AbhishekGhosh1998 commited on
Commit
91b144d
·
verified ·
1 Parent(s): 7ea7007

granularity results tiera-rerun-fixed-moonep @ 2026-08-30T06:09:23Z from sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0

Browse files
Files changed (17) hide show
  1. .gitattributes +1 -0
  2. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/nsys/train.log +44 -0
  3. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/G2__echo__one__seq4096__gbs16__cap1p08203125.nsys-rep +3 -0
  4. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/run.env +70 -0
  5. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/train.log +1106 -0
  6. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1/mem_samples.csv +256 -0
  7. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1/run.env +70 -0
  8. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1/train.log +1107 -0
  9. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2/mem_samples.csv +224 -0
  10. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2/run.env +70 -0
  11. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2/train.log +1143 -0
  12. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3/mem_samples.csv +248 -0
  13. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3/run.env +70 -0
  14. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3/train.log +1107 -0
  15. tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/results.tsv +3 -0
  16. tiera-rerun-fixed-moonep/hf_watcher.log +3 -0
  17. tiera-rerun-fixed-moonep/study.log +15 -0
.gitattributes CHANGED
@@ -69,3 +69,4 @@ tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_30B__moonep__mat
69
  tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
70
  tiera-rerun-fixed-moonep/echo/tierbprobe/tierbprobe_gbs16/G4__echo__matched__seq4096__gbs16__cap1p068359375__forcebal/rep1/train.log filter=lfs diff=lfs merge=lfs -text
71
  tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/nsys/G1__echo__one__seq4096__gbs16__cap2p40625.nsys-rep filter=lfs diff=lfs merge=lfs -text
 
 
69
  tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
70
  tiera-rerun-fixed-moonep/echo/tierbprobe/tierbprobe_gbs16/G4__echo__matched__seq4096__gbs16__cap1p068359375__forcebal/rep1/train.log filter=lfs diff=lfs merge=lfs -text
71
  tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/nsys/G1__echo__one__seq4096__gbs16__cap2p40625.nsys-rep filter=lfs diff=lfs merge=lfs -text
72
+ tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/G2__echo__one__seq4096__gbs16__cap1p08203125.nsys-rep filter=lfs diff=lfs merge=lfs -text
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/nsys/train.log CHANGED
@@ -982,3 +982,47 @@ Capture range ended in the application.
982
  Generating '/tmp/nsys-report-cfc2.qdstrm'
983
 
984
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
982
  Generating '/tmp/nsys-report-cfc2.qdstrm'
983
 
984
 
985
+ Collection stop is not allowed in this state.
986
+ W0830 06:02:08.796000 555662 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
987
+ W0830 06:02:08.796000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555774 closing signal SIGTERM
988
+ Generated:
989
+ /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/nsys/G1__echo__one__seq4096__gbs16__cap2p40625.nsys-rep
990
+ W0830 06:02:08.799000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555775 closing signal SIGTERM
991
+ W0830 06:02:08.803000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555776 closing signal SIGTERM
992
+ W0830 06:02:08.806000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555777 closing signal SIGTERM
993
+ W0830 06:02:08.814000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555778 closing signal SIGTERM
994
+ W0830 06:02:08.823000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555779 closing signal SIGTERM
995
+ W0830 06:02:08.826000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555780 closing signal SIGTERM
996
+ W0830 06:02:08.833000 555662 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 555781 closing signal SIGTERM
997
+ Traceback (most recent call last):
998
+ File "/usr/local/bin/torchrun", line 8, in <module>
999
+ sys.exit(main())
1000
+ ^^^^^^
1001
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1002
+ return f(*args, **kwargs)
1003
+ ^^^^^^^^^^^^^^^^^^
1004
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1005
+ run(args)
1006
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1007
+ elastic_launch(
1008
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1009
+ return launch_agent(self._config, self._entrypoint, list(args))
1010
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1011
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1012
+ result = agent.run()
1013
+ ^^^^^^^^^^^
1014
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1015
+ result = f(*args, **kwargs)
1016
+ ^^^^^^^^^^^^^^^^^^
1017
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1018
+ result = self._invoke_run(role)
1019
+ ^^^^^^^^^^^^^^^^^^^^^^
1020
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1021
+ time.sleep(monitor_interval)
1022
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1023
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1024
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 555662 got signal: 15
1025
+ The target application terminated. One or more process it created re-parented.
1026
+ Waiting for termination of re-parented processes.
1027
+ Use the `--wait` option to modify this behavior.
1028
+ There are no active sessions.
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/G2__echo__one__seq4096__gbs16__cap1p08203125.nsys-rep ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38463bd88b8b76ead2365f7a654a95def82c20e271ae8696d220e9ee30a908be
3
+ size 15407944
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=echo
2
+ config=G2
3
+ family=tierA
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=bracing-regions-through-kernels
6
+ commit=9a236793
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=3584
11
+ num_experts=32
12
+ topk=8
13
+ experts_per_rank=4
14
+ expert_params=44040192
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=32768
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.08203125
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=10
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=
59
+ ultraep_weight_sync_join=
60
+ profile_enable=1
61
+ kernel_trace=1
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/train.log ADDED
@@ -0,0 +1,1106 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G2 / echo (slots/rank=1 policy=one)
3
+ E=32 K=8 H'=3584 m=4
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.08203125 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/nsys/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 9a236793
8
+ ==============================================================
9
+ W0830 06:06:34.849000 574991 torch/distributed/run.py:766]
10
+ W0830 06:06:34.849000 574991 torch/distributed/run.py:766] *****************************************
11
+ W0830 06:06:34.849000 574991 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 06:06:34.849000 574991 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
+ 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
16
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
17
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
18
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
19
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
20
+ Number of virtual stages per pipeline stage: None
21
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
22
+ using torch.bfloat16 for parameters ...
23
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
24
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
25
+ ------------------------ arguments ------------------------
26
+ account_for_embedding_in_pipeline_split ......... False
27
+ account_for_loss_in_pipeline_split .............. False
28
+ accumulate_allreduce_grads_in_fp32 .............. True
29
+ adam_beta1 ...................................... 0.9
30
+ adam_beta2 ...................................... 0.999
31
+ adam_eps ........................................ 1e-08
32
+ add_bias_linear ................................. False
33
+ add_position_embedding .......................... False
34
+ add_qkv_bias .................................... False
35
+ adlr_autoresume ................................. False
36
+ adlr_autoresume_interval ........................ 1000
37
+ align_grad_reduce ............................... True
38
+ align_param_gather .............................. False
39
+ app_tag_run_name ................................ None
40
+ app_tag_run_version ............................. 0.0.0
41
+ apply_layernorm_1p .............................. False
42
+ apply_query_key_layer_scaling ................... False
43
+ apply_residual_connection_post_layernorm ........ False
44
+ apply_rope_fusion ............................... True
45
+ async_save ...................................... None
46
+ async_tensor_model_parallel_allreduce ........... True
47
+ attention_backend ............................... AttnBackend.auto
48
+ attention_dropout ............................... 0.0
49
+ attention_softmax_in_fp32 ....................... False
50
+ auto_detect_ckpt_format ......................... False
51
+ barrier_with_L1_time ............................ True
52
+ bert_binary_head ................................ True
53
+ bert_embedder_type .............................. megatron
54
+ bert_load ....................................... None
55
+ bf16 ............................................ True
56
+ bias_dropout_fusion ............................. True
57
+ bias_gelu_fusion ................................ False
58
+ bias_swiglu_fusion .............................. True
59
+ biencoder_projection_dim ........................ 0
60
+ biencoder_shared_query_context_model ............ False
61
+ block_data_path ................................. None
62
+ calc_ft_timeouts ................................ False
63
+ calculate_per_token_loss ........................ False
64
+ check_for_large_grads ........................... False
65
+ check_for_nan_in_loss_and_grad .................. False
66
+ check_for_spiky_loss ............................ False
67
+ check_weight_hash_across_dp_replicas_interval ... None
68
+ ckpt_assume_constant_structure .................. False
69
+ ckpt_convert_format ............................. None
70
+ ckpt_convert_save ............................... None
71
+ ckpt_convert_update_legacy_dist_opt_format ...... False
72
+ ckpt_format ..................................... torch_dist
73
+ ckpt_fully_parallel_load ........................ False
74
+ ckpt_fully_parallel_save ........................ True
75
+ ckpt_fully_parallel_save_deprecated ............. False
76
+ ckpt_step ....................................... None
77
+ classes_fraction ................................ 1.0
78
+ clip_grad ....................................... 1.0
79
+ clone_scatter_output_in_embedding ............... True
80
+ config_logger_dir ...............................
81
+ consumed_train_samples .......................... 0
82
+ consumed_valid_samples .......................... 0
83
+ context_parallel_size ........................... 1
84
+ cp_comm_type .................................... ['p2p']
85
+ create_attention_mask_in_dataloader ............. True
86
+ cross_entropy_fusion_impl ....................... native
87
+ cross_entropy_loss_fusion ....................... False
88
+ cuda_graph_scope ................................ full_iteration
89
+ cuda_graph_warmup_steps ......................... 3
90
+ data_args_path .................................. None
91
+ data_cache_path ................................. None
92
+ data_parallel_random_init ....................... False
93
+ data_parallel_sharding_strategy ................. no_shard
94
+ data_parallel_size .............................. 8
95
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
96
+ data_per_class_fraction ......................... 1.0
97
+ data_sharding ................................... True
98
+ dataloader_type ................................. single
99
+ ddp_average_in_collective ....................... False
100
+ ddp_bucket_size ................................. None
101
+ ddp_num_buckets ................................. None
102
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
103
+ decoder_first_pipeline_num_layers ............... None
104
+ decoder_last_pipeline_num_layers ................ None
105
+ decoder_num_layers .............................. None
106
+ decoder_seq_length .............................. None
107
+ decoupled_lr .................................... None
108
+ decoupled_min_lr ................................ None
109
+ decrease_batch_size_if_needed ................... False
110
+ defer_embedding_wgrad_compute ................... False
111
+ delay_wgrad_compute ............................. False
112
+ deprecated_use_mcore_models ..................... True
113
+ deterministic_mode .............................. False
114
+ dino_bottleneck_size ............................ 256
115
+ dino_freeze_last_layer .......................... 1
116
+ dino_head_hidden_size ........................... 2048
117
+ dino_local_crops_number ......................... 10
118
+ dino_local_img_size ............................. 96
119
+ dino_norm_last_layer ............................ False
120
+ dino_teacher_temp ............................... 0.07
121
+ dino_warmup_teacher_temp ........................ 0.04
122
+ dino_warmup_teacher_temp_epochs ................. 30
123
+ disable_bf16_reduced_precision_matmul ........... False
124
+ disable_mamba_mem_eff_path ...................... False
125
+ disable_straggler_on_startup .................... False
126
+ dist_ckpt_format_deprecated ..................... None
127
+ dist_ckpt_strictness ............................ assume_ok_unexpected
128
+ distribute_saved_activations .................... False
129
+ distributed_backend ............................. nccl
130
+ distributed_timeout_minutes ..................... 10
131
+ embedding_init_method_std ....................... None
132
+ embedding_path .................................. None
133
+ empty_unused_memory_level ....................... 0
134
+ enable_cuda_graph ............................... True
135
+ enable_experimental ............................. False
136
+ enable_ft_package ............................... False
137
+ enable_gloo_process_groups ...................... True
138
+ enable_msc ...................................... True
139
+ enable_one_logger ............................... True
140
+ encoder_num_layers .............................. 16
141
+ encoder_seq_length .............................. 4096
142
+ end_weight_decay ................................ 0.1
143
+ eod_mask_loss ................................... False
144
+ error_injection_rate ............................ 0
145
+ error_injection_type ............................ transient_error
146
+ eval_interval ................................... 1000000
147
+ eval_iters ...................................... 0
148
+ evidence_data_path .............................. None
149
+ exit_duration_in_mins ........................... None
150
+ exit_interval ................................... None
151
+ exit_on_missing_checkpoint ...................... False
152
+ exit_signal_handler ............................. False
153
+ exp_avg_dtype ................................... torch.float32
154
+ exp_avg_sq_dtype ................................ torch.float32
155
+ expert_model_parallel_size ...................... 8
156
+ expert_tensor_parallel_size ..................... 1
157
+ export_force_local_attention .................... False
158
+ export_kd_cfg ................................... None
159
+ export_kd_teacher_ckpt_format ................... None
160
+ export_kd_teacher_load .......................... None
161
+ export_kv_cache_quant ........................... False
162
+ export_legacy_megatron .......................... False
163
+ export_model_type ............................... GPTModel
164
+ export_moe_apply_probs_on_input ................. False
165
+ export_qk_l2_norm ............................... False
166
+ export_quant_cfg ................................ None
167
+ export_real_quant_cfg ........................... None
168
+ export_te_mcore_model ........................... False
169
+ external_cuda_graph ............................. False
170
+ ffn_hidden_size ................................. 14336
171
+ finetune ........................................ False
172
+ finetune_data_split ............................. train
173
+ finetune_hf_dataset ............................. None
174
+ first_last_layers_bf16 .......................... False
175
+ flash_decode .................................... False
176
+ fp16 ............................................ False
177
+ fp16_lm_cross_entropy ........................... False
178
+ fp32_residual_connection ........................ False
179
+ fp8 ............................................. e4m3
180
+ fp8_amax_compute_algo ........................... most_recent
181
+ fp8_amax_history_len ............................ 1
182
+ fp8_interval .................................... 1
183
+ fp8_margin ...................................... 0
184
+ fp8_param_gather ................................ False
185
+ fp8_recipe ...................................... mxfp8
186
+ fp8_wgrad ....................................... True
187
+ fsdp_double_buffer .............................. False
188
+ global_batch_size ............................... 16
189
+ grad_reduce_in_bf16 ............................. False
190
+ gradient_accumulation_fusion .................... True
191
+ gradient_reduce_div_fusion ...................... True
192
+ group_query_attention ........................... True
193
+ head_lr_mult .................................... 1.0
194
+ heterogeneous_layers_config_encoded_json ........ None
195
+ heterogeneous_layers_config_path ................ None
196
+ hidden_dropout .................................. 0.0
197
+ hidden_size ..................................... 4096
198
+ hierarchical_context_parallel_sizes ............. None
199
+ high_priority_stream_groups ..................... []
200
+ hybrid_attention_ratio .......................... 0.0
201
+ hybrid_mlp_ratio ................................ 0.0
202
+ hybrid_override_pattern ......................... None
203
+ hysteresis ...................................... 2
204
+ ict_head_size ................................... None
205
+ ict_load ........................................ None
206
+ img_h ........................................... 224
207
+ img_w ........................................... 224
208
+ indexer_batch_size .............................. 128
209
+ indexer_log_interval ............................ 1000
210
+ inference_batch_times_seqlen_threshold .......... -1
211
+ inference_dynamic_batching ...................... False
212
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
213
+ inference_dynamic_batching_buffer_overflow_factor None
214
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
215
+ inference_dynamic_batching_chunk_size ........... 256
216
+ inference_dynamic_batching_max_requests_override None
217
+ inference_dynamic_batching_max_tokens_override .. None
218
+ inference_dynamic_batching_num_cuda_graphs ...... 16
219
+ inference_max_batch_size ........................ 8
220
+ inference_max_seq_length ........................ 2560
221
+ inference_rng_tracker ........................... False
222
+ init_method_std ................................. 0.01
223
+ init_method_xavier_uniform ...................... False
224
+ init_model_with_meta_device ..................... False
225
+ initial_loss_scale .............................. 4294967296
226
+ inprocess_active_world_size ..................... 8
227
+ inprocess_barrier_timeout ....................... 120
228
+ inprocess_completion_timeout .................... 120
229
+ inprocess_empty_cuda_cache ...................... False
230
+ inprocess_granularity ........................... node
231
+ inprocess_hard_timeout .......................... 90
232
+ inprocess_heartbeat_interval .................... 30
233
+ inprocess_heartbeat_timeout ..................... 60
234
+ inprocess_last_call_wait ........................ 1
235
+ inprocess_max_iterations ........................ None
236
+ inprocess_monitor_process_interval .............. 1.0
237
+ inprocess_monitor_thread_interval ............... 1.0
238
+ inprocess_progress_watchdog_interval ............ 1.0
239
+ inprocess_restart ............................... False
240
+ inprocess_soft_timeout .......................... 60
241
+ inprocess_termination_grace_time ................ 1
242
+ is_hybrid_model ................................. False
243
+ iter_per_epoch .................................. 1250
244
+ iterations_to_skip .............................. []
245
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
246
+ kitchen_config_file ............................. None
247
+ kitchen_recipe_number ........................... None
248
+ kv_channels ..................................... 128
249
+ kv_lora_rank .................................... 32
250
+ lazy_mpu_init ................................... None
251
+ load ............................................ None
252
+ load_main_params_from_ckpt ...................... None
253
+ load_model_opt_format ........................... False
254
+ local_rank ...................................... 0
255
+ log_energy ...................................... False
256
+ log_interval .................................... 1
257
+ log_loss_scale_to_tensorboard ................... True
258
+ log_memory_to_tensorboard ....................... False
259
+ log_num_zeros_in_grad ........................... False
260
+ log_params_norm ................................. False
261
+ log_progress .................................... False
262
+ log_straggler ................................... False
263
+ log_throughput .................................. True
264
+ log_timers_to_tensorboard ....................... False
265
+ log_validation_ppl_to_tensorboard ............... False
266
+ log_world_size_to_tensorboard ................... False
267
+ logging_level ................................... None
268
+ loss_scale ...................................... None
269
+ loss_scale_window ............................... 1000
270
+ lr .............................................. 0.0001
271
+ lr_decay_iters .................................. 8
272
+ lr_decay_samples ................................ None
273
+ lr_decay_style .................................. cosine
274
+ lr_warmup_fraction .............................. None
275
+ lr_warmup_init .................................. 0.0
276
+ lr_warmup_iters ................................. 5
277
+ lr_warmup_samples ............................... 0
278
+ lr_wsd_decay_iters .............................. None
279
+ lr_wsd_decay_samples ............................ None
280
+ lr_wsd_decay_style .............................. exponential
281
+ main_grads_dtype ................................ torch.float32
282
+ main_params_dtype ............................... torch.float32
283
+ make_vocab_size_divisible_by .................... 128
284
+ mamba_head_dim .................................. 64
285
+ mamba_num_groups ................................ 8
286
+ mamba_num_heads ................................. None
287
+ mamba_state_dim ................................. 128
288
+ manual_gc ....................................... False
289
+ manual_gc_eval .................................. True
290
+ manual_gc_interval .............................. 0
291
+ mask_factor ..................................... 1.0
292
+ mask_prob ....................................... 0.15
293
+ mask_type ....................................... random
294
+ masked_softmax_fusion ........................... False
295
+ max_position_embeddings ......................... 32768
296
+ max_tokens_to_oom ............................... 12000
297
+ memory_snapshot_path ............................ snapshot.pickle
298
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
299
+ micro_batch_size ................................ 1
300
+ microbatch_group_size_per_vp_stage .............. None
301
+ mid_level_dataset_surplus ....................... 0.005
302
+ min_loss_scale .................................. 1.0
303
+ min_lr .......................................... 1e-05
304
+ mlp_chunks_for_prefill .......................... 1
305
+ mmap_bin_files .................................. True
306
+ mock_data ....................................... False
307
+ moe_apply_probs_on_input ........................ False
308
+ moe_aux_loss_coeff .............................. 0.01
309
+ moe_deepep_num_sms .............................. 20
310
+ moe_echo_enable_random_offloading ............... False
311
+ moe_echo_expert_dispatch_overlap ................ True
312
+ moe_echo_expert_dispatcher_type ................. hybridep
313
+ moe_echo_recompute_expert_dispatch .............. False
314
+ moe_enable_deepep ............................... False
315
+ moe_enable_echo ................................. True
316
+ moe_enable_hybridep ............................. True
317
+ moe_expert_capacity_factor ...................... None
318
+ moe_extended_tp ................................. False
319
+ moe_ffn_hidden_size ............................. 3584
320
+ moe_grouped_gemm ................................ True
321
+ moe_hybridep_num_sms ............................ 16
322
+ moe_input_jitter_eps ............................ None
323
+ moe_layer_freq .................................. 1
324
+ moe_layer_recompute ............................. False
325
+ moe_num_echo_experts ............................ 8
326
+ moe_pad_expert_input_to_capacity ................ False
327
+ moe_per_layer_logging ........................... False
328
+ moe_permute_fusion .............................. False
329
+ moe_received_token_capacity ..................... 1.08203125
330
+ moe_router_bias_update_rate ..................... 0.001
331
+ moe_router_dtype ................................ fp32
332
+ moe_router_enable_expert_bias ................... False
333
+ moe_router_force_load_balancing ................. False
334
+ moe_router_group_topk ........................... None
335
+ moe_router_load_balancing_type .................. aux_loss
336
+ moe_router_num_groups ........................... None
337
+ moe_router_padding_for_fp8 ...................... False
338
+ moe_router_pre_softmax .......................... False
339
+ moe_router_score_function ....................... softmax
340
+ moe_router_topk ................................. 8
341
+ moe_router_topk_scaling_factor .................. None
342
+ moe_shared_expert_intermediate_size ............. None
343
+ moe_shared_expert_overlap ....................... False
344
+ moe_token_dispatcher_type ....................... flex
345
+ moe_token_drop_policy ........................... probs
346
+ moe_upcycling_granularity ....................... 1
347
+ moe_use_device_initiated_grouped_gemm ........... True
348
+ moe_use_legacy_grouped_gemm ..................... False
349
+ moe_use_upcycling ............................... False
350
+ moe_z_loss_coeff ................................ None
351
+ mrope_section ................................... None
352
+ mscale .......................................... 1.0
353
+ mscale_all_dim .................................. 0.0
354
+ mtp_loss_scaling_factor ......................... 0.1
355
+ mtp_num_layers .................................. None
356
+ multi_latent_attention .......................... False
357
+ nccl_all_reduce_for_prefill ..................... False
358
+ nccl_communicator_config_path ................... None
359
+ nccl_ub ......................................... False
360
+ no_load_optim ................................... None
361
+ no_load_rng ..................................... None
362
+ no_persist_layer_norm ........................... False
363
+ no_rope_freq .................................... None
364
+ no_save_optim ................................... None
365
+ no_save_rng ..................................... None
366
+ non_persistent_ckpt_type ........................ None
367
+ non_persistent_global_ckpt_dir .................. None
368
+ non_persistent_local_ckpt_algo .................. fully_parallel
369
+ non_persistent_local_ckpt_dir ................... None
370
+ non_persistent_save_interval .................... None
371
+ norm_epsilon .................................... 1e-05
372
+ normalization ................................... RMSNorm
373
+ num_attention_heads ............................. 32
374
+ num_channels .................................... 3
375
+ num_classes ..................................... 1000
376
+ num_dataset_builder_threads ..................... 1
377
+ num_distributed_optimizer_instances ............. 1
378
+ num_experts ..................................... 32
379
+ num_layers ...................................... 16
380
+ num_layers_at_end_in_bf16 ....................... 1
381
+ num_layers_at_start_in_bf16 ..................... 1
382
+ num_layers_per_virtual_pipeline_stage ........... None
383
+ num_query_groups ................................ 8
384
+ num_virtual_stages_per_pipeline_rank ............ None
385
+ num_workers ..................................... 2
386
+ object_storage_cache_path ....................... None
387
+ one_logger_async ................................ False
388
+ one_logger_project .............................. megatron-lm
389
+ one_logger_run_name ............................. None
390
+ onnx_safe ....................................... None
391
+ openai_gelu ..................................... False
392
+ optimizer ....................................... adam
393
+ optimizer_cpu_offload ........................... False
394
+ optimizer_offload_fraction ...................... 1.0
395
+ output_bert_embeddings .......................... False
396
+ overlap_cpu_optimizer_d2h_h2d ................... False
397
+ overlap_grad_reduce ............................. True
398
+ overlap_moe_expert_parallel_comm ................ False
399
+ overlap_p2p_comm ................................ False
400
+ overlap_p2p_comm_warmup_flush ................... False
401
+ overlap_param_gather ............................ True
402
+ overlap_param_gather_with_optimizer_step ........ False
403
+ override_opt_param_scheduler .................... False
404
+ params_dtype .................................... torch.bfloat16
405
+ patch_dim ....................................... 16
406
+ per_split_data_args_path ........................ None
407
+ perform_initialization .......................... True
408
+ pin_cpu_grads ................................... True
409
+ pin_cpu_params .................................. True
410
+ pipeline_model_parallel_comm_backend ............ None
411
+ pipeline_model_parallel_layout .................. None
412
+ pipeline_model_parallel_size .................... 1
413
+ position_embedding_type ......................... rope
414
+ pretrained_checkpoint ........................... None
415
+ profile ......................................... True
416
+ profile_ranks ................................... [0, 1, 2, 3, 4, 5, 6, 7]
417
+ profile_step_end ................................ 8
418
+ profile_step_start .............................. 6
419
+ q_lora_rank ..................................... None
420
+ qk_head_dim ..................................... 128
421
+ qk_l2_norm ...................................... False
422
+ qk_layernorm .................................... False
423
+ qk_pos_emb_head_dim ............................. 64
424
+ query_in_block_prob ............................. 0.1
425
+ rampup_batch_size ............................... None
426
+ rank ............................................ 0
427
+ recompute_granularity ........................... None
428
+ recompute_method ................................ None
429
+ recompute_modules ............................... None
430
+ recompute_num_layers ............................ None
431
+ record_memory_history ........................... False
432
+ relative_attention_max_distance ................. 128
433
+ relative_attention_num_buckets .................. 32
434
+ replication ..................................... False
435
+ replication_factor .............................. 2
436
+ replication_jump ................................ None
437
+ rerun_mode ...................................... validate_results
438
+ reset_attention_mask ............................ False
439
+ reset_position_ids .............................. False
440
+ result_rejected_tracker_filename ................ None
441
+ retriever_report_topk_accuracies ................ []
442
+ retriever_score_scaling ......................... False
443
+ retriever_seq_length ............................ 256
444
+ retro_add_retriever ............................. False
445
+ retro_attention_gate ............................ 1
446
+ retro_cyclic_train_iters ........................ None
447
+ retro_encoder_attention_dropout ................. 0.1
448
+ retro_encoder_hidden_dropout .................... 0.1
449
+ retro_encoder_layers ............................ 2
450
+ retro_num_neighbors ............................. 2
451
+ retro_num_retrieved_chunks ...................... 2
452
+ retro_project_dir ............................... None
453
+ retro_verify_neighbor_count ..................... True
454
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
455
+ rope_scaling_factor ............................. 8.0
456
+ rope_type ....................................... None
457
+ rotary_base ..................................... 1000000
458
+ rotary_interleaved .............................. False
459
+ rotary_percent .................................. 1.0
460
+ rotary_scaling_factor ........................... 1.0
461
+ rotary_seq_len_interpolation_factor ............. None
462
+ run_workload_inspector_server ................... False
463
+ sample_rate ..................................... 1.0
464
+ save ............................................ None
465
+ save_interval ................................... None
466
+ save_retain_interval ............................ None
467
+ scatter_gather_tensors_in_pipeline .............. True
468
+ seed ............................................ 1234
469
+ seq_length ...................................... 4096
470
+ sequence_parallel ............................... False
471
+ sft ............................................. False
472
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
473
+ sgd_momentum .................................... 0.9
474
+ short_seq_prob .................................. 0.1
475
+ skip_train ...................................... False
476
+ skipped_train_samples ........................... 0
477
+ spec ............................................ None
478
+ split ........................................... 100,0,0
479
+ squared_relu .................................... False
480
+ start_weight_decay .............................. 0.1
481
+ straggler_ctrlr_port ............................ 65535
482
+ straggler_minmax_count .......................... 1
483
+ suggested_communication_unit_size ............... None
484
+ swiglu .......................................... True
485
+ swin_backbone_type .............................. tiny
486
+ symmetric_ar_type ............................... None
487
+ te_rng_tracker .................................. True
488
+ tensor_model_parallel_size ...................... 1
489
+ tensorboard_dir ................................. None
490
+ tensorboard_log_interval ........................ 1
491
+ tensorboard_queue_size .......................... 1000
492
+ test_data_path .................................. None
493
+ test_mode ....................................... False
494
