
+ moe_router_group_topk ........................... None
330
+ moe_router_load_balancing_type .................. aux_loss
331
+ moe_router_num_groups ........................... None
332
+ moe_router_padding_for_fp8 ...................... False
333
+ moe_router_pre_softmax .......................... False
334
+ moe_router_score_function ....................... softmax
335
+ moe_router_topk ................................. 8
336
+ moe_router_topk_scaling_factor .................. None
337
+ moe_shared_expert_intermediate_size ............. None
338
+ moe_shared_expert_overlap ....................... False
339
+ moe_token_dispatcher_type ....................... flex
340
+ moe_token_drop_policy ........................... probs
341
+ moe_upcycling_granularity ....................... 1
342
+ moe_use_device_initiated_grouped_gemm ........... True
343
+ moe_use_legacy_grouped_gemm ..................... False
344
+ moe_use_upcycling ............................... False
345
+ moe_z_loss_coeff ................................ None
346
+ mrope_section ................................... None
347
+ mscale .......................................... 1.0
348
+ mscale_all_dim .................................. 0.0
349
+ mtp_loss_scaling_factor ......................... 0.1
350
+ mtp_num_layers .................................. None
351
+ multi_latent_attention .......................... False
352
+ nccl_all_reduce_for_prefill ..................... False
353
+ nccl_communicator_config_path ................... None
354
+ nccl_ub ......................................... False
355
+ no_load_optim ................................... None
356
+ no_load_rng ..................................... None
357
+ no_persist_layer_norm ........................... False
358
+ no_rope_freq .................................... None
359
+ no_save_optim ................................... None
360
+ no_save_rng ..................................... None
361
+ non_persistent_ckpt_type ........................ None
362
+ non_persistent_global_ckpt_dir .................. None
363
+ non_persistent_local_ckpt_algo .................. fully_parallel
364
+ non_persistent_local_ckpt_dir ................... None
365
+ non_persistent_save_interval .................... None
366
+ norm_epsilon .................................... 1e-05
367
+ normalization ................................... RMSNorm
368
+ num_attention_heads ............................. 32
369
+ num_channels .................................... 3
370
+ num_classes ..................................... 1000
371
+ num_dataset_builder_threads ..................... 1
372
+ num_distributed_optimizer_instances ............. 1
373
+ num_experts ..................................... 128
374
+ num_layers ...................................... 16
375
+ num_layers_at_end_in_bf16 ....................... 1
376
+ num_layers_at_start_in_bf16 ..................... 1
377
+ num_layers_per_virtual_pipeline_stage ........... None
378
+ num_query_groups ................................ 8
379
+ num_virtual_stages_per_pipeline_rank ............ None
380
+ num_workers ..................................... 2
381
+ object_storage_cache_path ....................... None
382
+ one_logger_async ................................ False
383
+ one_logger_project .............................. megatron-lm
384
+ one_logger_run_name ............................. None
385
+ onnx_safe ....................................... None
386
+ openai_gelu ..................................... False
387
+ optimizer ....................................... adam
388
+ optimizer_cpu_offload ........................... False
389
+ optimizer_offload_fraction ...................... 1.0
390
+ output_bert_embeddings .......................... False
391
+ overlap_cpu_optimizer_d2h_h2d ................... False
392
+ overlap_grad_reduce ............................. True
393
+ overlap_moe_expert_parallel_comm ................ False
394
+ overlap_p2p_comm ................................ False
395
+ overlap_p2p_comm_warmup_flush ................... False
396
+ overlap_param_gather ............................ True
397
+ overlap_param_gather_with_optimizer_step ........ False
398
+ override_opt_param_scheduler .................... False
399
+ params_dtype .................................... torch.bfloat16
400
+ patch_dim ....................................... 16
401
+ per_split_data_args_path ........................ None
402
+ perform_initialization .......................... True
403
+ pin_cpu_grads ................................... True
404
+ pin_cpu_params .................................. True
405
+ pipeline_model_parallel_comm_backend ............ None
406
+ pipeline_model_parallel_layout .................. None
407
+ pipeline_model_parallel_size .................... 1
408
+ position_embedding_type ......................... rope
409
+ pretrained_checkpoint ........................... None
410
+ profile ......................................... True
411
+ profile_ranks ................................... [0, 1, 2, 3, 4, 5, 6, 7]
412
+ profile_step_end ................................ 8
413
+ profile_step_start .............................. 6
414
+ q_lora_rank ..................................... None
415
+ qk_head_dim ..................................... 128
416
+ qk_l2_norm ...................................... False
417
+ qk_layernorm .................................... False
418
+ qk_pos_emb_head_dim ............................. 64
419
+ query_in_block_prob ............................. 0.1
420
+ rampup_batch_size ............................... None
421
+ rank ............................................ 0
422
+ recompute_granularity ........................... None
423
+ recompute_method ................................ None
424
+ recompute_modules ............................... None
425
+ recompute_num_layers ............................ None
426
+ record_memory_history ........................... False
427
+ relative_attention_max_distance ................. 128
428
+ relative_attention_num_buckets .................. 32
429
+ replication ..................................... False
430
+ replication_factor .............................. 2
431
+ replication_jump ................................ None
432
+ rerun_mode ...................................... validate_results
433
+ reset_attention_mask ............................ False
434
+ reset_position_ids .............................. False
435
+ result_rejected_tracker_filename ................ None
436
+ retriever_report_topk_accuracies ................ []
437
+ retriever_score_scaling ......................... False
438
+ retriever_seq_length ............................ 256
439
+ retro_add_retriever ............................. False
440
+ retro_attention_gate ............................ 1
441
+ retro_cyclic_train_iters ........................ None
442
+ retro_encoder_attention_dropout ................. 0.1
443
+ retro_encoder_hidden_dropout .................... 0.1
444
+ retro_encoder_layers ............................ 2
445
+ retro_num_neighbors ............................. 2
446
+ retro_num_retrieved_chunks ...................... 2
447
+ retro_project_dir ............................... None
448
+ retro_verify_neighbor_count ..................... True
449
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
450
+ rope_scaling_factor ............................. 8.0
451
+ rope_type ....................................... None
452
+ rotary_base ..................................... 1000000
453
+ rotary_interleaved .............................. False
454
+ rotary_percent .................................. 1.0
455
+ rotary_scaling_factor ........................... 1.0
456
+ rotary_seq_len_interpolation_factor ............. None
457
+ run_workload_inspector_server ................... False
458
+ sample_rate ..................................... 1.0
459
+ save ............................................ None
460
+ save_interval ................................... None
461
+ save_retain_interval ............................ None
462
+ scatter_gather_tensors_in_pipeline .............. True
463
+ seed ............................................ 1234
464
+ seq_length ...................................... 4096
465
+ sequence_parallel ............................... False
466
+ sft ............................................. False
467
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
468
+ sgd_momentum .................................... 0.9
469
+ short_seq_prob .................................. 0.1
470
+ skip_train ...................................... False
471
+ skipped_train_samples ........................... 0
472
+ spec ............................................ None
473
+ split ........................................... 100,0,0
474
+ squared_relu .................................... False
475
+ start_weight_decay .............................. 0.1
476
+ straggler_ctrlr_port ............................ 65535
477
+ straggler_minmax_count .......................... 1
478
+ suggested_communication_unit_size ............... None
479
+ swiglu .......................................... True
480
+ swin_backbone_type .............................. tiny
481
+ symmetric_ar_type ............................... None
482
+ te_rng_tracker .................................. True
483
+ tensor_model_parallel_size ...................... 1
484
+ tensorboard_dir ................................. None
485
+ tensorboard_log_interval ........................ 1
486
+ tensorboard_queue_size .......................... 1000
487
+ test_data_path .................................. None
488
+ test_mode ....................................... False
489
+ tiktoken_num_special_tokens ..................... 1000
490
+ tiktoken_pattern ................................ None
491
+ tiktoken_special_tokens ......................... None
492
+ timing_log_level ................................ 0
493
+ timing_log_option ............................... minmax
494
+ titles_data_path ................................ None
495
+ tokenizer_model ................................. None
496
+ tokenizer_type .................................. GPT2BPETokenizer
497
+ torch_fsdp2_reshard_after_forward ............... True
498
+ tp_comm_bootstrap_backend ....................... nccl
499
+ tp_comm_bulk_dgrad .............................. True
500
+ tp_comm_bulk_wgrad .............................. True
501
+ tp_comm_overlap ................................. False
502
+ tp_comm_overlap_ag .............................. True
503
+ tp_comm_overlap_cfg ............................. None
504
+ tp_comm_overlap_rs .............................. True
505
+ tp_comm_overlap_rs_dgrad ........................ False
506
+ tp_comm_split_ag ................................ True
507
+ tp_comm_split_rs ................................ True
508
+ train_data_path ................................. None
509
+ train_iters ..................................... 10
510
+ train_samples ................................... None
511
+ train_sync_interval ............................. None
512
+ transformer_impl ................................ transformer_engine
513
+ transformer_pipeline_model_parallel_size ........ 1
514
+ untie_embeddings_and_output_weights ............. True
515
+ use_checkpoint_args ............................. False
516
+ use_checkpoint_opt_param_scheduler .............. False
517
+ use_cpu_initialization .......................... None
518
+ use_custom_fsdp ................................. False
519
+ use_dist_ckpt ................................... True
520
+ use_dist_ckpt_deprecated ........................ False
521
+ use_distributed_optimizer ....................... True
522
+ use_flash_attn .................................. False
523
+ use_legacy_models ............................... False
524
+ use_mp_args_from_checkpoint_args ................ False
525
+ use_one_sent_docs ............................... False
526
+ use_persistent_ckpt_worker ...................... False
527
+ use_precision_aware_optimizer ................... False
528
+ use_pytorch_profiler ............................ False
529
+ use_ring_exchange_p2p ........................... False
530
+ use_rope_scaling ................................ False
531
+ use_rotary_position_embeddings .................. False
532
+ use_sharp ....................................... False
533
+ use_tokenizer_model_from_checkpoint_args ........ True
534
+ use_torch_fsdp2 ................................. False
535
+ use_torch_optimizer_for_cpu_offload ............. False
536
+ use_tp_pp_dp_mapping ............................ False
537
+ v_head_dim ...................................... 128
538
+ valid_data_path ................................. None
539
+ variable_seq_lengths ............................ False
540
+ virtual_pipeline_model_parallel_size ............ None
541
+ vision_backbone_type ............................ vit
542
+ vision_pretraining .............................. False
543
+ vision_pretraining_type ......................... classify
544
+ vocab_extra_ids ................................. 0
545
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
546
+ vocab_size ...................................... None
547
+ wandb_exp_name ..................................
548
+ wandb_project ...................................
549
+ wandb_save_dir ..................................
550
+ weight_decay .................................... 0.1
551
+ weight_decay_incr_style ......................... constant
552
+ wgrad_deferral_limit ............................ 0
553
+ world_size ...................................... 8
554
+ yaml_cfg ........................................ None
555
+ -------------------- end of arguments ---------------------
556
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
557
+ > building GPT2BPETokenizer tokenizer ...
558
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
559
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
560
+ > initializing torch distributed ...
561
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
562
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/trace_results.tsv CHANGED
@@ -3,3 +3,5 @@ G2_K2 echo 1.15625 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tie
3
  G3_K2 echo 1.1875 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G3_K2__echo__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__echo__matched__seq4096__gbs16__cap1p1875.nsys-rep
4
  G3_K8 echo 1.09375 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G3_K8__echo__matched__seq4096__gbs16__cap1p09375/nsys/G3_K8__echo__matched__seq4096__gbs16__cap1p09375.nsys-rep
5
  G4_K2 echo 1.265625 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K2__echo__matched__seq4096__gbs16__cap1p265625/nsys/G4_K2__echo__matched__seq4096__gbs16__cap1p265625.nsys-rep
 