+ tiktoken_num_special_tokens ..................... 1000
495
+ tiktoken_pattern ................................ None
496
+ tiktoken_special_tokens ......................... None
497
+ timing_log_level ................................ 0
498
+ timing_log_option ............................... minmax
499
+ titles_data_path ................................ None
500
+ tokenizer_model ................................. None
501
+ tokenizer_type .................................. GPT2BPETokenizer
502
+ torch_fsdp2_reshard_after_forward ............... True
503
+ tp_comm_bootstrap_backend ....................... nccl
504
+ tp_comm_bulk_dgrad .............................. True
505
+ tp_comm_bulk_wgrad .............................. True
506
+ tp_comm_overlap ................................. False
507
+ tp_comm_overlap_ag .............................. True
508
+ tp_comm_overlap_cfg ............................. None
509
+ tp_comm_overlap_rs .............................. True
510
+ tp_comm_overlap_rs_dgrad ........................ False
511
+ tp_comm_split_ag ................................ True
512
+ tp_comm_split_rs ................................ True
513
+ train_data_path ................................. None
514
+ train_iters ..................................... 10
515
+ train_samples ................................... None
516
+ train_sync_interval ............................. None
517
+ transformer_impl ................................ transformer_engine
518
+ transformer_pipeline_model_parallel_size ........ 1
519
+ untie_embeddings_and_output_weights ............. True
520
+ use_checkpoint_args ............................. False
521
+ use_checkpoint_opt_param_scheduler .............. False
522
+ use_cpu_initialization .......................... None
523
+ use_custom_fsdp ................................. False
524
+ use_dist_ckpt ................................... True
525
+ use_dist_ckpt_deprecated ........................ False
526
+ use_distributed_optimizer ....................... True
527
+ use_flash_attn .................................. False
528
+ use_legacy_models ............................... False
529
+ use_mp_args_from_checkpoint_args ................ False
530
+ use_one_sent_docs ............................... False
531
+ use_persistent_ckpt_worker ...................... False
532
+ use_precision_aware_optimizer ................... False
533
+ use_pytorch_profiler ............................ False
534
+ use_ring_exchange_p2p ........................... False
535
+ use_rope_scaling ................................ False
536
+ use_rotary_position_embeddings .................. False
537
+ use_sharp ....................................... False
538
+ use_tokenizer_model_from_checkpoint_args ........ True
539
+ use_torch_fsdp2 ................................. False
540
+ use_torch_optimizer_for_cpu_offload ............. False
541
+ use_tp_pp_dp_mapping ............................ False
542
+ v_head_dim ...................................... 128
543
+ valid_data_path ................................. None
544
+ variable_seq_lengths ............................ False
545
+ virtual_pipeline_model_parallel_size ............ None
546
+ vision_backbone_type ............................ vit
547
+ vision_pretraining .............................. False
548
+ vision_pretraining_type ......................... classify
549
+ vocab_extra_ids ................................. 0
550
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
551
+ vocab_size ...................................... None
552
+ wandb_exp_name ..................................
553
+ wandb_project ...................................
554
+ wandb_save_dir ..................................
555
+ weight_decay .................................... 0.1
556
+ weight_decay_incr_style ......................... constant
557
+ wgrad_deferral_limit ............................ 0
558
+ world_size ...................................... 8
559
+ yaml_cfg ........................................ None
560
+ -------------------- end of arguments ---------------------
561
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
562
+ > building GPT2BPETokenizer tokenizer ...
563
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
564
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
565
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
566
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
567
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
568
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
569
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
570
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
571
+ > initializing torch distributed ...
572
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
573
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
574
+ /workspace/Megatron-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
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
577
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
578
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
579
+ [Gloo] Rank [Gloo] Rank 45 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
580
+
581
+ [Gloo] Rank [Gloo] Rank 36 is connected to is connected to 77 peer ranks. Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is : 7
582
+
583
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank 0 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 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
593
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
594
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
595
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
596
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
597
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [rank5]:[W830 06:06:48.697220859 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.
604
+ [rank1]:[W830 06:06:48.697233451 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.
605
+ [rank4]:[W830 06:06:48.697237132 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.
606
+ [rank2]:[W830 06:06:48.697284206 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.
607
+ [rank7]:[W830 06:06:48.701312795 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.
608
+ [rank6]:[W830 06:06:48.701367667 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.
609
+ [rank3]:[W830 06:06:48.701409986 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ > compiling dataset index builder ...
611
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
612
+ make: Nothing to be done for 'default'.
613
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
614
+ >>> done with dataset index builder. Compilation time: 0.425 seconds
615
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
616
+ > compiling and loading fused kernels ...
617
+ [rank0]:[W830 06:06:49.345557786 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.
618
+ >>> done with compiling and loading fused kernels. Compilation time: 3.478 seconds
619
+ time to initialize megatron (seconds): 9.029
620
+ [after megatron is initialized] datetime: 2026-08-30 06:06:54
621
+ building GPT model ...
622
+ /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.
623
+ warnings.warn(
624
+ /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.
625
+ warnings.warn(
626
+ /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.
627
+ warnings.warn(
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3903983616
639
+ 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)
640
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
641
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
642
+ module.output_layer.weight
643
+ Params for bucket 2 (42086400 elements, 42086400 padded size):
644
+ module.decoder.final_layernorm.weight
645
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
646
+ module.decoder.layers.15.mlp.router.weight
647
+ module.decoder.layers.15.self_attention.linear_proj.weight
648
+ module.decoder.layers.15.self_attention.linear_qkv.weight
649
+ module.decoder.layers.15.pre_mlp_layernorm.weight
650
+ Params for bucket 3 (42082304 elements, 42082304 padded size):
651
+ module.decoder.layers.14.self_attention.linear_qkv.weight
652
+ module.decoder.layers.14.pre_mlp_layernorm.weight
653
+ module.decoder.layers.14.mlp.router.weight
654
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
655
+ module.decoder.layers.14.self_attention.linear_proj.weight
656
+ Params for bucket 4 (42082304 elements, 42082304 padded size):
657
+ module.decoder.layers.13.mlp.router.weight
658
+ module.decoder.layers.13.self_attention.linear_proj.weight
659
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
660
+ module.decoder.layers.13.pre_mlp_layernorm.weight
661
+ module.decoder.layers.13.self_attention.linear_qkv.weight
662
+ Params for bucket 5 (42082304 elements, 42082304 padded size):
663
+ module.decoder.layers.12.self_attention.linear_qkv.weight
664
+ module.decoder.layers.12.pre_mlp_layernorm.weight
665
+ module.decoder.layers.12.mlp.router.weight
666
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
667
+ module.decoder.layers.12.self_attention.linear_proj.weight
668
+ Params for bucket 6 (42082304 elements, 42082304 padded size):
669
+ module.decoder.layers.11.mlp.router.weight
670
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
671
+ module.decoder.layers.11.self_attention.linear_proj.weight
672
+ module.decoder.layers.11.self_attention.linear_qkv.weight
673
+ module.decoder.layers.11.pre_mlp_layernorm.weight
674
+ Params for bucket 7 (42082304 elements, 42082304 padded size):
675
+ module.decoder.layers.10.self_attention.linear_qkv.weight
676
+ module.decoder.layers.10.pre_mlp_layernorm.weight
677
+ module.decoder.layers.10.mlp.router.weight
678
+ module.decoder.layers.10.self_attention.linear_proj.weight
679
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
680
+ Params for bucket 8 (42082304 elements, 42082304 padded size):
681
+ module.decoder.layers.9.mlp.router.weight
682
+ module.decoder.layers.9.self_attention.linear_proj.weight
683
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
684
+ module.decoder.layers.9.pre_mlp_layernorm.weight
685
+ module.decoder.layers.9.self_attention.linear_qkv.weight
686
+ Params for bucket 9 (42082304 elements, 42082304 padded size):
687
+ module.decoder.layers.8.mlp.router.weight
688
+ module.decoder.layers.8.self_attention.linear_proj.weight
689
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
690
+ module.decoder.layers.8.pre_mlp_layernorm.weight
691
+ module.decoder.layers.8.self_attention.linear_qkv.weight
692
+ Params for bucket 10 (42082304 elements, 42082304 padded size):
693
+ module.decoder.layers.7.self_attention.linear_qkv.weight
694
+ module.decoder.layers.7.pre_mlp_layernorm.weight
695
+ module.decoder.layers.7.mlp.router.weight
696
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
697
+ module.decoder.layers.7.self_attention.linear_proj.weight
698
+ Params for bucket 11 (42082304 elements, 42082304 padded size):
699
+ module.decoder.layers.6.mlp.router.weight
700
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
701
+ module.decoder.layers.6.self_attention.linear_proj.weight
702
+ module.decoder.layers.6.self_attention.linear_qkv.weight
703
+ module.decoder.layers.6.pre_mlp_layernorm.weight
704
+ Params for bucket 12 (42082304 elements, 42082304 padded size):
705
+ module.decoder.layers.5.mlp.router.weight
706
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
707
+ module.decoder.layers.5.self_attention.linear_qkv.weight
708
+ module.decoder.layers.5.self_attention.linear_proj.weight
709
+ module.decoder.layers.5.pre_mlp_layernorm.weight
710
+ Params for bucket 13 (42082304 elements, 42082304 padded size):
711
+ module.decoder.layers.4.self_attention.linear_qkv.weight
712
+ module.decoder.layers.4.pre_mlp_layernorm.weight
713
+ module.decoder.layers.4.mlp.router.weight
714
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
715
+ module.decoder.layers.4.self_attention.linear_proj.weight
716
+ Params for bucket 14 (42082304 elements, 42082304 padded size):
717
+ module.decoder.layers.3.self_attention.linear_qkv.weight
718
+ module.decoder.layers.3.pre_mlp_layernorm.weight
719
+ module.decoder.layers.3.mlp.router.weight
720
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
721
+ module.decoder.layers.3.self_attention.linear_proj.weight
722
+ Params for bucket 15 (42082304 elements, 42082304 padded size):
723
+ module.decoder.layers.2.mlp.router.weight
724
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
725
+ module.decoder.layers.2.self_attention.linear_proj.weight
726
+ module.decoder.layers.2.self_attention.linear_qkv.weight
727
+ module.decoder.layers.2.pre_mlp_layernorm.weight
728
+ Params for bucket 16 (42082304 elements, 42082304 padded size):
729
+ module.decoder.layers.1.pre_mlp_layernorm.weight
730
+ module.decoder.layers.1.self_attention.linear_qkv.weight
731
+ module.decoder.layers.1.self_attention.linear_proj.weight
732
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
733
+ module.decoder.layers.1.mlp.router.weight
734
+ Params for bucket 17 (42082304 elements, 42082304 padded size):
735
+ module.decoder.layers.0.mlp.router.weight
736
+ module.decoder.layers.0.self_attention.linear_qkv.weight
737
+ module.decoder.layers.0.self_attention.linear_proj.weight
738
+ module.decoder.layers.0.pre_mlp_layernorm.weight
739
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
740
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
741
+ module.embedding.word_embeddings.weight
742
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56
743
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
744
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
745
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
746
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
747
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
748
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
750
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
751
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
752
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
753
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
754
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
755
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
756
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
757
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
758
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
759
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
760
+ Params for bucket 6 (58720256 elements, 58720256 padded size):
761
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
762
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
763
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
764
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
765
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
766
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
767
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
768
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
769
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
770
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
771
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
772
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
773
+ Params for bucket 10 (58720256 elements, 58720256 padded size):
774
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
775
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
776
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
777
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
778
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
779
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
780
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
781
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
782
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
783
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
784
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
785
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
786
+ Params for bucket 14 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
788
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
790
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
791
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
792
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
793
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
794
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
795
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
796
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
797
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
798
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
799
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
800
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
801
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
802
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
803
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
804
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
805
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
806
+ Params for bucket 20 (58720256 elements, 58720256 padded size):
807
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
808
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
809
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
811
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
812
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
813
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
814
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
815
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
816
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
817
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
818
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
819
+ Params for bucket 24 (58720256 elements, 58720256 padded size):
820
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
821
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
822
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
823
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
824
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
825
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
827
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
828
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
829
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
830
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
831
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
832
+ Params for bucket 28 (58720256 elements, 58720256 padded size):
833
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
834
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
835
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
836
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
837
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
838
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
839
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
840
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
841
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
842
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
843
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
844
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
845
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
846
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
847
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
848
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
849
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
850
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
851
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
852
+ Params for bucket 34 (58720256 elements, 58720256 padded size):
853
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
854
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
855
+ Params for bucket 35 (58720256 elements, 58720256 padded size):
856
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
857
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
858
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
859
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
860
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
861
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
862
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
863
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
864
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
865
+ Params for bucket 38 (58720256 elements, 58720256 padded size):
866
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
867
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
868
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
869
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
870
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
871
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
872
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
873
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
874
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
875
+ Params for bucket 41 (58720256 elements, 58720256 padded size):
876
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
877
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
878
+ Params for bucket 42 (58720256 elements, 58720256 padded size):
879
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
880
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
881
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
882
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
883
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
884
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
885
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
886
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
887
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
888
+ Params for bucket 45 (58720256 elements, 58720256 padded size):
889
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
890
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
891
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
892
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
893
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
894
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
895
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
896
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
897
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
898
+ Params for bucket 48 (58720256 elements, 58720256 padded size):
899
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
900
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
901
+ Params for bucket 49 (58720256 elements, 58720256 padded size):
902
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
903
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
904
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
905
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
906
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
907
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
908
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
909
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
910
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
911
+ Params for bucket 52 (58720256 elements, 58720256 padded size):
912
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
913
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
914
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
915
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
916
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
917
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
918
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
919
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
920
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
921
+ Params for bucket 55 (58720256 elements, 58720256 padded size):
922
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
923
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
924
+ Params for bucket 56 (58720256 elements, 58720256 padded size):
925
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
926
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
927
+ 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 0x75146ec44830>, config_logger_dir='')
928
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
929
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 06:06:55
930
+ > building train, validation, and test datasets ...
931
+ > datasets target sizes (minimum size):
932
+ train: 160
933
+ validation: 0
934
+ test: 0
935
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
936
+ > building train, validation, and test datasets for GPT ...
937
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(160, 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 0x75146ee56840>, 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)
938
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
939
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
940
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
941
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
942
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
943
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
944
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
945
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-document_index.npy
946
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-sample_index.npy
947
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-shuffle_index.npy
948
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
949
+ > finished creating GPT datasets ...
950
+ [after dataloaders are built] datetime: 2026-08-30 06:06:55
951
+ done with setup ...
952
+ (min, max) time across ranks (ms):
953
+ model-and-optimizer-setup ......................: (344.00, 349.41)
954
+ train/valid/test-data-iterators-setup ..........: (85.40, 108.30)
955
+ training ...
956
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
957
+ [before the start of training step] datetime: 2026-08-30 06:06:55
958
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
959
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
960
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
961
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
962
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
963
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
964
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
965
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
966
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
967
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
968
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
969
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
970
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
971
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
972
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
973
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
974
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
975
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
976
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
977
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
978
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
979
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
980
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
981
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
982
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
983
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
985
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
987
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
989
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
991
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
992
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
993
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
994
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
995
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
996
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
997
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
998
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
999
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1000
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1001
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1002
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1003
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1004
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1005
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1006
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1007
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1008
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1009
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1010
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1011
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1012
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1013
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1014
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1015
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1016
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1017
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1018
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1019
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1020
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1021
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1022
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1023
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1024
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1025
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1026
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1027
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1028
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1029
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1030
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1031
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1032
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1033
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1034
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1035
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1036
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1037
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1038
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1039
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1040
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1041
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1042
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1043
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1044
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1045
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1046
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1047
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1048
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1049
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1050
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1051
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1052
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1053
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1054
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1055
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1056
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1057
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1058
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1059
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1060
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1061
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1062
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1063
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1064
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1065
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1066
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1067
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1068
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1069
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1070
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1071
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1072
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1073
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1074
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1075
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1076
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1077
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1078
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1079
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1080
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1081
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1082
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1083
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1084
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1085
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1086
+ Number of parameters in transformer block in billions: 23.22
1087
+ Number of parameters in embedding layers in billions: 0.41
1088
+ Total number of parameters in billions: 23.63 [2026-08-30 06:07:38] iteration 1/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 42685.7 | throughput per GPU (TFLOP/s/GPU): 7.8 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103012E+01 | loss scale: 1.0 | grad norm: 44.244 | number of skipped iterations: 0 | number of nan iterations: 0 |
1089
+
1090
+ Number of parameters in most loaded shard in billions: 23.6320
1091
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1092
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 59522.919921875 | max allocated: 68963.671875 | reserved: 72832.0 | max reserved: 72832.0
1093
+ [2026-08-30 06:07:38] iteration 2/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 579.1 | throughput per GPU (TFLOP/s/GPU): 575.7 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101632E+01 | loss scale: 1.0 | grad norm: 45.757 | number of skipped iterations: 0 | number of nan iterations: 0 |
1094
+ [2026-08-30 06:07:39] iteration 3/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 515.4 | throughput per GPU (TFLOP/s/GPU): 646.9 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.022607E+01 | loss scale: 1.0 | grad norm: 192.191 | number of skipped iterations: 0 | number of nan iterations: 0 |
1095
+ Capture CUDA graph for training!!!
1096
+ CUDA graph capture done!!!
1097
+ [2026-08-30 06:07:41] iteration 4/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 2403.1 | throughput per GPU (TFLOP/s/GPU): 138.7 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.325139E+01 | loss scale: 1.0 | grad norm: 30.074 | number of skipped iterations: 0 | number of nan iterations: 0 |
1098
+ [2026-08-30 06:07:41] iteration 5/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 397.5 | throughput per GPU (TFLOP/s/GPU): 838.6 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.038957E+01 | loss scale: 1.0 | grad norm: 16.508 | number of skipped iterations: 0 | number of nan iterations: 0 |
1099
+ [2026-08-30 06:07:42] iteration 6/ 10 | consumed samples: 96 | elapsed time per iteration (ms): 396.3 | throughput per GPU (TFLOP/s/GPU): 841.3 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.050723E+01 | loss scale: 1.0 | grad norm: 41.640 | number of skipped iterations: 0 | number of nan iterations: 0 |
1100
+ [2026-08-30 06:07:45] iteration 7/ 10 | consumed samples: 112 | elapsed time per iteration (ms): 3113.2 | throughput per GPU (TFLOP/s/GPU): 107.1 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.010912E+01 | loss scale: 1.0 | grad norm: 8.139 | number of skipped iterations: 0 | number of nan iterations: 0 |
1101
+ [2026-08-30 06:07:45] iteration 8/ 10 | consumed samples: 128 | elapsed time per iteration (ms): 405.8 | throughput per GPU (TFLOP/s/GPU): 821.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.570869E+00 | loss scale: 1.0 | grad norm: 5.901 | number of skipped iterations: 0 | number of nan iterations: 0 |
1102
+ Capture range started in the application.