 
 
3
  G3_K2 echo 1.1875 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G3_K2__echo__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__echo__matched__seq4096__gbs16__cap1p1875.nsys-rep
4
  G3_K8 echo 1.09375 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G3_K8__echo__matched__seq4096__gbs16__cap1p09375/nsys/G3_K8__echo__matched__seq4096__gbs16__cap1p09375.nsys-rep
5
  G4_K2 echo 1.265625 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K2__echo__matched__seq4096__gbs16__cap1p265625/nsys/G4_K2__echo__matched__seq4096__gbs16__cap1p265625.nsys-rep
6
+ G4_K4 echo 1.171875 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K4__echo__matched__seq4096__gbs16__cap1p171875/nsys/G4_K4__echo__matched__seq4096__gbs16__cap1p171875.nsys-rep
7
+ G4_K6 echo 1.1354166666666667 OK /workspace/granularity_next_b200/tiera_topk_trace/echo/tierbtrace/tierbtrace_gbs16/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667/nsys/G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667.nsys-rep
granularity_next_b200/tiera_topk_trace/hf_watcher.log CHANGED
@@ -16,3 +16,6 @@ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
16
  uploaded: /workspace/granularity_next_b200/tiera_topk_trace -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/granularity_next_b200/tiera_topk_trace
17
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
18
  [2026-09-17T02:44:49+00:00] backup pass
 