1103
+ Capture range ended in the application.
1104
+ Generating '/tmp/nsys-report-7467.qdstrm'
1105
+
1106
+
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1/mem_samples.csv ADDED
@@ -0,0 +1,256 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 4
18
+ 1, 2
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 4
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 821
42
+ 1, 821
43
+ 2, 819
44
+ 3, 821
45
+ 4, 819
46
+ 5, 819
47
+ 6, 819
48
+ 7, 821
49
+ 0, 1537
50
+ 1, 1537
51
+ 2, 1537
52
+ 3, 1537
53
+ 4, 1537
54
+ 5, 1537
55
+ 6, 1537
56
+ 7, 1537
57
+ 0, 2231
58
+ 1, 2231
59
+ 2, 2231
60
+ 3, 2231
61
+ 4, 2231
62
+ 5, 2231
63
+ 6, 2231
64
+ 7, 2231
65
+ 0, 39035
66
+ 1, 39085
67
+ 2, 39065
68
+ 3, 39095
69
+ 4, 39113
70
+ 5, 39085
71
+ 6, 39095
72
+ 7, 39099
73
+ 0, 39403
74
+ 1, 39413
75
+ 2, 39437
76
+ 3, 39419
77
+ 4, 39417
78
+ 5, 39417
79
+ 6, 39437
80
+ 7, 39437
81
+ 0, 41893
82
+ 1, 41885
83
+ 2, 41885
84
+ 3, 41885
85
+ 4, 41885
86
+ 5, 41885
87
+ 6, 41885
88
+ 7, 41885
89
+ 0, 41951
90
+ 1, 41943
91
+ 2, 41943
92
+ 3, 41943
93
+ 4, 41943
94
+ 5, 41943
95
+ 6, 41943
96
+ 7, 41943
97
+ 0, 41951
98
+ 1, 41943
99
+ 2, 41943
100
+ 3, 41943
101
+ 4, 41943
102
+ 5, 41943
103
+ 6, 41943
104
+ 7, 41943
105
+ 0, 42865
106
+ 1, 42857
107
+ 2, 42857
108
+ 3, 42857
109
+ 4, 42857
110
+ 5, 42857
111
+ 6, 42857
112
+ 7, 42857
113
+ 0, 43143
114
+ 1, 43135
115
+ 2, 43135
116
+ 3, 43135
117
+ 4, 43135
118
+ 5, 43135
119
+ 6, 43135
120
+ 7, 43135
121
+ 0, 43919
122
+ 1, 43911
123
+ 2, 43911
124
+ 3, 43911
125
+ 4, 43911
126
+ 5, 43911
127
+ 6, 43911
128
+ 7, 43911
129
+ 0, 44567
130
+ 1, 44559
131
+ 2, 44559
132
+ 3, 44559
133
+ 4, 44559
134
+ 5, 44559
135
+ 6, 44559
136
+ 7, 44559
137
+ 0, 67177
138
+ 1, 67155
139
+ 2, 67155
140
+ 3, 67155
141
+ 4, 67169
142
+ 5, 67155
143
+ 6, 67155
144
+ 7, 67155
145
+ 0, 67663
146
+ 1, 67641
147
+ 2, 67641
148
+ 3, 67641
149
+ 4, 67655
150
+ 5, 67641
151
+ 6, 67641
152
+ 7, 67641
153
+ 0, 67663
154
+ 1, 67641
155
+ 2, 67641
156
+ 3, 67641
157
+ 4, 67655
158
+ 5, 67641
159
+ 6, 67641
160
+ 7, 67641
161
+ 0, 67663
162
+ 1, 67641
163
+ 2, 67641
164
+ 3, 67641
165
+ 4, 67655
166
+ 5, 67641
167
+ 6, 67641
168
+ 7, 67641
169
+ 0, 78523
170
+ 1, 77887
171
+ 2, 78813
172
+ 3, 78807
173
+ 4, 78187
174
+ 5, 78827
175
+ 6, 78817
176
+ 7, 77837
177
+ 0, 79247
178
+ 1, 78695
179
+ 2, 79625
180
+ 3, 79595
181
+ 4, 79021
182
+ 5, 79655
183
+ 6, 79679
184
+ 7, 78631
185
+ 0, 80125
186
+ 1, 80103
187
+ 2, 80551
188
+ 3, 80551
189
+ 4, 80453
190
+ 5, 80775
191
+ 6, 80551
192
+ 7, 79991
193
+ 0, 80923
194
+ 1, 80919
195
+ 2, 81373
196
+ 3, 81389
197
+ 4, 81183
198
+ 5, 81643
199
+ 6, 81407
200
+ 7, 80855
201
+ 0, 88861
202
+ 1, 82575
203
+ 2, 84941
204
+ 3, 83875
205
+ 4, 83429
206
+ 5, 86415
207
+ 6, 86205
208
+ 7, 89377
209
+ 0, 119841
210
+ 1, 120227
211
+ 2, 120001
212
+ 3, 120113
213
+ 4, 119665
214
+ 5, 120001
215
+ 6, 120001
216
+ 7, 119889
217
+ 0, 119841
218
+ 1, 120227
219
+ 2, 120001
220
+ 3, 120113
221
+ 4, 119665
222
+ 5, 120001
223
+ 6, 120001
224
+ 7, 119889
225
+ 0, 118929
226
+ 1, 119315
227
+ 2, 119089
228
+ 3, 119201
229
+ 4, 118753
230
+ 5, 119089
231
+ 6, 119089
232
+ 7, 118977
233
+ 0, 2089
234
+ 1, 119315
235
+ 2, 3114
236
+ 3, 1834
237
+ 4, 1834
238
+ 5, 1834
239
+ 6, 1834
240
+ 7, 1834
241
+ 0, 0
242
+ 1, 0
243
+ 2, 0
244
+ 3, 0
245
+ 4, 0
246
+ 5, 0
247
+ 6, 0
248
+ 7, 0
249
+ 0, 0
250
+ 1, 0
251
+ 2, 0
252
+ 3, 0
253
+ 4, 0
254
+ 5, 0
255
+ 6, 0
256
+ 7, 0
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=echo
2
+ config=G2
3
+ family=tierA
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=bracing-regions-through-kernels
6
+ commit=9a236793
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=3584
11
+ num_experts=32
12
+ topk=8
13
+ experts_per_rank=4
14
+ expert_params=44040192
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=32768
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.08203125
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1/train.log ADDED
@@ -0,0 +1,1107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G2 / echo (slots/rank=1 policy=one)
3
+ E=32 K=8 H'=3584 m=4
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.08203125 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 9a236793
8
+ ==============================================================
9
+ W0830 06:02:25.515000 564682 torch/distributed/run.py:766]
10
+ W0830 06:02:25.515000 564682 torch/distributed/run.py:766] *****************************************
11
+ W0830 06:02:25.515000 564682 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 06:02:25.515000 564682 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
+ 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
16
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
17
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
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
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
34
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
35
+ ------------------------ arguments ------------------------
36
+ account_for_embedding_in_pipeline_split ......... False
37
+ account_for_loss_in_pipeline_split .............. False
38
+ accumulate_allreduce_grads_in_fp32 .............. True
39
+ adam_beta1 ...................................... 0.9
40
+ adam_beta2 ...................................... 0.999
41
+ adam_eps ........................................ 1e-08
42
+ add_bias_linear ................................. False
43
+ add_position_embedding .......................... False
44
+ add_qkv_bias .................................... False
45
+ adlr_autoresume ................................. False
46
+ adlr_autoresume_interval ........................ 1000
47
+ align_grad_reduce ............................... True
48
+ align_param_gather .............................. False
49
+ app_tag_run_name ................................ None
50
+ app_tag_run_version ............................. 0.0.0
51
+ apply_layernorm_1p .............................. False
52
+ apply_query_key_layer_scaling ................... False
53
+ apply_residual_connection_post_layernorm ........ False
54
+ apply_rope_fusion ............................... True
55
+ async_save ...................................... None
56
+ async_tensor_model_parallel_allreduce ........... True
57
+ attention_backend ............................... AttnBackend.auto
58
+ attention_dropout ............................... 0.0
59
+ attention_softmax_in_fp32 ....................... False
60
+ auto_detect_ckpt_format ......................... False
61
+ barrier_with_L1_time ............................ True
62
+ bert_binary_head ................................ True
63
+ bert_embedder_type .............................. megatron
64
+ bert_load ....................................... None
65
+ bf16 ............................................ True
66
+ bias_dropout_fusion ............................. True
67
+ bias_gelu_fusion ................................ False
68
+ bias_swiglu_fusion .............................. True
69
+ biencoder_projection_dim ........................ 0
70
+ biencoder_shared_query_context_model ............ False
71
+ block_data_path ................................. None
72
+ calc_ft_timeouts ................................ False
73
+ calculate_per_token_loss ........................ False
74
+ check_for_large_grads ........................... False
75
+ check_for_nan_in_loss_and_grad .................. False
76
+ check_for_spiky_loss ............................ False
77
+ check_weight_hash_across_dp_replicas_interval ... None
78
+ ckpt_assume_constant_structure .................. False
79
+ ckpt_convert_format ............................. None
80
+ ckpt_convert_save ............................... None
81
+ ckpt_convert_update_legacy_dist_opt_format ...... False
82
+ ckpt_format ..................................... torch_dist
83
+ ckpt_fully_parallel_load ........................ False
84
+ ckpt_fully_parallel_save ........................ True
85
+ ckpt_fully_parallel_save_deprecated ............. False
86
+ ckpt_step ....................................... None
87
+ classes_fraction ................................ 1.0
88
+ clip_grad ....................................... 1.0
89
+ clone_scatter_output_in_embedding ............... True
90
+ config_logger_dir ...............................
91
+ consumed_train_samples .......................... 0
92
+ consumed_valid_samples .......................... 0
93
+ context_parallel_size ........................... 1
94
+ cp_comm_type .................................... ['p2p']
95
+ create_attention_mask_in_dataloader ............. True
96
+ cross_entropy_fusion_impl ....................... native
97
+ cross_entropy_loss_fusion ....................... False
98
+ cuda_graph_scope ................................ full_iteration
99
+ cuda_graph_warmup_steps ......................... 3
100
+ data_args_path .................................. None
101
+ data_cache_path ................................. None
102
+ data_parallel_random_init ....................... False
103
+ data_parallel_sharding_strategy ................. no_shard
104
+ data_parallel_size .............................. 8
105
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
106
+ data_per_class_fraction ......................... 1.0
107
+ data_sharding ................................... True
108
+ dataloader_type ................................. single
109
+ ddp_average_in_collective ....................... False
110
+ ddp_bucket_size ................................. None
111
+ ddp_num_buckets ................................. None
112
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
113
+ decoder_first_pipeline_num_layers ............... None
114
+ decoder_last_pipeline_num_layers ................ None
115
+ decoder_num_layers .............................. None
116
+ decoder_seq_length .............................. None
117
+ decoupled_lr .................................... None
118
+ decoupled_min_lr ................................ None
119
+ decrease_batch_size_if_needed ................... False
120
+ defer_embedding_wgrad_compute ................... False
121
+ delay_wgrad_compute ............................. False
122
+ deprecated_use_mcore_models ..................... True
123
+ deterministic_mode .............................. False
124
+ dino_bottleneck_size ............................ 256
125
+ dino_freeze_last_layer .......................... 1
126
+ dino_head_hidden_size ........................... 2048
127
+ dino_local_crops_number ......................... 10
128
+ dino_local_img_size ............................. 96
129
+ dino_norm_last_layer ............................ False
130
+ dino_teacher_temp ............................... 0.07
131
+ dino_warmup_teacher_temp ........................ 0.04
132
+ dino_warmup_teacher_temp_epochs ................. 30
133
+ disable_bf16_reduced_precision_matmul ........... False
134
+ disable_mamba_mem_eff_path ...................... False
135
+ disable_straggler_on_startup .................... False
136
+ dist_ckpt_format_deprecated ..................... None
137
+ dist_ckpt_strictness ............................ assume_ok_unexpected
138
+ distribute_saved_activations .................... False
139
+ distributed_backend ............................. nccl
140
+ distributed_timeout_minutes ..................... 10
141
+ embedding_init_method_std ....................... None
142
+ embedding_path .................................. None
143
+ empty_unused_memory_level ....................... 0
144
+ enable_cuda_graph ............................... True
145
+ enable_experimental ............................. False
146
+ enable_ft_package ............................... False
147
+ enable_gloo_process_groups ...................... True
148
+ enable_msc ...................................... True
149
+ enable_one_logger ............................... True
150
+ encoder_num_layers .............................. 16
151
+ encoder_seq_length .............................. 4096
152
+ end_weight_decay ................................ 0.1
153
+ eod_mask_loss ................................... False
154
+ error_injection_rate ............................ 0
155
+ error_injection_type ............................ transient_error
156
+ eval_interval ................................... 1000000
157
+ eval_iters ...................................... 0
158
+ evidence_data_path .............................. None
159
+ exit_duration_in_mins ........................... None
160
+ exit_interval ................................... None
161
+ exit_on_missing_checkpoint ...................... False
162
+ exit_signal_handler ............................. False
163
+ exp_avg_dtype ................................... torch.float32
164
+ exp_avg_sq_dtype ................................ torch.float32
165
+ expert_model_parallel_size ...................... 8
166
+ expert_tensor_parallel_size ..................... 1
167
+ export_force_local_attention .................... False
168
+ export_kd_cfg ................................... None
169
+ export_kd_teacher_ckpt_format ................... None
170
+ export_kd_teacher_load .......................... None
171
+ export_kv_cache_quant ........................... False
172
+ export_legacy_megatron .......................... False
173
+ export_model_type ............................... GPTModel
174
+ export_moe_apply_probs_on_input ................. False
175
+ export_qk_l2_norm ............................... False
176
+ export_quant_cfg ................................ None
177
+ export_real_quant_cfg ........................... None
178
+ export_te_mcore_model ........................... False
179
+ external_cuda_graph ............................. False
180
+ ffn_hidden_size ................................. 14336
181
+ finetune ........................................ False
182
+ finetune_data_split ............................. train
183
+ finetune_hf_dataset ............................. None
184
+ first_last_layers_bf16 .......................... False
185
+ flash_decode .................................... False
186
+ fp16 ............................................ False
187
+ fp16_lm_cross_entropy ........................... False
188
+ fp32_residual_connection ........................ False
189
+ fp8 ............................................. e4m3
190
+ fp8_amax_compute_algo ........................... most_recent
191
+ fp8_amax_history_len ............................ 1
192
+ fp8_interval .................................... 1
193
+ fp8_margin ...................................... 0
194
+ fp8_param_gather ................................ False
195
+ fp8_recipe ...................................... mxfp8
196
+ fp8_wgrad ....................................... True
197
+ fsdp_double_buffer .............................. False
198
+ global_batch_size ............................... 16
199
+ grad_reduce_in_bf16 ............................. False
200
+ gradient_accumulation_fusion .................... True
201
+ gradient_reduce_div_fusion ...................... True
202
+ group_query_attention ........................... True
203
+ head_lr_mult .................................... 1.0
204
+ heterogeneous_layers_config_encoded_json ........ None
205
+ heterogeneous_layers_config_path ................ None
206
+ hidden_dropout .................................. 0.0
207
+ hidden_size ..................................... 4096
208
+ hierarchical_context_parallel_sizes ............. None
209
+ high_priority_stream_groups ..................... []
210
+ hybrid_attention_ratio .......................... 0.0
211
+ hybrid_mlp_ratio ................................ 0.0
212
+ hybrid_override_pattern ......................... None
213
+ hysteresis ...................................... 2
214
+ ict_head_size ................................... None
215
+ ict_load ........................................ None
216
+ img_h ........................................... 224
217
+ img_w ........................................... 224
218
+ indexer_batch_size .............................. 128
219
+ indexer_log_interval ............................ 1000
220
+ inference_batch_times_seqlen_threshold .......... -1
221
+ inference_dynamic_batching ...................... False
222
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
223
+ inference_dynamic_batching_buffer_overflow_factor None
224
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
225
+ inference_dynamic_batching_chunk_size ........... 256
226
+ inference_dynamic_batching_max_requests_override None
227
+ inference_dynamic_batching_max_tokens_override .. None
228
+ inference_dynamic_batching_num_cuda_graphs ...... 16
229
+ inference_max_batch_size ........................ 8
230
+ inference_max_seq_length ........................ 2560
231
+ inference_rng_tracker ........................... False
232
+ init_method_std ................................. 0.01
233
+ init_method_xavier_uniform ...................... False
234
+ init_model_with_meta_device ..................... False
235
+ initial_loss_scale .............................. 4294967296
236
+ inprocess_active_world_size ..................... 8
237
+ inprocess_barrier_timeout ....................... 120
238
+ inprocess_completion_timeout .................... 120
239
+ inprocess_empty_cuda_cache ...................... False
240
+ inprocess_granularity ........................... node
241
+ inprocess_hard_timeout .......................... 90
242
+ inprocess_heartbeat_interval .................... 30
243
+ inprocess_heartbeat_timeout ..................... 60
244
+ inprocess_last_call_wait ........................ 1
245
+ inprocess_max_iterations ........................ None
246
+ inprocess_monitor_process_interval .............. 1.0
247
+ inprocess_monitor_thread_interval ............... 1.0
248
+ inprocess_progress_watchdog_interval ............ 1.0
249
+ inprocess_restart ............................... False
250
+ inprocess_soft_timeout .......................... 60
251
+ inprocess_termination_grace_time ................ 1
252
+ is_hybrid_model ................................. False
253
+ iter_per_epoch .................................. 1250
254
+ iterations_to_skip .............................. []
255
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
256
+ kitchen_config_file ............................. None
257
+ kitchen_recipe_number ........................... None
258
+ kv_channels ..................................... 128
259
+ kv_lora_rank .................................... 32
260
+ lazy_mpu_init ................................... None
261
+ load ............................................ None
262
+ load_main_params_from_ckpt ...................... None
263
+ load_model_opt_format ........................... False
264
+ local_rank ...................................... 0
265
+ log_energy ...................................... False
266
+ log_interval .................................... 1
267
+ log_loss_scale_to_tensorboard ................... True
268
+ log_memory_to_tensorboard ....................... False
269
+ log_num_zeros_in_grad ........................... False
270
+ log_params_norm ................................. False
271
+ log_progress .................................... False
272
+ log_straggler ................................... False
273
+ log_throughput .................................. True
274
+ log_timers_to_tensorboard ....................... False
275
+ log_validation_ppl_to_tensorboard ............... False
276
+ log_world_size_to_tensorboard ................... False
277
+ logging_level ................................... None
278
+ loss_scale ...................................... None
279
+ loss_scale_window ............................... 1000
280
+ lr .............................................. 0.0001
281
+ lr_decay_iters .................................. 8
282
+ lr_decay_samples ................................ None
283
+ lr_decay_style .................................. cosine
284
+ lr_warmup_fraction .............................. None
285
+ lr_warmup_init .................................. 0.0
286
+ lr_warmup_iters ................................. 5
287
+ lr_warmup_samples ............................... 0
288
+ lr_wsd_decay_iters .............................. None
289
+ lr_wsd_decay_samples ............................ None
290
+ lr_wsd_decay_style .............................. exponential
291
+ main_grads_dtype ................................ torch.float32
292
+ main_params_dtype ............................... torch.float32
293
+ make_vocab_size_divisible_by .................... 128
294
+ mamba_head_dim .................................. 64
295
+ mamba_num_groups ................................ 8
296
+ mamba_num_heads ................................. None
297
+ mamba_state_dim ................................. 128
298
+ manual_gc ....................................... False
299
+ manual_gc_eval .................................. True
300
+ manual_gc_interval .............................. 0
301
+ mask_factor ..................................... 1.0
302
+ mask_prob ....................................... 0.15
303
+ mask_type ....................................... random
304
+ masked_softmax_fusion ........................... False
305
+ max_position_embeddings ......................... 32768
306
+ max_tokens_to_oom ............................... 12000
307
+ memory_snapshot_path ............................ snapshot.pickle
308
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
309
+ micro_batch_size ................................ 1
310
+ microbatch_group_size_per_vp_stage .............. None
311
+ mid_level_dataset_surplus ....................... 0.005
312
+ min_loss_scale .................................. 1.0
313
+ min_lr .......................................... 1e-05
314
+ mlp_chunks_for_prefill .......................... 1
315
+ mmap_bin_files .................................. True
316
+ mock_data ....................................... False
317
+ moe_apply_probs_on_input ........................ False
318
+ moe_aux_loss_coeff .............................. 0.01
319
+ moe_deepep_num_sms .............................. 20
320
+ moe_echo_enable_random_offloading ............... False
321
+ moe_echo_expert_dispatch_overlap ................ True
322
+ moe_echo_expert_dispatcher_type ................. hybridep
323
+ moe_echo_recompute_expert_dispatch .............. False
324
+ moe_enable_deepep ............................... False
325
+ moe_enable_echo ................................. True
326
+ moe_enable_hybridep ............................. True
327
+ moe_expert_capacity_factor ...................... None
328
+ moe_extended_tp ................................. False
329
+ moe_ffn_hidden_size ............................. 3584
330
+ moe_grouped_gemm ................................ True
331
+ moe_hybridep_num_sms ............................ 16
332
+ moe_input_jitter_eps ............................ None
333
+ moe_layer_freq .................................. 1
334
+ moe_layer_recompute ............................. False
335
+ moe_num_echo_experts ............................ 8
336
+ moe_pad_expert_input_to_capacity ................ False
337
+ moe_per_layer_logging ........................... False
338
+ moe_permute_fusion .............................. False
339
+ moe_received_token_capacity ..................... 1.08203125
340
+ moe_router_bias_update_rate ..................... 0.001
341
+ moe_router_dtype ................................ fp32
342
+ moe_router_enable_expert_bias ................... False
343
+ moe_router_force_load_balancing ................. False
344
+ moe_router_group_topk ........................... None
345
+ moe_router_load_balancing_type .................. aux_loss
346
+ moe_router_num_groups ........................... None
347
+ moe_router_padding_for_fp8 ...................... False
348
+ moe_router_pre_softmax .......................... False
349
+ moe_router_score_function ....................... softmax
350
+ moe_router_topk ................................. 8
351
+ moe_router_topk_scaling_factor .................. None
352
+ moe_shared_expert_intermediate_size ............. None
353
+ moe_shared_expert_overlap ....................... False
354
+ moe_token_dispatcher_type ....................... flex
355
+ moe_token_drop_policy ........................... probs
356
+ moe_upcycling_granularity ....................... 1
357
+ moe_use_device_initiated_grouped_gemm ........... True
358
+ moe_use_legacy_grouped_gemm ..................... False
359
+ moe_use_upcycling ............................... False
360
+ moe_z_loss_coeff ................................ None
361
+ mrope_section ................................... None
362
+ mscale .......................................... 1.0
363
+ mscale_all_dim .................................. 0.0
364
+ mtp_loss_scaling_factor ......................... 0.1
365
+ mtp_num_layers .................................. None
366
+ multi_latent_attention .......................... False
367
+ nccl_all_reduce_for_prefill ..................... False
368
+ nccl_communicator_config_path ................... None
369
+ nccl_ub ......................................... False
370
+ no_load_optim ................................... None
371
+ no_load_rng ..................................... None
372
+ no_persist_layer_norm ........................... False
373
+ no_rope_freq .................................... None
374
+ no_save_optim ................................... None
375
+ no_save_rng ..................................... None
376
+ non_persistent_ckpt_type ........................ None
377
+ non_persistent_global_ckpt_dir .................. None
378
+ non_persistent_local_ckpt_algo .................. fully_parallel
379
+ non_persistent_local_ckpt_dir ................... None
380
+ non_persistent_save_interval .................... None
381
+ norm_epsilon .................................... 1e-05
382
+ normalization ................................... RMSNorm
383
+ num_attention_heads ............................. 32
384
+ num_channels .................................... 3
385
+ num_classes ..................................... 1000
386
+ num_dataset_builder_threads ..................... 1
387
+ num_distributed_optimizer_instances ............. 1
388
+ num_experts ..................................... 32
389
+ num_layers ...................................... 16
390
+ num_layers_at_end_in_bf16 ....................... 1
391
+ num_layers_at_start_in_bf16 ..................... 1
392
+ num_layers_per_virtual_pipeline_stage ........... None
393
+ num_query_groups ................................ 8
394
+ num_virtual_stages_per_pipeline_rank ............ None
395
+ num_workers ..................................... 2
396
+ object_storage_cache_path ....................... None
397
+ one_logger_async ................................ False
398
+ one_logger_project .............................. megatron-lm
399
+ one_logger_run_name ............................. None
400
+ onnx_safe ....................................... None
401
+ openai_gelu ..................................... False
402
+ optimizer ....................................... adam
403
+ optimizer_cpu_offload ........................... False
404
+ optimizer_offload_fraction ...................... 1.0
405
+ output_bert_embeddings .......................... False
406
+ overlap_cpu_optimizer_d2h_h2d ................... False
407
+ overlap_grad_reduce ............................. True
408
+ overlap_moe_expert_parallel_comm ................ False
409
+ overlap_p2p_comm ................................ False
410
+ overlap_p2p_comm_warmup_flush ................... False
411
+ overlap_param_gather ............................ True
412
+ overlap_param_gather_with_optimizer_step ........ False
413
+ override_opt_param_scheduler .................... False
414
+ params_dtype .................................... torch.bfloat16
415
+ patch_dim ....................................... 16
416
+ per_split_data_args_path ........................ None
417
+ perform_initialization .......................... True
418
+ pin_cpu_grads ................................... True
419
+ pin_cpu_params .................................. True
420
+ pipeline_model_parallel_comm_backend ............ None
421
+ pipeline_model_parallel_layout .................. None
422
+ pipeline_model_parallel_size .................... 1
423
+ position_embedding_type ......................... rope
424
+ pretrained_checkpoint ........................... None
425
+ profile ......................................... False
426
+ profile_ranks ................................... [0]
427
+ profile_step_end ................................ 12
428
+ profile_step_start .............................. 10
429
+ q_lora_rank ..................................... None
430
+ qk_head_dim ..................................... 128
431
+ qk_l2_norm ...................................... False
432
+ qk_layernorm .................................... False
433
+ qk_pos_emb_head_dim ............................. 64
434
+ query_in_block_prob ............................. 0.1
435
+ rampup_batch_size ............................... None
436
+ rank ............................................ 0
437
+ recompute_granularity ........................... None
438
+ recompute_method ................................ None
439
+ recompute_modules ............................... None
440
+ recompute_num_layers ............................ None
441
+ record_memory_history ........................... False
442
+ relative_attention_max_distance ................. 128
443
+ relative_attention_num_buckets .................. 32
444
+ replication ..................................... False
445
+ replication_factor .............................. 2
446
+ replication_jump ................................ None
447
+ rerun_mode ...................................... validate_results
448
+ reset_attention_mask ............................ False
449
+ reset_position_ids .............................. False
450
+ result_rejected_tracker_filename ................ None
451
+ retriever_report_topk_accuracies ................ []
452
+ retriever_score_scaling ......................... False
453
+ retriever_seq_length ............................ 256
454
+ retro_add_retriever ............................. False
455
+ retro_attention_gate ............................ 1
456
+ retro_cyclic_train_iters ........................ None
457
+ retro_encoder_attention_dropout ................. 0.1
458
+ retro_encoder_hidden_dropout .................... 0.1
459
+ retro_encoder_layers ............................ 2
460
+ retro_num_neighbors ............................. 2
461
+ retro_num_retrieved_chunks ...................... 2
462
+ retro_project_dir ............................... None
463
+ retro_verify_neighbor_count ..................... True
464
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
465
+ rope_scaling_factor ............................. 8.0
466
+ rope_type ....................................... None
467
+ rotary_base ..................................... 1000000
468
+ rotary_interleaved .............................. False
469
+ rotary_percent .................................. 1.0
470
+ rotary_scaling_factor ........................... 1.0
471
+ rotary_seq_len_interpolation_factor ............. None
472
+ run_workload_inspector_server ................... False
473
+ sample_rate ..................................... 1.0
474
+ save ............................................ None
475
+ save_interval ................................... None
476
+ save_retain_interval ............................ None
477
+ scatter_gather_tensors_in_pipeline .............. True
478
+ seed ............................................ 1234
479
+ seq_length ...................................... 4096
480
+ sequence_parallel ............................... False
481
+ sft ............................................. False
482
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
483
+ sgd_momentum .................................... 0.9
484
+ short_seq_prob .................................. 0.1
485
+ skip_train ...................................... False
486
+ skipped_train_samples ........................... 0
487
+ spec ............................................ None
488
+ split ........................................... 100,0,0
489
+ squared_relu .................................... False
490
+ start_weight_decay .............................. 0.1
491
+ straggler_ctrlr_port ............................ 65535
492
+ straggler_minmax_count .......................... 1
493
+ suggested_communication_unit_size ............... None
494
+ swiglu .......................................... True
495
+ swin_backbone_type .............................. tiny
496
+ symmetric_ar_type ............................... None
497
+ te_rng_tracker .................................. True
498
+ tensor_model_parallel_size ...................... 1
499
+ tensorboard_dir ................................. None
500
+ tensorboard_log_interval ........................ 1
501
+ tensorboard_queue_size .......................... 1000
502
+ test_data_path .................................. None
503
+ test_mode ....................................... False
504
+ tiktoken_num_special_tokens ..................... 1000
505
+ tiktoken_pattern ................................ None
506
+ tiktoken_special_tokens ......................... None
507
+ timing_log_level ................................ 0
508
+ timing_log_option ............................... minmax
509
+ titles_data_path ................................ None
510
+ tokenizer_model ................................. None
511
+ tokenizer_type .................................. GPT2BPETokenizer
512
+ torch_fsdp2_reshard_after_forward ............... True
513
+ tp_comm_bootstrap_backend ....................... nccl
514
+ tp_comm_bulk_dgrad .............................. True
515
+ tp_comm_bulk_wgrad .............................. True
516
+ tp_comm_overlap ................................. False
517
+ tp_comm_overlap_ag .............................. True
518
+ tp_comm_overlap_cfg ............................. None
519
+ tp_comm_overlap_rs .............................. True
520
+ tp_comm_overlap_rs_dgrad ........................ False
521
+ tp_comm_split_ag ................................ True
522
+ tp_comm_split_rs ................................ True
523
+ train_data_path ................................. None
524
+ train_iters ..................................... 12
525
+ train_samples ................................... None
526
+ train_sync_interval ............................. None
527
+ transformer_impl ................................ transformer_engine
528
+ transformer_pipeline_model_parallel_size ........ 1
529
+ untie_embeddings_and_output_weights ............. True
530
+ use_checkpoint_args ............................. False
531
+ use_checkpoint_opt_param_scheduler .............. False
532
+ use_cpu_initialization .......................... None
533
+ use_custom_fsdp ................................. False
534
+ use_dist_ckpt ................................... True
535
+ use_dist_ckpt_deprecated ........................ False
536
+ use_distributed_optimizer ....................... True
537
+ use_flash_attn .................................. False
538
+ use_legacy_models ............................... False
539
+ use_mp_args_from_checkpoint_args ................ False
540
+ use_one_sent_docs ............................... False
541
+ use_persistent_ckpt_worker ...................... False
542
+ use_precision_aware_optimizer ................... False
543
+ use_pytorch_profiler ............................ False
544
+ use_ring_exchange_p2p ........................... False
545
+ use_rope_scaling ................................ False
546
+ use_rotary_position_embeddings .................. False
547
+ use_sharp ....................................... False
548
+ use_tokenizer_model_from_checkpoint_args ........ True
549
+ use_torch_fsdp2 ................................. False
550
+ use_torch_optimizer_for_cpu_offload ............. False
551
+ use_tp_pp_dp_mapping ............................ False
552
+ v_head_dim ...................................... 128
553
+ valid_data_path ................................. None
554
+ variable_seq_lengths ............................ False
555
+ virtual_pipeline_model_parallel_size ............ None
556
+ vision_backbone_type ............................ vit
557
+ vision_pretraining .............................. False
558
+ vision_pretraining_type ......................... classify
559
+ vocab_extra_ids ................................. 0
560
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
561
+ vocab_size ...................................... None
562
+ wandb_exp_name ..................................
563
+ wandb_project ...................................
564
+ wandb_save_dir ..................................
565
+ weight_decay .................................... 0.1
566
+ weight_decay_incr_style ......................... constant
567
+ wgrad_deferral_limit ............................ 0
568
+ world_size ...................................... 8
569
+ yaml_cfg ........................................ None
570
+ -------------------- end of arguments ---------------------
571
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
572
+ > building GPT2BPETokenizer tokenizer ...
573
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
574
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
575
+ > initializing torch distributed ...
576
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
577
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
578
+ [Gloo] Rank [Gloo] Rank 3 is connected to 17 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
579
+ 7
580
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank
582
+ [Gloo] Rank 65 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 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank 12365 is connected to is connected to is connected to is connected to [Gloo] Rank is connected to 777747 peer ranks. peer ranks. peer ranks. peer ranks. is connected to peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 7777 peer ranks. [Gloo] Rank 7
586
+
587
+
588
+
589
+ Expected number of connected peer ranks is : 7
590
+ 7 is connected to
591
+ 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
593
+ > initialized tensor model parallel with size 1
594
+ > initialized pipeline model parallel with size 1
595
+ > setting random seeds to 1234 ...
596
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
597
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
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
+ [rank6]:[W830 06:02:43.340598712 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.
604
+ [rank3]:[W830 06:02:43.342978654 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.
605
+ [rank1]:[W830 06:02:43.343887894 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.
606
+ [rank2]:[W830 06:02:43.345573975 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.
607
+ [rank4]:[W830 06:02:43.345590399 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.
608
+ [rank7]:[W830 06:02:43.345739375 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.
609
+ [rank5]:[W830 06:02:43.348933651 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ > compiling dataset index builder ...
611
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
612
+ make: Nothing to be done for 'default'.
613
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
614
+ >>> done with dataset index builder. Compilation time: 0.034 seconds
615
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
616
+ > compiling and loading fused kernels ...
617
+ [rank0]:[W830 06:02:43.503803142 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.
618
+ >>> done with compiling and loading fused kernels. Compilation time: 3.792 seconds
619
+ time to initialize megatron (seconds): 7.412
620
+ [after megatron is initialized] datetime: 2026-08-30 06:02:49
621
+ building GPT model ...
622
+ /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.
623
+ warnings.warn(
624
+ /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.
625
+ warnings.warn(
626
+ /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.
627
+ warnings.warn(
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3903983616
639
+ 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)
640
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
641
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
642
+ module.output_layer.weight
643
+ Params for bucket 2 (42086400 elements, 42086400 padded size):
644
+ module.decoder.final_layernorm.weight
645
+ module.decoder.layers.15.self_attention.linear_qkv.weight
646
+ module.decoder.layers.15.pre_mlp_layernorm.weight
647
+ module.decoder.layers.15.mlp.router.weight
648
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
649
+ module.decoder.layers.15.self_attention.linear_proj.weight
650
+ Params for bucket 3 (42082304 elements, 42082304 padded size):
651
+ module.decoder.layers.14.mlp.router.weight
652
+ module.decoder.layers.14.self_attention.linear_qkv.weight
653
+ module.decoder.layers.14.pre_mlp_layernorm.weight
654
+ module.decoder.layers.14.self_attention.linear_proj.weight
655
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
656
+ Params for bucket 4 (42082304 elements, 42082304 padded size):
657
+ module.decoder.layers.13.self_attention.linear_qkv.weight
658
+ module.decoder.layers.13.pre_mlp_layernorm.weight
659
+ module.decoder.layers.13.mlp.router.weight
660
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
661
+ module.decoder.layers.13.self_attention.linear_proj.weight
662
+ Params for bucket 5 (42082304 elements, 42082304 padded size):
663
+ module.decoder.layers.12.mlp.router.weight
664
+ module.decoder.layers.12.self_attention.linear_proj.weight
665
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
666
+ module.decoder.layers.12.pre_mlp_layernorm.weight
667
+ module.decoder.layers.12.self_attention.linear_qkv.weight
668
+ Params for bucket 6 (42082304 elements, 42082304 padded size):
669
+ module.decoder.layers.11.mlp.router.weight
670
+ module.decoder.layers.11.self_attention.linear_proj.weight
671
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
672
+ module.decoder.layers.11.self_attention.linear_qkv.weight
673
+ module.decoder.layers.11.pre_mlp_layernorm.weight
674
+ Params for bucket 7 (42082304 elements, 42082304 padded size):
675
+ module.decoder.layers.10.self_attention.linear_proj.weight
676
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
677
+ module.decoder.layers.10.mlp.router.weight
678
+ module.decoder.layers.10.self_attention.linear_qkv.weight
679
+ module.decoder.layers.10.pre_mlp_layernorm.weight
680
+ Params for bucket 8 (42082304 elements, 42082304 padded size):
681
+ module.decoder.layers.9.self_attention.linear_qkv.weight
682
+ module.decoder.layers.9.pre_mlp_layernorm.weight
683
+ module.decoder.layers.9.mlp.router.weight
684
+ module.decoder.layers.9.self_attention.linear_proj.weight
685
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
686
+ Params for bucket 9 (42082304 elements, 42082304 padded size):
687
+ module.decoder.layers.8.pre_mlp_layernorm.weight
688
+ module.decoder.layers.8.self_attention.linear_qkv.weight
689
+ module.decoder.layers.8.mlp.router.weight
690
+ module.decoder.layers.8.self_attention.linear_proj.weight
691
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
692
+ Params for bucket 10 (42082304 elements, 42082304 padded size):
693
+ module.decoder.layers.7.mlp.router.weight
694
+ module.decoder.layers.7.pre_mlp_layernorm.weight
695
+ module.decoder.layers.7.self_attention.linear_qkv.weight
696
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
697
+ module.decoder.layers.7.self_attention.linear_proj.weight
698
+ Params for bucket 11 (42082304 elements, 42082304 padded size):
699
+ module.decoder.layers.6.mlp.router.weight
700
+ module.decoder.layers.6.self_attention.linear_proj.weight
701
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
702
+ module.decoder.layers.6.pre_mlp_layernorm.weight
703
+ module.decoder.layers.6.self_attention.linear_qkv.weight
704
+ Params for bucket 12 (42082304 elements, 42082304 padded size):
705
+ module.decoder.layers.5.mlp.router.weight
706
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
707
+ module.decoder.layers.5.self_attention.linear_qkv.weight
708
+ module.decoder.layers.5.self_attention.linear_proj.weight
709
+ module.decoder.layers.5.pre_mlp_layernorm.weight
710
+ Params for bucket 13 (42082304 elements, 42082304 padded size):
711
+ module.decoder.layers.4.mlp.router.weight
712
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
713
+ module.decoder.layers.4.self_attention.linear_proj.weight
714
+ module.decoder.layers.4.pre_mlp_layernorm.weight
715
+ module.decoder.layers.4.self_attention.linear_qkv.weight
716
+ Params for bucket 14 (42082304 elements, 42082304 padded size):
717
+ module.decoder.layers.3.mlp.router.weight
718
+ module.decoder.layers.3.self_attention.linear_proj.weight
719
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
720
+ module.decoder.layers.3.pre_mlp_layernorm.weight
721
+ module.decoder.layers.3.self_attention.linear_qkv.weight
722
+ Params for bucket 15 (42082304 elements, 42082304 padded size):
723
+ module.decoder.layers.2.self_attention.linear_qkv.weight
724
+ module.decoder.layers.2.pre_mlp_layernorm.weight
725
+ module.decoder.layers.2.mlp.router.weight
726
+ module.decoder.layers.2.self_attention.linear_proj.weight
727
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
728
+ Params for bucket 16 (42082304 elements, 42082304 padded size):
729
+ module.decoder.layers.1.mlp.router.weight
730
+ module.decoder.layers.1.self_attention.linear_proj.weight
731
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
732
+ module.decoder.layers.1.pre_mlp_layernorm.weight
733
+ module.decoder.layers.1.self_attention.linear_qkv.weight
734
+ Params for bucket 17 (42082304 elements, 42082304 padded size):
735
+ module.decoder.layers.0.pre_mlp_layernorm.weight
736
+ module.decoder.layers.0.self_attention.linear_proj.weight
737
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
738
+ module.decoder.layers.0.mlp.router.weight
739
+ module.decoder.layers.0.self_attention.linear_qkv.weight
740
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
741
+ module.embedding.word_embeddings.weight
742
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56
743
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
744
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
745
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
746
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
747
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
748
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
750
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
751
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
752
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
753
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
754
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
755
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
756
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
757
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
758
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
759
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
760
+ Params for bucket 6 (58720256 elements, 58720256 padded size):
761
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
762
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
763
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
764
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
765
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
766
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
767
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
768
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
769
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
770
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
771
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
772
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
773
+ Params for bucket 10 (58720256 elements, 58720256 padded size):
774
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
775
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
776
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
777
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
778
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
779
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
780
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
781
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
782
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
783
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
784
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
785
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
786
+ Params for bucket 14 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
788
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
790
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
791
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
792
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
793
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
794
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
795
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
796
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
797
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
798
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
799
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
800
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
801
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
802
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
803
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
804
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
805
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
806
+ Params for bucket 20 (58720256 elements, 58720256 padded size):
807
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
808
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
809
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
811
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
812
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
813
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
814
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
815
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
816
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
817
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
818
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
819
+ Params for bucket 24 (58720256 elements, 58720256 padded size):
820
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
821
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
822
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
823
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
824
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
825
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
827
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
828
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
829
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
830
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
831
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
832
+ Params for bucket 28 (58720256 elements, 58720256 padded size):
833
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
834
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
835
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
836
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
837
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
838
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
839
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
840
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
841
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
842
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
843
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
844
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
845
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
846
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
847
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
848
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
849
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
850
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
851
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
852
+ Params for bucket 34 (58720256 elements, 58720256 padded size):
853
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
854
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
855
+ Params for bucket 35 (58720256 elements, 58720256 padded size):
856
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
857
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
858
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
859
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
860
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
861
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
862
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
863
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
864
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
865
+ Params for bucket 38 (58720256 elements, 58720256 padded size):
866
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
867
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
868
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
869
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
870
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
871
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
872
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
873
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
874
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
875
+ Params for bucket 41 (58720256 elements, 58720256 padded size):
876
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
877
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
878
+ Params for bucket 42 (58720256 elements, 58720256 padded size):
879
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
880
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
881
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
882
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
883
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
884
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
885
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
886
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
887
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
888
+ Params for bucket 45 (58720256 elements, 58720256 padded size):
889
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
890
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
891
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
892
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
893
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
894
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
895
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
896
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
897
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
898
+ Params for bucket 48 (58720256 elements, 58720256 padded size):
899
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
900
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
901
+ Params for bucket 49 (58720256 elements, 58720256 padded size):
902
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
903
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
904
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
905
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
906
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
907
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
908
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
909
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
910
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
911
+ Params for bucket 52 (58720256 elements, 58720256 padded size):
912
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
913
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
914
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
915
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
916
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
917
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
918
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
919
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
920
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
921
+ Params for bucket 55 (58720256 elements, 58720256 padded size):
922
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
923
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
924
+ Params for bucket 56 (58720256 elements, 58720256 padded size):
925
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
926
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
927
+ 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 0x7b4d913ec6b0>, config_logger_dir='')
928
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
929
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 06:02:49
930
+ > building train, validation, and test datasets ...
931
+ > datasets target sizes (minimum size):
932
+ train: 192
933
+ validation: 0
934
+ test: 0
935
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
936
+ > building train, validation, and test datasets for GPT ...
937
+ 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 0x7b4d91972330>, 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)
938
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
939
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
940
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
941
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
942
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
943
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
944
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
945
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
946
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
947
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
948
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
949
+ > finished creating GPT datasets ...
950
+ [after dataloaders are built] datetime: 2026-08-30 06:02:49
951
+ done with setup ...
952
+ (min, max) time across ranks (ms):
953
+ model-and-optimizer-setup ......................: (326.59, 334.17)
954
+ train/valid/test-data-iterators-setup ..........: (64.07, 74.15)
955
+ training ...
956
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
957
+ [before the start of training step] datetime: 2026-08-30 06:02:49
958
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
959
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
960
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
961
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
962
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
963
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
964
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
965
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
966
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
967
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
968
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
969
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
970
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
971
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
972
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
973
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
974
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
975
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
976
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
977
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
978
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
979
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
980
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
981
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
982
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
983
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
985
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
989
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
991
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
992
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
993
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
994
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
995
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
996
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
997
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
998
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
999
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1000
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1001
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1002
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1003
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1004
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1005
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1006
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1007
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1008
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1009
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1010
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1011
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1012
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1013
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1014
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1015
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1016
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1017
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1018
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1019
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1020
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1021
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1022
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1023
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1024
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1025
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1026
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1027
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1028
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1029
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1030
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1031
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1032
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1033
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1034
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1035
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1036
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1037
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1038
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1039
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1040
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1041
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1042
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1043
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1044
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1045
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1046
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1047
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1048
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1049
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1050
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1051
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1052
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1053
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1054
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1055
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1056
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1057
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1058
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1059
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1060
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1061
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1062
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1063
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1064
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1065
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1066
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1067
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1068
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1069
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1070
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1071
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1072
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1073
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1074
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1075
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1076
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1077
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1078
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1079
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1080
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1081
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1082
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1083
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1084
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1085
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1086
+ [2026-08-30 06:03:28] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38699.5 | throughput per GPU (TFLOP/s/GPU): 8.6 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103012E+01 | loss scale: 1.0 | grad norm: 44.246 | number of skipped iterations: 0 | number of nan iterations: 0 |Number of parameters in transformer block in billions: 23.22
1087
+
1088
+ Number of parameters in embedding layers in billions: 0.41
1089
+ Total number of parameters in billions: 23.63
1090
+ Number of parameters in most loaded shard in billions: 23.6320
1091
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1092
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 59522.912109375 | max allocated: 68963.6640625 | reserved: 72944.0 | max reserved: 72944.0
1093
+ [2026-08-30 06:03:28] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 546.9 | throughput per GPU (TFLOP/s/GPU): 609.6 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101632E+01 | loss scale: 1.0 | grad norm: 45.758 | number of skipped iterations: 0 | number of nan iterations: 0 |
1094
+ [2026-08-30 06:03:29] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 485.9 | throughput per GPU (TFLOP/s/GPU): 686.1 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.022673E+01 | loss scale: 1.0 | grad norm: 192.285 | number of skipped iterations: 0 | number of nan iterations: 0 |
1095
+ Capture CUDA graph for training!!!