 
 
 
16
  uploaded: /workspace/granularity_next_b200/tiera_topk_trace -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/granularity_next_b200/tiera_topk_trace
17
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
18
  [2026-09-17T02:44:49+00:00] backup pass
19
+ uploaded: /workspace/granularity_next_b200/tiera_topk_trace -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/granularity_next_b200/tiera_topk_trace
20
+ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
21
+ [2026-09-17T02:54:52+00:00] backup pass
granularity_next_b200/tiera_topk_trace/study.log CHANGED
@@ -50,3 +50,13 @@ driver_repo: /workspace/blackwell_run_scripts @ 10756d4
50
  [guard] iter 8/10 (90s)
51
  [guard] no progress for 420s at iter 8/10; treating as done
52
  [guard] nsys stop (flushing report)
 
 
 
 
 
 
 
 
 
 
 
50
  [guard] iter 8/10 (90s)
51
  [guard] no progress for 420s at iter 8/10; treating as done
52
  [guard] nsys stop (flushing report)
53
+ [guard] tearing down process group (outcome=STALLED)
54
+ [02:45:17] trace verified -> G4_K4__echo__matched__seq4096__gbs16__cap1p171875.nsys-rep (20M, all 8 GPUs + echo markers)
55
+ [02:45:17] TRACE-ONLY G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667 (ranks: 0 1 2 3 4 5 6 7)
56
+ [guard] iter 6/10 (80s)
57
+ [guard] iter 8/10 (90s)
58
+ [guard] no progress for 420s at iter 8/10; treating as done
59
+ [guard] nsys stop (flushing report)
60
+ [guard] tearing down process group (outcome=STALLED)
61
+ [02:54:36] trace verified -> G4_K6__echo__matched__seq4096__gbs16__cap1p1354166666666667.nsys-rep (19M, all 8 GPUs + echo markers)
62
+ [02:54:36] TRACE-ONLY G4_K8__echo__matched__seq4096__gbs16__cap1p109375 (ranks: 0 1 2 3 4 5 6 7)