1096
+ CUDA graph capture done!!!
1097
+ [2026-08-30 06:03:31] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2141.2 | throughput per GPU (TFLOP/s/GPU): 155.7 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.325120E+01 | loss scale: 1.0 | grad norm: 30.083 | number of skipped iterations: 0 | number of nan iterations: 0 |
1098
+ [2026-08-30 06:03:31] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 397.5 | throughput per GPU (TFLOP/s/GPU): 838.7 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.038933E+01 | loss scale: 1.0 | grad norm: 16.542 | number of skipped iterations: 0 | number of nan iterations: 0 |
1099
+ [2026-08-30 06:03:32] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 393.1 | throughput per GPU (TFLOP/s/GPU): 848.0 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.050315E+01 | loss scale: 1.0 | grad norm: 41.427 | number of skipped iterations: 0 | number of nan iterations: 0 |
1100
+ [2026-08-30 06:03:32] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 398.4 | throughput per GPU (TFLOP/s/GPU): 836.9 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.011129E+01 | loss scale: 1.0 | grad norm: 8.126 | number of skipped iterations: 0 | number of nan iterations: 0 |
1101
+ [2026-08-30 06:03:32] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 392.9 | throughput per GPU (TFLOP/s/GPU): 848.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.572981E+00 | loss scale: 1.0 | grad norm: 5.901 | number of skipped iterations: 0 | number of nan iterations: 0 |
1102
+ [2026-08-30 06:03:33] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 393.8 | throughput per GPU (TFLOP/s/GPU): 846.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.315355E+00 | loss scale: 1.0 | grad norm: 5.763 | number of skipped iterations: 0 | number of nan iterations: 0 |
1103
+ [2026-08-30 06:03:33] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 393.1 | throughput per GPU (TFLOP/s/GPU): 848.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.132029E+00 | loss scale: 1.0 | grad norm: 5.555 | number of skipped iterations: 0 | number of nan iterations: 0 |
1104
+ [2026-08-30 06:03:34] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 392.8 | throughput per GPU (TFLOP/s/GPU): 848.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.999208E+00 | loss scale: 1.0 | grad norm: 5.249 | number of skipped iterations: 0 | number of nan iterations: 0 |
1105
+ [2026-08-30 06:03:34] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 393.3 | throughput per GPU (TFLOP/s/GPU): 847.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.880781E+00 | loss scale: 1.0 | grad norm: 5.011 | number of skipped iterations: 0 | number of nan iterations: 0 |
1106
+ [after training is done] datetime: 2026-08-30 06:03:34
1107
+ [rank0]:[W830 06:03:35.976342965 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())
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2/mem_samples.csv ADDED
@@ -0,0 +1,224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 4
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 2
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 4
42
+ 1, 629
43
+ 2, 629
44
+ 3, 629
45
+ 4, 629
46
+ 5, 629
47
+ 6, 629
48
+ 7, 629
49
+ 0, 945
50
+ 1, 967
51
+ 2, 965
52
+ 3, 961
53
+ 4, 957
54
+ 5, 959
55
+ 6, 975
56
+ 7, 967
57
+ 0, 1895
58
+ 1, 1895
59
+ 2, 1895
60
+ 3, 1895
61
+ 4, 1895
62
+ 5, 1895
63
+ 6, 1895
64
+ 7, 1895
65
+ 0, 38989
66
+ 1, 38973
67
+ 2, 38961
68
+ 3, 38991
69
+ 4, 38985
70
+ 5, 39009
71
+ 6, 39011
72
+ 7, 39001
73
+ 0, 39327
74
+ 1, 39361
75
+ 2, 39345
76
+ 3, 39351
77
+ 4, 39351
78
+ 5, 39365
79
+ 6, 39375
80
+ 7, 39377
81
+ 0, 41893
82
+ 1, 41885
83
+ 2, 41885
84
+ 3, 41885
85
+ 4, 41885
86
+ 5, 41885
87
+ 6, 41885
88
+ 7, 41885
89
+ 0, 41951
90
+ 1, 41943
91
+ 2, 41943
92
+ 3, 41943
93
+ 4, 41943
94
+ 5, 41943
95
+ 6, 41943
96
+ 7, 41943
97
+ 0, 42033
98
+ 1, 42035
99
+ 2, 42041
100
+ 3, 42035
101
+ 4, 42039
102
+ 5, 42031
103
+ 6, 42055
104
+ 7, 42045
105
+ 0, 42865
106
+ 1, 42857
107
+ 2, 42857
108
+ 3, 42857
109
+ 4, 42857
110
+ 5, 42857
111
+ 6, 42857
112
+ 7, 42857
113
+ 0, 43143
114
+ 1, 43135
115
+ 2, 43135
116
+ 3, 43135
117
+ 4, 43135
118
+ 5, 43135
119
+ 6, 43135
120
+ 7, 43135
121
+ 0, 43919
122
+ 1, 43911
123
+ 2, 43911
124
+ 3, 43911
125
+ 4, 43911
126
+ 5, 43911
127
+ 6, 43911
128
+ 7, 43911
129
+ 0, 44567
130
+ 1, 44559
131
+ 2, 44559
132
+ 3, 44559
133
+ 4, 44559
134
+ 5, 44559
135
+ 6, 44559
136
+ 7, 44559
137
+ 0, 67177
138
+ 1, 67155
139
+ 2, 67155
140
+ 3, 67155
141
+ 4, 67169
142
+ 5, 67155
143
+ 6, 67155
144
+ 7, 67155
145
+ 0, 67663
146
+ 1, 67641
147
+ 2, 67641
148
+ 3, 67641
149
+ 4, 67655
150
+ 5, 67641
151
+ 6, 67641
152
+ 7, 67641
153
+ 0, 67663
154
+ 1, 67641
155
+ 2, 67641
156
+ 3, 67641
157
+ 4, 67655
158
+ 5, 67641
159
+ 6, 67641
160
+ 7, 67641
161
+ 0, 67663
162
+ 1, 67641
163
+ 2, 67641
164
+ 3, 67641
165
+ 4, 67655
166
+ 5, 67641
167
+ 6, 67641
168
+ 7, 67641
169
+ 0, 78791
170
+ 1, 78809
171
+ 2, 78757
172
+ 3, 78777
173
+ 4, 78805
174
+ 5, 78793
175
+ 6, 78551
176
+ 7, 77895
177
+ 0, 79615
178
+ 1, 79571
179
+ 2, 79601
180
+ 3, 79583
181
+ 4, 79693
182
+ 5, 79609
183
+ 6, 79455
184
+ 7, 78739
185
+ 0, 80573
186
+ 1, 80551
187
+ 2, 80775
188
+ 3, 80551
189
+ 4, 80789
190
+ 5, 80551
191
+ 6, 80775
192
+ 7, 80439
193
+ 0, 81415
194
+ 1, 81391
195
+ 2, 81627
196
+ 3, 81399
197
+ 4, 81565
198
+ 5, 81419
199
+ 6, 81677
200
+ 7, 81331
201
+ 0, 79241
202
+ 1, 78219
203
+ 2, 88081
204
+ 3, 79357
205
+ 4, 78023
206
+ 5, 77713
207
+ 6, 78879
208
+ 7, 77365
209
+ 0, 119951
210
+ 1, 120225
211
+ 2, 120113
212
+ 3, 119891
213
+ 4, 120115
214
+ 5, 119891
215
+ 6, 120001
216
+ 7, 119875
217
+ 0, 119951
218
+ 1, 120225
219
+ 2, 120113
220
+ 3, 119891
221
+ 4, 120115
222
+ 5, 119891
223
+ 6, 120001
224
+ 7, 119875
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=echo
2
+ config=G2
3
+ family=tierA
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=bracing-regions-through-kernels
6
+ commit=9a236793
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=3584
11
+ num_experts=32
12
+ topk=8
13
+ experts_per_rank=4
14
+ expert_params=44040192
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=32768
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.08203125
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2/train.log ADDED
@@ -0,0 +1,1143 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G2 / echo (slots/rank=1 policy=one)
3
+ E=32 K=8 H'=3584 m=4
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.08203125 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 9a236793
8
+ ==============================================================
9
+ W0830 06:03:45.722000 568159 torch/distributed/run.py:766]
10
+ W0830 06:03:45.722000 568159 torch/distributed/run.py:766] *****************************************
11
+ W0830 06:03:45.722000 568159 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 06:03:45.722000 568159 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
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
20
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
21
+ Number of virtual stages per pipeline stage: None
22
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
23
+ using torch.bfloat16 for parameters ...
24
+ /workspace/Megatron-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
+ ------------------------ arguments ------------------------
27
+ account_for_embedding_in_pipeline_split ......... False
28
+ account_for_loss_in_pipeline_split .............. False
29
+ accumulate_allreduce_grads_in_fp32 .............. True
30
+ adam_beta1 ...................................... 0.9
31
+ adam_beta2 ...................................... 0.999
32
+ adam_eps ........................................ 1e-08
33
+ add_bias_linear ................................. False
34
+ add_position_embedding .......................... False
35
+ add_qkv_bias .................................... False
36
+ adlr_autoresume ................................. False
37
+ adlr_autoresume_interval ........................ 1000
38
+ align_grad_reduce ............................... True
39
+ align_param_gather .............................. False
40
+ app_tag_run_name ................................ None
41
+ app_tag_run_version ............................. 0.0.0
42
+ apply_layernorm_1p .............................. False
43
+ apply_query_key_layer_scaling ................... False
44
+ apply_residual_connection_post_layernorm ........ False
45
+ apply_rope_fusion ............................... True
46
+ async_save ...................................... None
47
+ async_tensor_model_parallel_allreduce ........... True
48
+ attention_backend ............................... AttnBackend.auto
49
+ attention_dropout ............................... 0.0
50
+ attention_softmax_in_fp32 ....................... False
51
+ auto_detect_ckpt_format ......................... False
52
+ barrier_with_L1_time ............................ True
53
+ bert_binary_head ................................ True
54
+ bert_embedder_type .............................. megatron
55
+ bert_load ....................................... None
56
+ bf16 ............................................ True
57
+ bias_dropout_fusion ............................. True
58
+ bias_gelu_fusion ................................ False
59
+ bias_swiglu_fusion .............................. True
60
+ biencoder_projection_dim ........................ 0
61
+ biencoder_shared_query_context_model ............ False
62
+ block_data_path ................................. None
63
+ calc_ft_timeouts ................................ False
64
+ calculate_per_token_loss ........................ False
65
+ check_for_large_grads ........................... False
66
+ check_for_nan_in_loss_and_grad .................. False
67
+ check_for_spiky_loss ............................ False
68
+ check_weight_hash_across_dp_replicas_interval ... None
69
+ ckpt_assume_constant_structure .................. False
70
+ ckpt_convert_format ............................. None
71
+ ckpt_convert_save ............................... None
72
+ ckpt_convert_update_legacy_dist_opt_format ...... False
73
+ ckpt_format ..................................... torch_dist
74
+ ckpt_fully_parallel_load ........................ False
75
+ ckpt_fully_parallel_save ........................ True
76
+ ckpt_fully_parallel_save_deprecated ............. False
77
+ ckpt_step ....................................... None
78
+ classes_fraction ................................ 1.0
79
+ clip_grad ....................................... 1.0
80
+ clone_scatter_output_in_embedding ............... True
81
+ config_logger_dir ...............................
82
+ consumed_train_samples .......................... 0
83
+ consumed_valid_samples .......................... 0
84
+ context_parallel_size ........................... 1
85
+ cp_comm_type .................................... ['p2p']
86
+ create_attention_mask_in_dataloader ............. True
87
+ cross_entropy_fusion_impl ....................... native
88
+ cross_entropy_loss_fusion ....................... False
89
+ cuda_graph_scope ................................ full_iteration
90
+ cuda_graph_warmup_steps ......................... 3
91
+ data_args_path .................................. None
92
+ data_cache_path ................................. None
93
+ data_parallel_random_init ....................... False
94
+ data_parallel_sharding_strategy ................. no_shard
95
+ data_parallel_size .............................. 8
96
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
97
+ data_per_class_fraction ......................... 1.0
98
+ data_sharding ................................... True
99
+ dataloader_type ................................. single
100
+ ddp_average_in_collective ....................... False
101
+ ddp_bucket_size ................................. None
102
+ ddp_num_buckets ................................. None
103
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
104
+ decoder_first_pipeline_num_layers ............... None
105
+ decoder_last_pipeline_num_layers ................ None
106
+ decoder_num_layers .............................. None
107
+ decoder_seq_length .............................. None
108
+ decoupled_lr .................................... None
109
+ decoupled_min_lr ................................ None
110
+ decrease_batch_size_if_needed ................... False
111
+ defer_embedding_wgrad_compute ................... False
112
+ delay_wgrad_compute ............................. False
113
+ deprecated_use_mcore_models ..................... True
114
+ deterministic_mode .............................. False
115
+ dino_bottleneck_size ............................ 256
116
+ dino_freeze_last_layer .......................... 1
117
+ dino_head_hidden_size ........................... 2048
118
+ dino_local_crops_number ......................... 10
119
+ dino_local_img_size ............................. 96
120
+ dino_norm_last_layer ............................ False
121
+ dino_teacher_temp ............................... 0.07
122
+ dino_warmup_teacher_temp ........................ 0.04
123
+ dino_warmup_teacher_temp_epochs ................. 30
124
+ disable_bf16_reduced_precision_matmul ........... False
125
+ disable_mamba_mem_eff_path ...................... False
126
+ disable_straggler_on_startup .................... False
127
+ dist_ckpt_format_deprecated ..................... None
128
+ dist_ckpt_strictness ............................ assume_ok_unexpected
129
+ distribute_saved_activations .................... False
130
+ distributed_backend ............................. nccl
131
+ distributed_timeout_minutes ..................... 10
132
+ embedding_init_method_std ....................... None
133
+ embedding_path .................................. None
134
+ empty_unused_memory_level ....................... 0
135
+ enable_cuda_graph ............................... True
136
+ enable_experimental ............................. False
137
+ enable_ft_package ............................... False
138
+ enable_gloo_process_groups ...................... True
139
+ enable_msc ...................................... True
140
+ enable_one_logger ............................... True
141
+ encoder_num_layers .............................. 16
142
+ encoder_seq_length .............................. 4096
143
+ end_weight_decay ................................ 0.1
144
+ eod_mask_loss ................................... False
145
+ error_injection_rate ............................ 0
146
+ error_injection_type ............................ transient_error
147
+ eval_interval ................................... 1000000
148
+ eval_iters ...................................... 0
149
+ evidence_data_path .............................. None
150
+ exit_duration_in_mins ........................... None
151
+ exit_interval ................................... None
152
+ exit_on_missing_checkpoint ...................... False
153
+ exit_signal_handler ............................. False
154
+ exp_avg_dtype ................................... torch.float32
155
+ exp_avg_sq_dtype ................................ torch.float32
156
+ expert_model_parallel_size ...................... 8
157
+ expert_tensor_parallel_size ..................... 1
158
+ export_force_local_attention .................... False
159
+ export_kd_cfg ................................... None
160
+ export_kd_teacher_ckpt_format ................... None
161
+ export_kd_teacher_load .......................... None
162
+ export_kv_cache_quant ........................... False
163
+ export_legacy_megatron .......................... False
164
+ export_model_type ............................... GPTModel
165
+ export_moe_apply_probs_on_input ................. False
166
+ export_qk_l2_norm ............................... False
167
+ export_quant_cfg ................................ None
168
+ export_real_quant_cfg ........................... None
169
+ export_te_mcore_model ........................... False
170
+ external_cuda_graph ............................. False
171
+ ffn_hidden_size ................................. 14336
172
+ finetune ........................................ False
173
+ finetune_data_split ............................. train
174
+ finetune_hf_dataset ............................. None
175
+ first_last_layers_bf16 .......................... False
176
+ flash_decode .................................... False
177
+ fp16 ............................................ False
178
+ fp16_lm_cross_entropy ........................... False
179
+ fp32_residual_connection ........................ False
180
+ fp8 ............................................. e4m3
181
+ fp8_amax_compute_algo ........................... most_recent
182
+ fp8_amax_history_len ............................ 1
183
+ fp8_interval .................................... 1
184
+ fp8_margin ...................................... 0
185
+ fp8_param_gather ................................ False
186
+ fp8_recipe ...................................... mxfp8
187
+ fp8_wgrad ....................................... True
188
+ fsdp_double_buffer .............................. False
189
+ global_batch_size ............................... 16
190
+ grad_reduce_in_bf16 ............................. False
191
+ gradient_accumulation_fusion .................... True
192
+ gradient_reduce_div_fusion ...................... True
193
+ group_query_attention ........................... True
194
+ head_lr_mult .................................... 1.0
195
+ heterogeneous_layers_config_encoded_json ........ None
196
+ heterogeneous_layers_config_path ................ None
197
+ hidden_dropout .................................. 0.0
198
+ hidden_size ..................................... 4096
199
+ hierarchical_context_parallel_sizes ............. None
200
+ high_priority_stream_groups ..................... []
201
+ hybrid_attention_ratio .......................... 0.0
202
+ hybrid_mlp_ratio ................................ 0.0
203
+ hybrid_override_pattern ......................... None
204
+ hysteresis ...................................... 2
205
+ ict_head_size ................................... None
206
+ ict_load ........................................ None
207
+ img_h ........................................... 224
208
+ img_w ........................................... 224
209
+ indexer_batch_size .............................. 128
210
+ indexer_log_interval ............................ 1000
211
+ inference_batch_times_seqlen_threshold .......... -1
212
+ inference_dynamic_batching ...................... False
213
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
214
+ inference_dynamic_batching_buffer_overflow_factor None
215
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
216
+ inference_dynamic_batching_chunk_size ........... 256
217
+ inference_dynamic_batching_max_requests_override None
218
+ inference_dynamic_batching_max_tokens_override .. None
219
+ inference_dynamic_batching_num_cuda_graphs ...... 16
220
+ inference_max_batch_size ........................ 8
221
+ inference_max_seq_length ........................ 2560
222
+ inference_rng_tracker ........................... False
223
+ init_method_std ................................. 0.01
224
+ init_method_xavier_uniform ...................... False
225
+ init_model_with_meta_device ..................... False
226
+ initial_loss_scale .............................. 4294967296
227
+ inprocess_active_world_size ..................... 8
228
+ inprocess_barrier_timeout ....................... 120
229
+ inprocess_completion_timeout .................... 120
230
+ inprocess_empty_cuda_cache ...................... False
231
+ inprocess_granularity ........................... node
232
+ inprocess_hard_timeout .......................... 90
233
+ inprocess_heartbeat_interval .................... 30
234
+ inprocess_heartbeat_timeout ..................... 60
235
+ inprocess_last_call_wait ........................ 1
236
+ inprocess_max_iterations ........................ None
237
+ inprocess_monitor_process_interval .............. 1.0
238
+ inprocess_monitor_thread_interval ............... 1.0
239
+ inprocess_progress_watchdog_interval ............ 1.0
240
+ inprocess_restart ............................... False
241
+ inprocess_soft_timeout .......................... 60
242
+ inprocess_termination_grace_time ................ 1
243
+ is_hybrid_model ................................. False
244
+ iter_per_epoch .................................. 1250
245
+ iterations_to_skip .............................. []
246
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
247
+ kitchen_config_file ............................. None
248
+ kitchen_recipe_number ........................... None
249
+ kv_channels ..................................... 128
250
+ kv_lora_rank .................................... 32
251
+ lazy_mpu_init ................................... None
252
+ load ............................................ None
253
+ load_main_params_from_ckpt ...................... None
254
+ load_model_opt_format ........................... False
255
+ local_rank ...................................... 0
256
+ log_energy ...................................... False
257
+ log_interval .................................... 1
258
+ log_loss_scale_to_tensorboard ................... True
259
+ log_memory_to_tensorboard ....................... False
260
+ log_num_zeros_in_grad ........................... False
261
+ log_params_norm ................................. False
262
+ log_progress .................................... False
263
+ log_straggler ................................... False
264
+ log_throughput .................................. True
265
+ log_timers_to_tensorboard ....................... False
266
+ log_validation_ppl_to_tensorboard ............... False
267
+ log_world_size_to_tensorboard ................... False
268
+ logging_level ................................... None
269
+ loss_scale ...................................... None
270
+ loss_scale_window ............................... 1000
271
+ lr .............................................. 0.0001
272
+ lr_decay_iters .................................. 8
273
+ lr_decay_samples ................................ None
274
+ lr_decay_style .................................. cosine
275
+ lr_warmup_fraction .............................. None
276
+ lr_warmup_init .................................. 0.0
277
+ lr_warmup_iters ................................. 5
278
+ lr_warmup_samples ............................... 0
279
+ lr_wsd_decay_iters .............................. None
280
+ lr_wsd_decay_samples ............................ None
281
+ lr_wsd_decay_style .............................. exponential
282
+ main_grads_dtype ................................ torch.float32
283
+ main_params_dtype ............................... torch.float32
284
+ make_vocab_size_divisible_by .................... 128
285
+ mamba_head_dim .................................. 64
286
+ mamba_num_groups ................................ 8
287
+ mamba_num_heads ................................. None
288
+ mamba_state_dim ................................. 128
289
+ manual_gc ....................................... False
290
+ manual_gc_eval .................................. True
291
+ manual_gc_interval .............................. 0
292
+ mask_factor ..................................... 1.0
293
+ mask_prob ....................................... 0.15
294
+ mask_type ....................................... random
295
+ masked_softmax_fusion ........................... False
296
+ max_position_embeddings ......................... 32768
297
+ max_tokens_to_oom ............................... 12000
298
+ memory_snapshot_path ............................ snapshot.pickle
299
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
300
+ micro_batch_size ................................ 1
301
+ microbatch_group_size_per_vp_stage .............. None
302
+ mid_level_dataset_surplus ....................... 0.005
303
+ min_loss_scale .................................. 1.0
304
+ min_lr .......................................... 1e-05
305
+ mlp_chunks_for_prefill .......................... 1
306
+ mmap_bin_files .................................. True
307
+ mock_data ....................................... False
308
+ moe_apply_probs_on_input ........................ False
309
+ moe_aux_loss_coeff .............................. 0.01
310
+ moe_deepep_num_sms .............................. 20
311
+ moe_echo_enable_random_offloading ............... False
312
+ moe_echo_expert_dispatch_overlap ................ True
313
+ moe_echo_expert_dispatcher_type ................. hybridep
314
+ moe_echo_recompute_expert_dispatch .............. False
315
+ moe_enable_deepep ............................... False
316
+ moe_enable_echo ................................. True
317
+ moe_enable_hybridep ............................. True
318
+ moe_expert_capacity_factor ...................... None
319
+ moe_extended_tp ................................. False
320
+ moe_ffn_hidden_size ............................. 3584
321
+ moe_grouped_gemm ................................ True
322
+ moe_hybridep_num_sms ............................ 16
323
+ moe_input_jitter_eps ............................ None
324
+ moe_layer_freq .................................. 1
325
+ moe_layer_recompute ............................. False
326
+ moe_num_echo_experts ............................ 8
327
+ moe_pad_expert_input_to_capacity ................ False
328
+ moe_per_layer_logging ........................... False
329
+ moe_permute_fusion .............................. False
330
+ moe_received_token_capacity ..................... 1.08203125
331
+ moe_router_bias_update_rate ..................... 0.001
332
+ moe_router_dtype ................................ fp32
333
+ moe_router_enable_expert_bias ................... False
334
+ moe_router_force_load_balancing ................. False
335
+ moe_router_group_topk ........................... None
336
+ moe_router_load_balancing_type .................. aux_loss
337
+ moe_router_num_groups ........................... None
338
+ moe_router_padding_for_fp8 ...................... False
339
+ moe_router_pre_softmax .......................... False
340
+ moe_router_score_function ....................... softmax
341
+ moe_router_topk ................................. 8
342
+ moe_router_topk_scaling_factor .................. None
343
+ moe_shared_expert_intermediate_size ............. None
344
+ moe_shared_expert_overlap ....................... False
345
+ moe_token_dispatcher_type ....................... flex
346
+ moe_token_drop_policy ........................... probs
347
+ moe_upcycling_granularity ....................... 1
348
+ moe_use_device_initiated_grouped_gemm ........... True
349
+ moe_use_legacy_grouped_gemm ..................... False
350
+ moe_use_upcycling ............................... False
351
+ moe_z_loss_coeff ................................ None
352
+ mrope_section ................................... None
353
+ mscale .......................................... 1.0
354
+ mscale_all_dim .................................. 0.0
355
+ mtp_loss_scaling_factor ......................... 0.1
356
+ mtp_num_layers .................................. None
357
+ multi_latent_attention .......................... False
358
+ nccl_all_reduce_for_prefill ..................... False
359
+ nccl_communicator_config_path ................... None
360
+ nccl_ub ......................................... False
361
+ no_load_optim ................................... None
362
+ no_load_rng ..................................... None
363
+ no_persist_layer_norm ........................... False
364
+ no_rope_freq .................................... None
365
+ no_save_optim ................................... None
366
+ no_save_rng ..................................... None
367
+ non_persistent_ckpt_type ........................ None
368
+ non_persistent_global_ckpt_dir .................. None
369
+ non_persistent_local_ckpt_algo .................. fully_parallel
370
+ non_persistent_local_ckpt_dir ................... None
371
+ non_persistent_save_interval .................... None
372
+ norm_epsilon .................................... 1e-05
373
+ normalization ................................... RMSNorm
374
+ num_attention_heads ............................. 32
375
+ num_channels .................................... 3
376
+ num_classes ..................................... 1000
377
+ num_dataset_builder_threads ..................... 1
378
+ num_distributed_optimizer_instances ............. 1
379
+ num_experts ..................................... 32
380
+ num_layers ...................................... 16
381
+ num_layers_at_end_in_bf16 ....................... 1
382
+ num_layers_at_start_in_bf16 ..................... 1
383
+ num_layers_per_virtual_pipeline_stage ........... None
384
+ num_query_groups ................................ 8
385
+ num_virtual_stages_per_pipeline_rank ............ None
386
+ num_workers ..................................... 2
387
+ object_storage_cache_path ....................... None
388
+ one_logger_async ................................ False
389
+ one_logger_project .............................. megatron-lm
390
+ one_logger_run_name ............................. None
391
+ onnx_safe ....................................... None
392
+ openai_gelu ..................................... False
393
+ optimizer ....................................... adam
394
+ optimizer_cpu_offload ........................... False
395
+ optimizer_offload_fraction ...................... 1.0
396
+ output_bert_embeddings .......................... False
397
+ overlap_cpu_optimizer_d2h_h2d ................... False
398
+ overlap_grad_reduce ............................. True
399
+ overlap_moe_expert_parallel_comm ................ False
400
+ overlap_p2p_comm ................................ False
401
+ overlap_p2p_comm_warmup_flush ................... False
402
+ overlap_param_gather ............................ True
403
+ overlap_param_gather_with_optimizer_step ........ False
404
+ override_opt_param_scheduler .................... False
405
+ params_dtype .................................... torch.bfloat16
406
+ patch_dim ....................................... 16
407
+ per_split_data_args_path ........................ None
408
+ perform_initialization .......................... True
409
+ pin_cpu_grads ................................... True
410
+ pin_cpu_params .................................. True
411
+ pipeline_model_parallel_comm_backend ............ None
412
+ pipeline_model_parallel_layout .................. None
413
+ pipeline_model_parallel_size .................... 1
414
+ position_embedding_type ......................... rope
415
+ pretrained_checkpoint ........................... None
416
+ profile ......................................... False
417
+ profile_ranks ................................... [0]
418
+ profile_step_end ................................ 12
419
+ profile_step_start .............................. 10
420
+ q_lora_rank ..................................... None
421
+ qk_head_dim ..................................... 128
422
+ qk_l2_norm ...................................... False
423
+ qk_layernorm .................................... False
424
+ qk_pos_emb_head_dim ............................. 64
425
+ query_in_block_prob ............................. 0.1
426
+ rampup_batch_size ............................... None
427
+ rank ............................................ 0
428
+ recompute_granularity ........................... None
429
+ recompute_method ................................ None
430
+ recompute_modules ............................... None
431
+ recompute_num_layers ............................ None
432
+ record_memory_history ........................... False
433
+ relative_attention_max_distance ................. 128
434
+ relative_attention_num_buckets .................. 32
435
+ replication ..................................... False
436
+ replication_factor .............................. 2
437
+ replication_jump ................................ None
438
+ rerun_mode ...................................... validate_results
439
+ reset_attention_mask ............................ False
440
+ reset_position_ids .............................. False
441
+ result_rejected_tracker_filename ................ None
442
+ retriever_report_topk_accuracies ................ []
443
+ retriever_score_scaling ......................... False
444
+ retriever_seq_length ............................ 256
445
+ retro_add_retriever ............................. False
446
+ retro_attention_gate ............................ 1
447
+ retro_cyclic_train_iters ........................ None
448
+ retro_encoder_attention_dropout ................. 0.1
449
+ retro_encoder_hidden_dropout .................... 0.1
450
+ retro_encoder_layers ............................ 2
451
+ retro_num_neighbors ............................. 2
452
+ retro_num_retrieved_chunks ...................... 2
453
+ retro_project_dir ............................... None
454
+ retro_verify_neighbor_count ..................... True
455
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
456
+ rope_scaling_factor ............................. 8.0
457
+ rope_type ....................................... None
458
+ rotary_base ..................................... 1000000
459
+ rotary_interleaved .............................. False
460
+ rotary_percent .................................. 1.0
461
+ rotary_scaling_factor ........................... 1.0
462
+ rotary_seq_len_interpolation_factor ............. None
463
+ run_workload_inspector_server ................... False
464
+ sample_rate ..................................... 1.0
465
+ save ............................................ None
466
+ save_interval ................................... None
467
+ save_retain_interval ............................ None
468
+ scatter_gather_tensors_in_pipeline .............. True
469
+ seed ............................................ 1234
470
+ seq_length ...................................... 4096
471
+ sequence_parallel ............................... False
472
+ sft ............................................. False
473
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
474
+ sgd_momentum .................................... 0.9
475
+ short_seq_prob .................................. 0.1
476
+ skip_train ...................................... False
477
+ skipped_train_samples ........................... 0
478
+ spec ............................................ None
479
+ split ........................................... 100,0,0
480
+ squared_relu .................................... False
481
+ start_weight_decay .............................. 0.1
482
+ straggler_ctrlr_port ............................ 65535
483
+ straggler_minmax_count .......................... 1
484
+ suggested_communication_unit_size ............... None
485
+ swiglu .......................................... True
486
+ swin_backbone_type .............................. tiny
487
+ symmetric_ar_type ............................... None
488
+ te_rng_tracker .................................. True
489
+ tensor_model_parallel_size ...................... 1
490
+ tensorboard_dir ................................. None
491
+ tensorboard_log_interval ........................ 1
492
+ tensorboard_queue_size .......................... 1000
493
+ test_data_path .................................. None
494
+ test_mode ....................................... False
495
+ tiktoken_num_special_tokens ..................... 1000
496
+ tiktoken_pattern ................................ None
497
+ tiktoken_special_tokens ......................... None
498
+ timing_log_level ................................ 0
499
+ timing_log_option ............................... minmax
500
+ titles_data_path ................................ None
501
+ tokenizer_model ................................. None
502
+ tokenizer_type .................................. GPT2BPETokenizer
503
+ torch_fsdp2_reshard_after_forward ............... True
504
+ tp_comm_bootstrap_backend ....................... nccl
505
+ tp_comm_bulk_dgrad .............................. True
506
+ tp_comm_bulk_wgrad .............................. True
507
+ tp_comm_overlap ................................. False
508
+ tp_comm_overlap_ag .............................. True
509
+ tp_comm_overlap_cfg ............................. None
510
+ tp_comm_overlap_rs .............................. True
511
+ tp_comm_overlap_rs_dgrad ........................ False
512
+ tp_comm_split_ag ................................ True
513
+ tp_comm_split_rs ................................ True
514
+ train_data_path ................................. None
515
+ train_iters ..................................... 12
516
+ train_samples ................................... None
517
+ train_sync_interval ............................. None
518
+ transformer_impl ................................ transformer_engine
519
+ transformer_pipeline_model_parallel_size ........ 1
520
+ untie_embeddings_and_output_weights ............. True
521
+ use_checkpoint_args ............................. False
522
+ use_checkpoint_opt_param_scheduler .............. False
523
+ use_cpu_initialization .......................... None
524
+ use_custom_fsdp ................................. False
525
+ use_dist_ckpt ................................... True
526
+ use_dist_ckpt_deprecated ........................ False
527
+ use_distributed_optimizer ....................... True
528
+ use_flash_attn .................................. False
529
+ use_legacy_models ............................... False
530
+ use_mp_args_from_checkpoint_args ................ False
531
+ use_one_sent_docs ............................... False
532
+ use_persistent_ckpt_worker ...................... False
533
+ use_precision_aware_optimizer ................... False
534
+ use_pytorch_profiler ............................ False
535
+ use_ring_exchange_p2p ........................... False
536
+ use_rope_scaling ................................ False
537
+ use_rotary_position_embeddings .................. False
538
+ use_sharp ....................................... False
539
+ use_tokenizer_model_from_checkpoint_args ........ True
540
+ use_torch_fsdp2 ................................. False
541
+ use_torch_optimizer_for_cpu_offload ............. False
542
+ use_tp_pp_dp_mapping ............................ False
543
+ v_head_dim ...................................... 128
544
+ valid_data_path ................................. None
545
+ variable_seq_lengths ............................ False
546
+ virtual_pipeline_model_parallel_size ............ None
547
+ vision_backbone_type ............................ vit
548
+ vision_pretraining .............................. False
549
+ vision_pretraining_type ......................... classify
550
+ vocab_extra_ids ................................. 0
551
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
552
+ vocab_size ...................................... None
553
+ wandb_exp_name ..................................
554
+ wandb_project ...................................
555
+ wandb_save_dir ..................................
556
+ weight_decay .................................... 0.1
557
+ weight_decay_incr_style ......................... constant
558
+ wgrad_deferral_limit ............................ 0
559
+ world_size ...................................... 8
560
+ yaml_cfg ........................................ None
561
+ -------------------- end of arguments ---------------------
562
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
563
+ > building GPT2BPETokenizer tokenizer ...
564
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
565
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
566
+ > initializing torch distributed ...
567
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
568
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
569
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
570
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
571
+ 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
572
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
573
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
574
+ /workspace/Megatron-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
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
577
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
578
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
579
+ [Gloo] Rank 3 is connected to 7[Gloo] Rank peer ranks. Expected number of connected peer ranks is : 72
580
+ is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 5 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 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 2[Gloo] Rank is connected to 70 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks.
586
+ Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank [Gloo] Rank 7 is connected to 76 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks. Expected number of connected peer ranks is :
591
+ 7
592
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
593
+ > initialized tensor model parallel with size 1
594
+ > initialized pipeline model parallel with size 1
595
+ > setting random seeds to 1234 ...
596
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
597
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
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
+ [rank1]:[W830 06:04:03.724086168 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.
604
+ [rank2]:[W830 06:04:03.724098729 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.
605
+ [rank5]:[W830 06:04:03.724182778 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.
606
+ [rank3]:[W830 06:04:03.724241005 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.
607
+ [rank6]:[W830 06:04:03.724266277 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 06:04:03.724295752 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
+ [rank7]:[W830 06:04:03.724305249 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ > compiling dataset index builder ...
611
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
612
+ make: Nothing to be done for 'default'.
613
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
614
+ >>> done with dataset index builder. Compilation time: 0.034 seconds
615
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
616
+ > compiling and loading fused kernels ...
617
+ [rank0]:[W830 06:04:03.879256173 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.
618
+ >>> done with compiling and loading fused kernels. Compilation time: 3.831 seconds
619
+ time to initialize megatron (seconds): 7.088
620
+ [after megatron is initialized] datetime: 2026-08-30 06:04:09
621
+ building GPT model ...
622
+ /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.
623
+ warnings.warn(
624
+ /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.
625
+ warnings.warn(
626
+ /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.
627
+ warnings.warn(
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3903983616
639
+ 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)
640
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
641
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
642
+ module.output_layer.weight
643
+ Params for bucket 2 (42086400 elements, 42086400 padded size):
644
+ module.decoder.layers.15.self_attention.linear_qkv.weight
645
+ module.decoder.layers.15.pre_mlp_layernorm.weight
646
+ module.decoder.final_layernorm.weight
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
+ Params for bucket 3 (42082304 elements, 42082304 padded size):
651
+ module.decoder.layers.14.mlp.router.weight
652
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
653
+ module.decoder.layers.14.self_attention.linear_proj.weight
654
+ module.decoder.layers.14.self_attention.linear_qkv.weight
655
+ module.decoder.layers.14.pre_mlp_layernorm.weight
656
+ Params for bucket 4 (42082304 elements, 42082304 padded size):
657
+ module.decoder.layers.13.mlp.router.weight
658
+ module.decoder.layers.13.self_attention.linear_proj.weight
659
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
660
+ module.decoder.layers.13.pre_mlp_layernorm.weight
661
+ module.decoder.layers.13.self_attention.linear_qkv.weight
662
+ Params for bucket 5 (42082304 elements, 42082304 padded size):
663
+ module.decoder.layers.12.self_attention.linear_qkv.weight
664
+ module.decoder.layers.12.pre_mlp_layernorm.weight
665
+ module.decoder.layers.12.mlp.router.weight
666
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
667
+ module.decoder.layers.12.self_attention.linear_proj.weight
668
+ Params for bucket 6 (42082304 elements, 42082304 padded size):
669
+ module.decoder.layers.11.mlp.router.weight
670
+ module.decoder.layers.11.self_attention.linear_proj.weight
671
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
672
+ module.decoder.layers.11.self_attention.linear_qkv.weight
673
+ module.decoder.layers.11.pre_mlp_layernorm.weight
674
+ Params for bucket 7 (42082304 elements, 42082304 padded size):
675
+ module.decoder.layers.10.pre_mlp_layernorm.weight
676
+ module.decoder.layers.10.self_attention.linear_qkv.weight
677
+ module.decoder.layers.10.mlp.router.weight
678
+ module.decoder.layers.10.self_attention.linear_proj.weight
679
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
680
+ Params for bucket 8 (42082304 elements, 42082304 padded size):
681
+ module.decoder.layers.9.mlp.router.weight
682
+ module.decoder.layers.9.self_attention.linear_proj.weight
683
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
684
+ module.decoder.layers.9.self_attention.linear_qkv.weight
685
+ module.decoder.layers.9.pre_mlp_layernorm.weight
686
+ Params for bucket 9 (42082304 elements, 42082304 padded size):
687
+ module.decoder.layers.8.self_attention.linear_qkv.weight
688
+ module.decoder.layers.8.pre_mlp_layernorm.weight
689
+ module.decoder.layers.8.mlp.router.weight
690
+ module.decoder.layers.8.self_attention.linear_proj.weight
691
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
692
+ Params for bucket 10 (42082304 elements, 42082304 padded size):
693
+ module.decoder.layers.7.mlp.router.weight
694
+ module.decoder.layers.7.self_attention.linear_proj.weight
695
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
696
+ module.decoder.layers.7.pre_mlp_layernorm.weight
697
+ module.decoder.layers.7.self_attention.linear_qkv.weight
698
+ Params for bucket 11 (42082304 elements, 42082304 padded size):
699
+ module.decoder.layers.6.self_attention.linear_proj.weight
700
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
701
+ module.decoder.layers.6.mlp.router.weight
702
+ module.decoder.layers.6.self_attention.linear_qkv.weight
703
+ module.decoder.layers.6.pre_mlp_layernorm.weight
704
+ Params for bucket 12 (42082304 elements, 42082304 padded size):
705
+ module.decoder.layers.5.mlp.router.weight
706
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
707
+ module.decoder.layers.5.self_attention.linear_qkv.weight
708
+ module.decoder.layers.5.self_attention.linear_proj.weight
709
+ module.decoder.layers.5.pre_mlp_layernorm.weight
710
+ Params for bucket 13 (42082304 elements, 42082304 padded size):
711
+ module.decoder.layers.4.pre_mlp_layernorm.weight
712
+ module.decoder.layers.4.self_attention.linear_qkv.weight
713
+ module.decoder.layers.4.mlp.router.weight
714
+ module.decoder.layers.4.self_attention.linear_proj.weight
715
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
716
+ Params for bucket 14 (42082304 elements, 42082304 padded size):
717
+ module.decoder.layers.3.mlp.router.weight
718
+ module.decoder.layers.3.self_attention.linear_proj.weight
719
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
720
+ module.decoder.layers.3.self_attention.linear_qkv.weight
721
+ module.decoder.layers.3.pre_mlp_layernorm.weight
722
+ Params for bucket 15 (42082304 elements, 42082304 padded size):
723
+ module.decoder.layers.2.self_attention.linear_qkv.weight
724
+ module.decoder.layers.2.pre_mlp_layernorm.weight
725
+ module.decoder.layers.2.mlp.router.weight
726
+ module.decoder.layers.2.self_attention.linear_proj.weight
727
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
728
+ Params for bucket 16 (42082304 elements, 42082304 padded size):
729
+ module.decoder.layers.1.pre_mlp_layernorm.weight
730
+ module.decoder.layers.1.self_attention.linear_qkv.weight
731
+ module.decoder.layers.1.self_attention.linear_proj.weight
732
+ module.decoder.layers.1.mlp.router.weight
733
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
734
+ Params for bucket 17 (42082304 elements, 42082304 padded size):
735
+ module.decoder.layers.0.mlp.router.weight
736
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
737
+ module.decoder.layers.0.self_attention.linear_proj.weight
738
+ module.decoder.layers.0.pre_mlp_layernorm.weight
739
+ module.decoder.layers.0.self_attention.linear_qkv.weight
740
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
741
+ module.embedding.word_embeddings.weight
742
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56
743
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
744
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
745
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
746
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
747
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
748
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
750
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
751
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
752
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
753
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
754
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
755
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
756
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
757
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
758
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
759
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
760
+ Params for bucket 6 (58720256 elements, 58720256 padded size):
761
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
762
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
763
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
764
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
765
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
766
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
767
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
768
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
769
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
770
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
771
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
772
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
773
+ Params for bucket 10 (58720256 elements, 58720256 padded size):
774
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
775
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
776
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
777
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
778
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
779
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
780
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
781
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
782
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
783
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
784
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
785
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
786
+ Params for bucket 14 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
788
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
790
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
791
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
792
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
793
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
794
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
795
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
796
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
797
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
798
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
799
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
800
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
801
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
802
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
803
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
804
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
805
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
806
+ Params for bucket 20 (58720256 elements, 58720256 padded size):
807
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
808
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
809
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
811
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
812
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
813
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
814
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
815
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
816
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
817
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
818
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
819
+ Params for bucket 24 (58720256 elements, 58720256 padded size):
820
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
821
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
822
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
823
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
824
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
825
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
827
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
828
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
829
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
830
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
831
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
832
+ Params for bucket 28 (58720256 elements, 58720256 padded size):
833
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
834
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
835
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
836
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
837
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
838
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
839
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
840
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
841
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
842
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
843
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
844
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
845
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
846
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
847
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
848
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
849
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
850
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
851
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
852
+ Params for bucket 34 (58720256 elements, 58720256 padded size):
853
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
854
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
855
+ Params for bucket 35 (58720256 elements, 58720256 padded size):
856
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
857
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
858
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
859
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
860
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
861
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
862
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
863
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
864
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
865
+ Params for bucket 38 (58720256 elements, 58720256 padded size):
866
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
867
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
868
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
869
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
870
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
871
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
872
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
873
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
874
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
875
+ Params for bucket 41 (58720256 elements, 58720256 padded size):
876
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
877
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
878
+ Params for bucket 42 (58720256 elements, 58720256 padded size):
879
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
880
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
881
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
882
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
883
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
884
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
885
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
886
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
887
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
888
+ Params for bucket 45 (58720256 elements, 58720256 padded size):
889
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
890
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
891
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
892
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
893
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
894
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
895
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
896
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
897
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
898
+ Params for bucket 48 (58720256 elements, 58720256 padded size):
899
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
900
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
901
+ Params for bucket 49 (58720256 elements, 58720256 padded size):
902
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
903
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
904
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
905
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
906
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
907
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
908
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
909
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
910
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
911
+ Params for bucket 52 (58720256 elements, 58720256 padded size):
912
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
913
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
914
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
915
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
916
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
917
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
918
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
919
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
920
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
921
+ Params for bucket 55 (58720256 elements, 58720256 padded size):
922
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
923
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
924
+ Params for bucket 56 (58720256 elements, 58720256 padded size):
925
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
926
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
927
+ 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 0x78d57e558740>, config_logger_dir='')
928
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
929
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 06:04:09
930
+ > building train, validation, and test datasets ...
931
+ > datasets target sizes (minimum size):
932
+ train: 192
933
+ validation: 0
934
+ test: 0
935
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
936
+ > building train, validation, and test datasets for GPT ...
937
+ 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 0x78d57e7c3a40>, 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)
938
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
939
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
940
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
941
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
942
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
943
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
944
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
945
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
946
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
947
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
948
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
949
+ > finished creating GPT datasets ...
950
+ [after dataloaders are built] datetime: 2026-08-30 06:04:09
951
+ done with setup ...
952
+ (min, max) time across ranks (ms):
953
+ model-and-optimizer-setup ......................: (339.92, 343.36)
954
+ train/valid/test-data-iterators-setup ..........: (62.14, 78.88)
955
+ training ...
956
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
957
+ [before the start of training step] datetime: 2026-08-30 06:04:09
958
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
959
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
960
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
961
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
962
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
963
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
964
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
965
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
966
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
967
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
968
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
969
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
970
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
971
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
972
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
973
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
974
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
975
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
976
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
977
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
978
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
979
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
980
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
981
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
982
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
983
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
985
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
989
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
991
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
992
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
993
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
994
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
995
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
996
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
997
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
998
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
999
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1000
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1001
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1002
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1003
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1004
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1005
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1006
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1007
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1008
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1009
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1010
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1011
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1012
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1013
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1014
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1015
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1016
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1017
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1018
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1019
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1020
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1021
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1022
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1023
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1024
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1025
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1026
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1027
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1028
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1029
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1030
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1031
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1032
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1033
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1034
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1035
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1036
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1037
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1038
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1039
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1040
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1041
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1042
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1043
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1044
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1045
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1046
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1047
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1048
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1049
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1050
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1051
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1052
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1053
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1054
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1055
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1056
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1057
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1058
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1059
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1060
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1061
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1062
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1063
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1064
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1065
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1066
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1067
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1068
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1069
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1070
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1071
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1072
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1073
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1074
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1075
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1076
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1077
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1078
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1079
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1080
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1081
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1082
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1083
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1084
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1085
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1086
+ Number of parameters in transformer block in billions: 23.22 [2026-08-30 06:04:47] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38010.6 | throughput per GPU (TFLOP/s/GPU): 8.8 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103012E+01 | loss scale: 1.0 | grad norm: 44.244 | number of skipped iterations: 0 | number of nan iterations: 0 |
1087
+
1088
+ Number of parameters in embedding layers in billions: 0.41
1089
+ Total number of parameters in billions: 23.63
1090
+ Number of parameters in most loaded shard in billions: 23.6320
1091
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1092
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 59522.912109375 | max allocated: 68963.6640625 | reserved: 73392.0 | max reserved: 73392.0
1093
+ [2026-08-30 06:04:48] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 575.0 | throughput per GPU (TFLOP/s/GPU): 579.8 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101632E+01 | loss scale: 1.0 | grad norm: 45.757 | number of skipped iterations: 0 | number of nan iterations: 0 |
1094
+ [2026-08-30 06:04:49] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 484.4 | throughput per GPU (TFLOP/s/GPU): 688.2 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.022507E+01 | loss scale: 1.0 | grad norm: 192.019 | number of skipped iterations: 0 | number of nan iterations: 0 |
1095
+ Capture CUDA graph for training!!!
1096
+ CUDA graph capture done!!!
1097
+ [2026-08-30 06:04:51] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2228.2 | throughput per GPU (TFLOP/s/GPU): 149.6 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.325150E+01 | loss scale: 1.0 | grad norm: 30.053 | number of skipped iterations: 0 | number of nan iterations: 0 |
1098
+ [2026-08-30 06:04:51] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 393.6 | throughput per GPU (TFLOP/s/GPU): 847.0 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.038983E+01 | loss scale: 1.0 | grad norm: 16.542 | number of skipped iterations: 0 | number of nan iterations: 0 |
1099
+ [2026-08-30 06:04:52] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 394.8 | throughput per GPU (TFLOP/s/GPU): 844.4 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.050704E+01 | loss scale: 1.0 | grad norm: 41.610 | number of skipped iterations: 0 | number of nan iterations: 0 |
1100
+ [2026-08-30 06:04:52] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 396.1 | throughput per GPU (TFLOP/s/GPU): 841.6 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.010933E+01 | loss scale: 1.0 | grad norm: 8.142 | number of skipped iterations: 0 | number of nan iterations: 0 |
1101
+ [2026-08-30 06:04:52] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 393.1 | throughput per GPU (TFLOP/s/GPU): 848.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.571067E+00 | loss scale: 1.0 | grad norm: 5.902 | number of skipped iterations: 0 | number of nan iterations: 0 |
1102
+ [2026-08-30 06:04:53] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 394.9 | throughput per GPU (TFLOP/s/GPU): 844.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.311386E+00 | loss scale: 1.0 | grad norm: 5.779 | number of skipped iterations: 0 | number of nan iterations: 0 |
1103
+ [2026-08-30 06:04:53] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 395.3 | throughput per GPU (TFLOP/s/GPU): 843.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.134122E+00 | loss scale: 1.0 | grad norm: 5.569 | number of skipped iterations: 0 | number of nan iterations: 0 |
1104
+ [2026-08-30 06:04:54] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 397.1 | throughput per GPU (TFLOP/s/GPU): 839.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.004296E+00 | loss scale: 1.0 | grad norm: 5.267 | number of skipped iterations: 0 | number of nan iterations: 0 |
1105
+ [2026-08-30 06:04:54] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 393.0 | throughput per GPU (TFLOP/s/GPU): 848.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.883189E+00 | loss scale: 1.0 | grad norm: 5.035 | number of skipped iterations: 0 | number of nan iterations: 0 |
1106
+ [after training is done] datetime: 2026-08-30 06:04:54
1107
+ W0830 06:04:54.413000 568159 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1108
+ W0830 06:04:54.414000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568248 closing signal SIGTERM
1109
+ W0830 06:04:54.417000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568249 closing signal SIGTERM
1110
+ W0830 06:04:54.419000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568250 closing signal SIGTERM
1111
+ W0830 06:04:54.420000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568251 closing signal SIGTERM
1112
+ W0830 06:04:54.426000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568252 closing signal SIGTERM
1113
+ W0830 06:04:54.428000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568253 closing signal SIGTERM
1114
+ W0830 06:04:54.435000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568254 closing signal SIGTERM
1115
+ W0830 06:04:54.443000 568159 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 568255 closing signal SIGTERM
1116
+ Traceback (most recent call last):
1117
+ File "/usr/local/bin/torchrun", line 8, in <module>
1118
+ sys.exit(main())
1119
+ ^^^^^^
1120
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1121
+ return f(*args, **kwargs)
1122
+ ^^^^^^^^^^^^^^^^^^
1123
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1124
+ run(args)
1125
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1126
+ elastic_launch(
1127
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1128
+ return launch_agent(self._config, self._entrypoint, list(args))
1129
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1130
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1131
+ result = agent.run()
1132
+ ^^^^^^^^^^^
1133
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1134
+ result = f(*args, **kwargs)
1135
+ ^^^^^^^^^^^^^^^^^^
1136
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1137
+ result = self._invoke_run(role)
1138
+ ^^^^^^^^^^^^^^^^^^^^^^
1139
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1140
+ time.sleep(monitor_interval)
1141
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1142
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1143
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 568159 got signal: 15
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3/mem_samples.csv ADDED
@@ -0,0 +1,248 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 0, 0
2
+ 1, 0
3
+ 2, 0
4
+ 3, 0
5
+ 4, 0
6
+ 5, 0
7
+ 6, 0
8
+ 7, 0
9
+ 0, 0
10
+ 1, 0
11
+ 2, 0
12
+ 3, 0
13
+ 4, 0
14
+ 5, 0
15
+ 6, 0
16
+ 7, 0
17
+ 0, 1
18
+ 1, 1
19
+ 2, 1
20
+ 3, 1
21
+ 4, 1
22
+ 5, 1
23
+ 6, 1
24
+ 7, 1
25
+ 0, 4
26
+ 1, 4
27
+ 2, 4
28
+ 3, 4
29
+ 4, 1
30
+ 5, 1
31
+ 6, 1
32
+ 7, 1
33
+ 0, 4
34
+ 1, 4
35
+ 2, 4
36
+ 3, 4
37
+ 4, 4
38
+ 5, 4
39
+ 6, 4
40
+ 7, 4
41
+ 0, 819
42
+ 1, 819
43
+ 2, 819
44
+ 3, 819
45
+ 4, 819
46
+ 5, 819
47
+ 6, 819
48
+ 7, 819
49
+ 0, 1537
50
+ 1, 1537
51
+ 2, 1537
52
+ 3, 1537
53
+ 4, 1537
54
+ 5, 1537
55
+ 6, 1537
56
+ 7, 1537
57
+ 0, 2231
58
+ 1, 2231
59
+ 2, 2231
60
+ 3, 2231
61
+ 4, 2231
62
+ 5, 2231
63
+ 6, 2231
64
+ 7, 2231
65
+ 0, 39055
66
+ 1, 39065
67
+ 2, 39067
68
+ 3, 39065
69
+ 4, 39073
70
+ 5, 39113
71
+ 6, 39107
72
+ 7, 39101
73
+ 0, 39359
74
+ 1, 39367
75
+ 2, 39395
76
+ 3, 39373
77
+ 4, 39391
78
+ 5, 39399
79
+ 6, 39391
80
+ 7, 39379
81
+ 0, 41893
82
+ 1, 41885
83
+ 2, 41885
84
+ 3, 41885
85
+ 4, 41885
86
+ 5, 41885
87
+ 6, 41885
88
+ 7, 41885
89
+ 0, 41951
90
+ 1, 41943
91
+ 2, 41943
92
+ 3, 41943
93
+ 4, 41943
94
+ 5, 41943
95
+ 6, 41943
96
+ 7, 41943
97
+ 0, 42071
98
+ 1, 42045
99
+ 2, 42061
100
+ 3, 42103
101
+ 4, 42073
102
+ 5, 42055
103
+ 6, 42109
104
+ 7, 42073
105
+ 0, 42865
106
+ 1, 42857
107
+ 2, 42857
108
+ 3, 42857
109
+ 4, 42857
110
+ 5, 42857
111
+ 6, 42857
112
+ 7, 42857
113
+ 0, 43143
114
+ 1, 43135
115
+ 2, 43135
116
+ 3, 43135
117
+ 4, 43135
118
+ 5, 43135
119
+ 6, 43135
120
+ 7, 43135
121
+ 0, 44567
122
+ 1, 44559
123
+ 2, 44559
124
+ 3, 44559
125
+ 4, 44559
126
+ 5, 44559
127
+ 6, 44559
128
+ 7, 44559
129
+ 0, 44567
130
+ 1, 44559
131
+ 2, 44559
132
+ 3, 44559
133
+ 4, 44559
134
+ 5, 44559
135
+ 6, 44559
136
+ 7, 44559
137
+ 0, 67177
138
+ 1, 67155
139
+ 2, 67155
140
+ 3, 67155
141
+ 4, 67169
142
+ 5, 67155
143
+ 6, 67155
144
+ 7, 67155
145
+ 0, 67663
146
+ 1, 67641
147
+ 2, 67641
148
+ 3, 67641
149
+ 4, 67655
150
+ 5, 67641
151
+ 6, 67641
152
+ 7, 67641
153
+ 0, 67663
154
+ 1, 67641
155
+ 2, 67641
156
+ 3, 67641
157
+ 4, 67655
158
+ 5, 67641
159
+ 6, 67641
160
+ 7, 67641
161
+ 0, 67663
162
+ 1, 67641
163
+ 2, 67641
164
+ 3, 67641
165
+ 4, 67655
166
+ 5, 67641
167
+ 6, 67641
168
+ 7, 67641
169
+ 0, 78809
170
+ 1, 78243
171
+ 2, 78825
172
+ 3, 78795
173
+ 4, 78525
174
+ 5, 78791
175
+ 6, 78471
176
+ 7, 78871
177
+ 0, 79699
178
+ 1, 79081
179
+ 2, 79663
180
+ 3, 79665
181
+ 4, 79357
182
+ 5, 79661
183
+ 6, 79343
184
+ 7, 79679
185
+ 0, 80573
186
+ 1, 80327
187
+ 2, 80775
188
+ 3, 80775
189
+ 4, 80565
190
+ 5, 80775
191
+ 6, 80439
192
+ 7, 80775
193
+ 0, 81429
194
+ 1, 81201
195
+ 2, 81621
196
+ 3, 81657
197
+ 4, 81315
198
+ 5, 81625
199
+ 6, 81321
200
+ 7, 81667
201
+ 0, 77541
202
+ 1, 81109
203
+ 2, 89883
204
+ 3, 87277
205
+ 4, 78711
206
+ 5, 79341
207
+ 6, 80381
208
+ 7, 79401
209
+ 0, 120065
210
+ 1, 120115
211
+ 2, 120115
212
+ 3, 120337
213
+ 4, 119777
214
+ 5, 120113
215
+ 6, 120003
216
+ 7, 120111
217
+ 0, 120065
218
+ 1, 120115
219
+ 2, 120115
220
+ 3, 120337
221
+ 4, 119777
222
+ 5, 120113
223
+ 6, 120003
224
+ 7, 120111
225
+ 0, 119153
226
+ 1, 119203
227
+ 2, 119203
228
+ 3, 119425
229
+ 4, 118865
230
+ 5, 119201
231
+ 6, 119091
232
+ 7, 119199
233
+ 0, 2087
234
+ 1, 119203
235
+ 2, 3114
236
+ 3, 1834
237
+ 4, 1834
238
+ 5, 1834
239
+ 6, 1834
240
+ 7, 1834
241
+ 0, 0
242
+ 1, 0
243
+ 2, 0
244
+ 3, 0
245
+ 4, 0
246
+ 5, 0
247
+ 6, 0
248
+ 7, 0
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=echo
2
+ config=G2
3
+ family=tierA
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=bracing-regions-through-kernels
6
+ commit=9a236793
7
+ num_layers=16
8
+ hidden=4096
9
+ ffn_hidden=14336
10
+ moe_ffn_hidden=3584
11
+ num_experts=32
12
+ topk=8
13
+ experts_per_rank=4
14
+ expert_params=44040192
15
+ heads=32
16
+ kv_groups=8
17
+ kv_channels=
18
+ qk_layernorm=0
19
+ shared_ffn=0
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=0
23
+ dense_first_layers=0
24
+ moe_layers=16
25
+ rotary_base=1000000
26
+ rotary_scaling=
27
+ qkv_bias=0
28
+ router_balancing=aux_loss
29
+ aux_loss_coeff=1e-2
30
+ router_pre_softmax=0
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-5
35
+ seq_length=4096
36
+ micro_batch_size=1
37
+ global_batch_size=16
38
+ num_microbatches=2
39
+ assignments_per_rank=32768
40
+ mean_tokens_per_expert=8192
41
+ slot_policy=one
42
+ slots_per_rank=1
43
+ capacity_mode=static
44
+ capacity=1.08203125
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=12
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=
59
+ ultraep_weight_sync_join=
60
+ profile_enable=0
61
+ kernel_trace=0
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3/train.log ADDED
@@ -0,0 +1,1107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ G2 / echo (slots/rank=1 policy=one)
3
+ E=32 K=8 H'=3584 m=4
4
+ L=16 seq=4096 gbs=16 (2 microbatches)
5
+ cap_mode=static cap=1.08203125 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 9a236793
8
+ ==============================================================
9
+ W0830 06:05:07.346000 571546 torch/distributed/run.py:766]
10
+ W0830 06:05:07.346000 571546 torch/distributed/run.py:766] *****************************************
11
+ W0830 06:05:07.346000 571546 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 06:05:07.346000 571546 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
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
22
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
23
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
24
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
25
+ 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
26
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
27
+ Number of virtual stages per pipeline stage: None
28
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
29
+ using torch.bfloat16 for parameters ...
30
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
31
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
32
+ ------------------------ arguments ------------------------
33
+ account_for_embedding_in_pipeline_split ......... False
34
+ account_for_loss_in_pipeline_split .............. False
35
+ accumulate_allreduce_grads_in_fp32 .............. True
36
+ adam_beta1 ...................................... 0.9
37
+ adam_beta2 ...................................... 0.999
38
+ adam_eps ........................................ 1e-08
39
+ add_bias_linear ................................. False
40
+ add_position_embedding .......................... False
41
+ add_qkv_bias .................................... False
42
+ adlr_autoresume ................................. False
43
+ adlr_autoresume_interval ........................ 1000
44
+ align_grad_reduce ............................... True
45
+ align_param_gather .............................. False
46
+ app_tag_run_name ................................ None
47
+ app_tag_run_version ............................. 0.0.0
48
+ apply_layernorm_1p .............................. False
49
+ apply_query_key_layer_scaling ................... False
50
+ apply_residual_connection_post_layernorm ........ False
51
+ apply_rope_fusion ............................... True
52
+ async_save ...................................... None
53
+ async_tensor_model_parallel_allreduce ........... True
54
+ attention_backend ............................... AttnBackend.auto
55
+ attention_dropout ............................... 0.0
56
+ attention_softmax_in_fp32 ....................... False
57
+ auto_detect_ckpt_format ......................... False
58
+ barrier_with_L1_time ............................ True
59
+ bert_binary_head ................................ True
60
+ bert_embedder_type .............................. megatron
61
+ bert_load ....................................... None
62
+ bf16 ............................................ True
63
+ bias_dropout_fusion ............................. True
64
+ bias_gelu_fusion ................................ False
65
+ bias_swiglu_fusion .............................. True
66
+ biencoder_projection_dim ........................ 0
67
+ biencoder_shared_query_context_model ............ False
68
+ block_data_path ................................. None
69
+ calc_ft_timeouts ................................ False
70
+ calculate_per_token_loss ........................ False
71
+ check_for_large_grads ........................... False
72
+ check_for_nan_in_loss_and_grad .................. False
73
+ check_for_spiky_loss ............................ False
74
+ check_weight_hash_across_dp_replicas_interval ... None
75
+ ckpt_assume_constant_structure .................. False
76
+ ckpt_convert_format ............................. None
77
+ ckpt_convert_save ............................... None
78
+ ckpt_convert_update_legacy_dist_opt_format ...... False
79
+ ckpt_format ..................................... torch_dist
80
+ ckpt_fully_parallel_load ........................ False
81
+ ckpt_fully_parallel_save ........................ True
82
+ ckpt_fully_parallel_save_deprecated ............. False
83
+ ckpt_step ....................................... None
84
+ classes_fraction ................................ 1.0
85
+ clip_grad ....................................... 1.0
86
+ clone_scatter_output_in_embedding ............... True
87
+ config_logger_dir ...............................
88
+ consumed_train_samples .......................... 0
89
+ consumed_valid_samples .......................... 0
90
+ context_parallel_size ........................... 1
91
+ cp_comm_type .................................... ['p2p']
92
+ create_attention_mask_in_dataloader ............. True
93
+ cross_entropy_fusion_impl ....................... native
94
+ cross_entropy_loss_fusion ....................... False
95
+ cuda_graph_scope ................................ full_iteration
96
+ cuda_graph_warmup_steps ......................... 3
97
+ data_args_path .................................. None
98
+ data_cache_path ................................. None
99
+ data_parallel_random_init ....................... False
100
+ data_parallel_sharding_strategy ................. no_shard
101
+ data_parallel_size .............................. 8
102
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
103
+ data_per_class_fraction ......................... 1.0
104
+ data_sharding ................................... True
105
+ dataloader_type ................................. single
106
+ ddp_average_in_collective ....................... False
107
+ ddp_bucket_size ................................. None
108
+ ddp_num_buckets ................................. None
109
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
110
+ decoder_first_pipeline_num_layers ............... None
111
+ decoder_last_pipeline_num_layers ................ None
112
+ decoder_num_layers .............................. None
113
+ decoder_seq_length .............................. None
114
+ decoupled_lr .................................... None
115
+ decoupled_min_lr ................................ None
116
+ decrease_batch_size_if_needed ................... False
117
+ defer_embedding_wgrad_compute ................... False
118
+ delay_wgrad_compute ............................. False
119
+ deprecated_use_mcore_models ..................... True
120
+ deterministic_mode .............................. False
121
+ dino_bottleneck_size ............................ 256
122
+ dino_freeze_last_layer .......................... 1
123
+ dino_head_hidden_size ........................... 2048
124
+ dino_local_crops_number ......................... 10
125
+ dino_local_img_size ............................. 96
126
+ dino_norm_last_layer ............................ False
127
+ dino_teacher_temp ............................... 0.07
128
+ dino_warmup_teacher_temp ........................ 0.04
129
+ dino_warmup_teacher_temp_epochs ................. 30
130
+ disable_bf16_reduced_precision_matmul ........... False
131
+ disable_mamba_mem_eff_path ...................... False
132
+ disable_straggler_on_startup .................... False
133
+ dist_ckpt_format_deprecated ..................... None
134
+ dist_ckpt_strictness ............................ assume_ok_unexpected
135
+ distribute_saved_activations .................... False
136
+ distributed_backend ............................. nccl
137
+ distributed_timeout_minutes ..................... 10
138
+ embedding_init_method_std ....................... None
139
+ embedding_path .................................. None
140
+ empty_unused_memory_level ....................... 0
141
+ enable_cuda_graph ............................... True
142
+ enable_experimental ............................. False
143
+ enable_ft_package ............................... False
144
+ enable_gloo_process_groups ...................... True
145
+ enable_msc ...................................... True
146
+ enable_one_logger ............................... True
147
+ encoder_num_layers .............................. 16
148
+ encoder_seq_length .............................. 4096
149
+ end_weight_decay ................................ 0.1
150
+ eod_mask_loss ................................... False
151
+ error_injection_rate ............................ 0
152
+ error_injection_type ............................ transient_error
153
+ eval_interval ................................... 1000000
154
+ eval_iters ...................................... 0
155
+ evidence_data_path .............................. None
156
+ exit_duration_in_mins ........................... None
157
+ exit_interval ................................... None
158
+ exit_on_missing_checkpoint ...................... False
159
+ exit_signal_handler ............................. False
160
+ exp_avg_dtype ................................... torch.float32
161
+ exp_avg_sq_dtype ................................ torch.float32
162
+ expert_model_parallel_size ...................... 8
163
+ expert_tensor_parallel_size ..................... 1
164
+ export_force_local_attention .................... False
165
+ export_kd_cfg ................................... None
166
+ export_kd_teacher_ckpt_format ................... None
167
+ export_kd_teacher_load .......................... None
168
+ export_kv_cache_quant ........................... False
169
+ export_legacy_megatron .......................... False
170
+ export_model_type ............................... GPTModel
171
+ export_moe_apply_probs_on_input ................. False
172
+ export_qk_l2_norm ............................... False
173
+ export_quant_cfg ................................ None
174
+ export_real_quant_cfg ........................... None
175
+ export_te_mcore_model ........................... False
176
+ external_cuda_graph ............................. False
177
+ ffn_hidden_size ................................. 14336
178
+ finetune ........................................ False
179
+ finetune_data_split ............................. train
180
+ finetune_hf_dataset ............................. None
181
+ first_last_layers_bf16 .......................... False
182
+ flash_decode .................................... False
183
+ fp16 ............................................ False
184
+ fp16_lm_cross_entropy ........................... False
185
+ fp32_residual_connection ........................ False
186
+ fp8 ............................................. e4m3
187
+ fp8_amax_compute_algo ........................... most_recent
188
+ fp8_amax_history_len ............................ 1
189
+ fp8_interval .................................... 1
190
+ fp8_margin ...................................... 0
191
+ fp8_param_gather ................................ False
192
+ fp8_recipe ...................................... mxfp8
193
+ fp8_wgrad ....................................... True
194
+ fsdp_double_buffer .............................. False
195
+ global_batch_size ............................... 16
196
+ grad_reduce_in_bf16 ............................. False
197
+ gradient_accumulation_fusion .................... True
198
+ gradient_reduce_div_fusion ...................... True
199
+ group_query_attention ........................... True
200
+ head_lr_mult .................................... 1.0
201
+ heterogeneous_layers_config_encoded_json ........ None
202
+ heterogeneous_layers_config_path ................ None
203
+ hidden_dropout .................................. 0.0
204
+ hidden_size ..................................... 4096
205
+ hierarchical_context_parallel_sizes ............. None
206
+ high_priority_stream_groups ..................... []
207
+ hybrid_attention_ratio .......................... 0.0
208
+ hybrid_mlp_ratio ................................ 0.0
209
+ hybrid_override_pattern ......................... None
210
+ hysteresis ...................................... 2
211
+ ict_head_size ................................... None
212
+ ict_load ........................................ None
213
+ img_h ........................................... 224
214
+ img_w ........................................... 224
215
+ indexer_batch_size .............................. 128
216
+ indexer_log_interval ............................ 1000
217
+ inference_batch_times_seqlen_threshold .......... -1
218
+ inference_dynamic_batching ...................... False
219
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
220
+ inference_dynamic_batching_buffer_overflow_factor None
221
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
222
+ inference_dynamic_batching_chunk_size ........... 256
223
+ inference_dynamic_batching_max_requests_override None
224
+ inference_dynamic_batching_max_tokens_override .. None
225
+ inference_dynamic_batching_num_cuda_graphs ...... 16
226
+ inference_max_batch_size ........................ 8
227
+ inference_max_seq_length ........................ 2560
228
+ inference_rng_tracker ........................... False
229
+ init_method_std ................................. 0.01
230
+ init_method_xavier_uniform ...................... False
231
+ init_model_with_meta_device ..................... False
232
+ initial_loss_scale .............................. 4294967296
233
+ inprocess_active_world_size ..................... 8
234
+ inprocess_barrier_timeout ....................... 120
235
+ inprocess_completion_timeout .................... 120
236
+ inprocess_empty_cuda_cache ...................... False
237
+ inprocess_granularity ........................... node
238
+ inprocess_hard_timeout .......................... 90
239
+ inprocess_heartbeat_interval .................... 30
240
+ inprocess_heartbeat_timeout ..................... 60
241
+ inprocess_last_call_wait ........................ 1
242
+ inprocess_max_iterations ........................ None
243
+ inprocess_monitor_process_interval .............. 1.0
244
+ inprocess_monitor_thread_interval ............... 1.0
245
+ inprocess_progress_watchdog_interval ............ 1.0
246
+ inprocess_restart ............................... False
247
+ inprocess_soft_timeout .......................... 60
248
+ inprocess_termination_grace_time ................ 1
249
+ is_hybrid_model ................................. False
250
+ iter_per_epoch .................................. 1250
251
+ iterations_to_skip .............................. []
252
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
253
+ kitchen_config_file ............................. None
254
+ kitchen_recipe_number ........................... None
255
+ kv_channels ..................................... 128
256
+ kv_lora_rank .................................... 32
257
+ lazy_mpu_init ................................... None
258
+ load ............................................ None
259
+ load_main_params_from_ckpt ...................... None
260
+ load_model_opt_format ........................... False
261
+ local_rank ...................................... 0
262
+ log_energy ...................................... False
263
+ log_interval .................................... 1
264
+ log_loss_scale_to_tensorboard ................... True
265
+ log_memory_to_tensorboard ....................... False
266
+ log_num_zeros_in_grad ........................... False
267
+ log_params_norm ................................. False
268
+ log_progress .................................... False
269
+ log_straggler ................................... False
270
+ log_throughput .................................. True
271
+ log_timers_to_tensorboard ....................... False
272
+ log_validation_ppl_to_tensorboard ............... False
273
+ log_world_size_to_tensorboard ................... False
274
+ logging_level ................................... None
275
+ loss_scale ...................................... None
276
+ loss_scale_window ............................... 1000
277
+ lr .............................................. 0.0001
278
+ lr_decay_iters .................................. 8
279
+ lr_decay_samples ................................ None
280
+ lr_decay_style .................................. cosine
281
+ lr_warmup_fraction .............................. None
282
+ lr_warmup_init .................................. 0.0
283
+ lr_warmup_iters ................................. 5
284
+ lr_warmup_samples ............................... 0
285
+ lr_wsd_decay_iters .............................. None
286
+ lr_wsd_decay_samples ............................ None
287
+ lr_wsd_decay_style .............................. exponential
288
+ main_grads_dtype ................................ torch.float32
289
+ main_params_dtype ............................... torch.float32
290
+ make_vocab_size_divisible_by .................... 128
291
+ mamba_head_dim .................................. 64
292
+ mamba_num_groups ................................ 8
293
+ mamba_num_heads ................................. None
294
+ mamba_state_dim ................................. 128
295
+ manual_gc ....................................... False
296
+ manual_gc_eval .................................. True
297
+ manual_gc_interval .............................. 0
298
+ mask_factor ..................................... 1.0
299
+ mask_prob ....................................... 0.15
300
+ mask_type ....................................... random
301
+ masked_softmax_fusion ........................... False
302
+ max_position_embeddings ......................... 32768
303
+ max_tokens_to_oom ............................... 12000
304
+ memory_snapshot_path ............................ snapshot.pickle
305
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
306
+ micro_batch_size ................................ 1
307
+ microbatch_group_size_per_vp_stage .............. None
308
+ mid_level_dataset_surplus ....................... 0.005
309
+ min_loss_scale .................................. 1.0
310
+ min_lr .......................................... 1e-05
311
+ mlp_chunks_for_prefill .......................... 1
312
+ mmap_bin_files .................................. True
313
+ mock_data ....................................... False
314
+ moe_apply_probs_on_input ........................ False
315
+ moe_aux_loss_coeff .............................. 0.01
316
+ moe_deepep_num_sms .............................. 20
317
+ moe_echo_enable_random_offloading ............... False
318
+ moe_echo_expert_dispatch_overlap ................ True
319
+ moe_echo_expert_dispatcher_type ................. hybridep
320
+ moe_echo_recompute_expert_dispatch .............. False
321
+ moe_enable_deepep ............................... False
322
+ moe_enable_echo ................................. True
323
+ moe_enable_hybridep ............................. True
324
+ moe_expert_capacity_factor ...................... None
325
+ moe_extended_tp ................................. False
326
+ moe_ffn_hidden_size ............................. 3584
327
+ moe_grouped_gemm ................................ True
328
+ moe_hybridep_num_sms ............................ 16
329
+ moe_input_jitter_eps ............................ None
330
+ moe_layer_freq .................................. 1
331
+ moe_layer_recompute ............................. False
332
+ moe_num_echo_experts ............................ 8
333
+ moe_pad_expert_input_to_capacity ................ False
334
+ moe_per_layer_logging ........................... False
335
+ moe_permute_fusion .............................. False
336
+ moe_received_token_capacity ..................... 1.08203125
337
+ moe_router_bias_update_rate ..................... 0.001
338
+ moe_router_dtype ................................ fp32
339
+ moe_router_enable_expert_bias ................... False
340
+ moe_router_force_load_balancing ................. False
341
+ moe_router_group_topk ........................... None
342
+ moe_router_load_balancing_type .................. aux_loss
343
+ moe_router_num_groups ........................... None
344
+ moe_router_padding_for_fp8 ...................... False
345
+ moe_router_pre_softmax .......................... False
346
+ moe_router_score_function ....................... softmax
347
+ moe_router_topk ................................. 8
348
+ moe_router_topk_scaling_factor .................. None
349
+ moe_shared_expert_intermediate_size ............. None
350
+ moe_shared_expert_overlap ....................... False
351
+ moe_token_dispatcher_type ....................... flex
352
+ moe_token_drop_policy ........................... probs
353
+ moe_upcycling_granularity ....................... 1
354
+ moe_use_device_initiated_grouped_gemm ........... True
355
+ moe_use_legacy_grouped_gemm ..................... False
356
+ moe_use_upcycling ............................... False
357
+ moe_z_loss_coeff ................................ None
358
+ mrope_section ................................... None
359
+ mscale .......................................... 1.0
360
+ mscale_all_dim .................................. 0.0
361
+ mtp_loss_scaling_factor ......................... 0.1
362
+ mtp_num_layers .................................. None
363
+ multi_latent_attention .......................... False
364
+ nccl_all_reduce_for_prefill ..................... False
365
+ nccl_communicator_config_path ................... None
366
+ nccl_ub ......................................... False
367
+ no_load_optim ................................... None
368
+ no_load_rng ..................................... None
369
+ no_persist_layer_norm ........................... False
370
+ no_rope_freq .................................... None
371
+ no_save_optim ................................... None
372
+ no_save_rng ..................................... None
373
+ non_persistent_ckpt_type ........................ None
374
+ non_persistent_global_ckpt_dir .................. None
375
+ non_persistent_local_ckpt_algo .................. fully_parallel
376
+ non_persistent_local_ckpt_dir ................... None
377
+ non_persistent_save_interval .................... None
378
+ norm_epsilon .................................... 1e-05
379
+ normalization ................................... RMSNorm
380
+ num_attention_heads ............................. 32
381
+ num_channels .................................... 3
382
+ num_classes ..................................... 1000
383
+ num_dataset_builder_threads ..................... 1
384
+ num_distributed_optimizer_instances ............. 1
385
+ num_experts ..................................... 32
386
+ num_layers ...................................... 16
387
+ num_layers_at_end_in_bf16 ....................... 1
388
+ num_layers_at_start_in_bf16 ..................... 1
389
+ num_layers_per_virtual_pipeline_stage ........... None
390
+ num_query_groups ................................ 8
391
+ num_virtual_stages_per_pipeline_rank ............ None
392
+ num_workers ..................................... 2
393
+ object_storage_cache_path ....................... None
394
+ one_logger_async ................................ False
395
+ one_logger_project .............................. megatron-lm
396
+ one_logger_run_name ............................. None
397
+ onnx_safe ....................................... None
398
+ openai_gelu ..................................... False
399
+ optimizer ....................................... adam
400
+ optimizer_cpu_offload ........................... False
401
+ optimizer_offload_fraction ...................... 1.0
402
+ output_bert_embeddings .......................... False
403
+ overlap_cpu_optimizer_d2h_h2d ................... False
404
+ overlap_grad_reduce ............................. True
405
+ overlap_moe_expert_parallel_comm ................ False
406
+ overlap_p2p_comm ................................ False
407
+ overlap_p2p_comm_warmup_flush ................... False
408
+ overlap_param_gather ............................ True
409
+ overlap_param_gather_with_optimizer_step ........ False
410
+ override_opt_param_scheduler .................... False
411
+ params_dtype .................................... torch.bfloat16
412
+ patch_dim ....................................... 16
413
+ per_split_data_args_path ........................ None
414
+ perform_initialization .......................... True
415
+ pin_cpu_grads ................................... True
416
+ pin_cpu_params .................................. True
417
+ pipeline_model_parallel_comm_backend ............ None
418
+ pipeline_model_parallel_layout .................. None
419
+ pipeline_model_parallel_size .................... 1
420
+ position_embedding_type ......................... rope
421
+ pretrained_checkpoint ........................... None
422
+ profile ......................................... False
423
+ profile_ranks ................................... [0]
424
+ profile_step_end ................................ 12
425
+ profile_step_start .............................. 10
426
+ q_lora_rank ..................................... None
427
+ qk_head_dim ..................................... 128
428
+ qk_l2_norm ...................................... False
429
+ qk_layernorm .................................... False
430
+ qk_pos_emb_head_dim ............................. 64
431
+ query_in_block_prob ............................. 0.1
432
+ rampup_batch_size ............................... None
433
+ rank ............................................ 0
434
+ recompute_granularity ........................... None
435
+ recompute_method ................................ None
436
+ recompute_modules ............................... None
437
+ recompute_num_layers ............................ None
438
+ record_memory_history ........................... False
439
+ relative_attention_max_distance ................. 128
440
+ relative_attention_num_buckets .................. 32
441
+ replication ..................................... False
442
+ replication_factor .............................. 2
443
+ replication_jump ................................ None
444
+ rerun_mode ...................................... validate_results
445
+ reset_attention_mask ............................ False
446
+ reset_position_ids .............................. False
447
+ result_rejected_tracker_filename ................ None
448
+ retriever_report_topk_accuracies ................ []
449
+ retriever_score_scaling ......................... False
450
+ retriever_seq_length ............................ 256
451
+ retro_add_retriever ............................. False
452
+ retro_attention_gate ............................ 1
453
+ retro_cyclic_train_iters ........................ None
454
+ retro_encoder_attention_dropout ................. 0.1
455
+ retro_encoder_hidden_dropout .................... 0.1
456
+ retro_encoder_layers ............................ 2
457
+ retro_num_neighbors ............................. 2
458
+ retro_num_retrieved_chunks ...................... 2
459
+ retro_project_dir ............................... None
460
+ retro_verify_neighbor_count ..................... True
461
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
462
+ rope_scaling_factor ............................. 8.0
463
+ rope_type ....................................... None
464
+ rotary_base ..................................... 1000000
465
+ rotary_interleaved .............................. False
466
+ rotary_percent .................................. 1.0
467
+ rotary_scaling_factor ........................... 1.0
468
+ rotary_seq_len_interpolation_factor ............. None
469
+ run_workload_inspector_server ................... False
470
+ sample_rate ..................................... 1.0
471
+ save ............................................ None
472
+ save_interval ................................... None
473
+ save_retain_interval ............................ None
474
+ scatter_gather_tensors_in_pipeline .............. True
475
+ seed ............................................ 1234
476
+ seq_length ...................................... 4096
477
+ sequence_parallel ............................... False
478
+ sft ............................................. False
479
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
480
+ sgd_momentum .................................... 0.9
481
+ short_seq_prob .................................. 0.1
482
+ skip_train ...................................... False
483
+ skipped_train_samples ........................... 0
484
+ spec ............................................ None
485
+ split ........................................... 100,0,0
486
+ squared_relu .................................... False
487
+ start_weight_decay .............................. 0.1
488
+ straggler_ctrlr_port ............................ 65535
489
+ straggler_minmax_count .......................... 1
490
+ suggested_communication_unit_size ............... None
491
+ swiglu .......................................... True
492
+ swin_backbone_type .............................. tiny
493
+ symmetric_ar_type ............................... None
494
+ te_rng_tracker .................................. True
495
+ tensor_model_parallel_size ...................... 1
496
+ tensorboard_dir ................................. None
497
+ tensorboard_log_interval ........................ 1
498
+ tensorboard_queue_size .......................... 1000
499
+ test_data_path .................................. None
500
+ test_mode ....................................... False
501
+ tiktoken_num_special_tokens ..................... 1000
502
+ tiktoken_pattern ................................ None
503
+ tiktoken_special_tokens ......................... None
504
+ timing_log_level ................................ 0
505
+ timing_log_option ............................... minmax
506
+ titles_data_path ................................ None
507
+ tokenizer_model ................................. None
508
+ tokenizer_type .................................. GPT2BPETokenizer
509
+ torch_fsdp2_reshard_after_forward ............... True
510
+ tp_comm_bootstrap_backend ....................... nccl
511
+ tp_comm_bulk_dgrad .............................. True
512
+ tp_comm_bulk_wgrad .............................. True
513
+ tp_comm_overlap ................................. False
514
+ tp_comm_overlap_ag .............................. True
515
+ tp_comm_overlap_cfg ............................. None
516
+ tp_comm_overlap_rs .............................. True
517
+ tp_comm_overlap_rs_dgrad ........................ False
518
+ tp_comm_split_ag ................................ True
519
+ tp_comm_split_rs ................................ True
520
+ train_data_path ................................. None
521
+ train_iters ..................................... 12
522
+ train_samples ................................... None
523
+ train_sync_interval ............................. None
524
+ transformer_impl ................................ transformer_engine
525
+ transformer_pipeline_model_parallel_size ........ 1
526
+ untie_embeddings_and_output_weights ............. True
527
+ use_checkpoint_args ............................. False
528
+ use_checkpoint_opt_param_scheduler .............. False
529
+ use_cpu_initialization .......................... None
530
+ use_custom_fsdp ................................. False
531
+ use_dist_ckpt ................................... True
532
+ use_dist_ckpt_deprecated ........................ False
533
+ use_distributed_optimizer ....................... True
534
+ use_flash_attn .................................. False
535
+ use_legacy_models ............................... False
536
+ use_mp_args_from_checkpoint_args ................ False
537
+ use_one_sent_docs ............................... False
538
+ use_persistent_ckpt_worker ...................... False
539
+ use_precision_aware_optimizer ................... False
540
+ use_pytorch_profiler ............................ False
541
+ use_ring_exchange_p2p ........................... False
542
+ use_rope_scaling ................................ False
543
+ use_rotary_position_embeddings .................. False
544
+ use_sharp ....................................... False
545
+ use_tokenizer_model_from_checkpoint_args ........ True
546
+ use_torch_fsdp2 ................................. False
547
+ use_torch_optimizer_for_cpu_offload ............. False
548
+ use_tp_pp_dp_mapping ............................ False
549
+ v_head_dim ...................................... 128
550
+ valid_data_path ................................. None
551
+ variable_seq_lengths ............................ False
552
+ virtual_pipeline_model_parallel_size ............ None
553
+ vision_backbone_type ............................ vit
554
+ vision_pretraining .............................. False
555
+ vision_pretraining_type ......................... classify
556
+ vocab_extra_ids ................................. 0
557
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
558
+ vocab_size ...................................... None
559
+ wandb_exp_name ..................................
560
+ wandb_project ...................................
561
+ wandb_save_dir ..................................
562
+ weight_decay .................................... 0.1
563
+ weight_decay_incr_style ......................... constant
564
+ wgrad_deferral_limit ............................ 0
565
+ world_size ...................................... 8
566
+ yaml_cfg ........................................ None
567
+ -------------------- end of arguments ---------------------
568
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2
569
+ > building GPT2BPETokenizer tokenizer ...
570
+ 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
571
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
572
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
573
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
574
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
575
+ > initializing torch distributed ...
576
+ [Gloo] Rank [Gloo] Rank 4 is connected to 07 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
577
+ 7
578
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
579
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
580
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
582
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank
583
+ 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank [Gloo] Rank 7 is connected to 67 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
586
+ 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 [Gloo] Rank [Gloo] Rank 345 is connected to is connected to is connected to 777 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 777
590
+
591
+
592
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
593
+ > initialized tensor model parallel with size 1
594
+ > initialized pipeline model parallel with size 1
595
+ > setting random seeds to 1234 ...
596
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
597
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : [Gloo] Rank 00
598
+ is connected to [Gloo] Rank 0 peer ranks. Expected number of connected peer ranks is : 00 is connected to
599
+ 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[Gloo] Rank
602
+ 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [rank6]:[W830 06:05:25.230246878 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.
604
+ [rank7]:[W830 06:05:25.230257242 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.
605
+ [rank5]:[W830 06:05:25.230336188 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.
606
+ [rank2]:[W830 06:05:25.230344507 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.
607
+ [rank1]:[W830 06:05:25.230397004 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 06:05:25.232249299 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
+ [rank4]:[W830 06:05:25.234189569 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ > compiling dataset index builder ...
611
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
612
+ make: Nothing to be done for 'default'.
613
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
614
+ >>> done with dataset index builder. Compilation time: 0.037 seconds
615
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
616
+ > compiling and loading fused kernels ...
617
+ [rank0]:[W830 06:05:25.394690893 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.
618
+ >>> done with compiling and loading fused kernels. Compilation time: 3.837 seconds
619
+ time to initialize megatron (seconds): 7.249
620
+ [after megatron is initialized] datetime: 2026-08-30 06:05:30
621
+ building GPT model ...
622
+ /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.
623
+ warnings.warn(
624
+ /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.
625
+ warnings.warn(
626
+ /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.
627
+ warnings.warn(
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3903983616
639
+ 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)
640
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18
641
+ Params for bucket 1 (206045184 elements, 206045184 padded size):
642
+ module.output_layer.weight
643
+ Params for bucket 2 (42086400 elements, 42086400 padded size):
644
+ module.decoder.final_layernorm.weight
645
+ module.decoder.layers.15.self_attention.linear_qkv.weight
646
+ module.decoder.layers.15.pre_mlp_layernorm.weight
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
+ Params for bucket 3 (42082304 elements, 42082304 padded size):
651
+ module.decoder.layers.14.self_attention.linear_qkv.weight
652
+ module.decoder.layers.14.pre_mlp_layernorm.weight
653
+ module.decoder.layers.14.mlp.router.weight
654
+ module.decoder.layers.14.self_attention.linear_proj.weight
655
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
656
+ Params for bucket 4 (42082304 elements, 42082304 padded size):
657
+ module.decoder.layers.13.pre_mlp_layernorm.weight
658
+ module.decoder.layers.13.self_attention.linear_qkv.weight
659
+ module.decoder.layers.13.mlp.router.weight
660
+ module.decoder.layers.13.self_attention.linear_proj.weight
661
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
662
+ Params for bucket 5 (42082304 elements, 42082304 padded size):
663
+ module.decoder.layers.12.pre_mlp_layernorm.weight
664
+ module.decoder.layers.12.self_attention.linear_qkv.weight
665
+ module.decoder.layers.12.mlp.router.weight
666
+ module.decoder.layers.12.self_attention.linear_proj.weight
667
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
668
+ Params for bucket 6 (42082304 elements, 42082304 padded size):
669
+ module.decoder.layers.11.pre_mlp_layernorm.weight
670
+ module.decoder.layers.11.self_attention.linear_qkv.weight
671
+ module.decoder.layers.11.mlp.router.weight
672
+ module.decoder.layers.11.self_attention.linear_proj.weight
673
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
674
+ Params for bucket 7 (42082304 elements, 42082304 padded size):
675
+ module.decoder.layers.10.pre_mlp_layernorm.weight
676
+ module.decoder.layers.10.self_attention.linear_qkv.weight
677
+ module.decoder.layers.10.mlp.router.weight
678
+ module.decoder.layers.10.self_attention.linear_proj.weight
679
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
680
+ Params for bucket 8 (42082304 elements, 42082304 padded size):
681
+ module.decoder.layers.9.mlp.router.weight
682
+ module.decoder.layers.9.pre_mlp_layernorm.weight
683
+ module.decoder.layers.9.self_attention.linear_qkv.weight
684
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
685
+ module.decoder.layers.9.self_attention.linear_proj.weight
686
+ Params for bucket 9 (42082304 elements, 42082304 padded size):
687
+ module.decoder.layers.8.self_attention.linear_proj.weight
688
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
689
+ module.decoder.layers.8.mlp.router.weight
690
+ module.decoder.layers.8.pre_mlp_layernorm.weight
691
+ module.decoder.layers.8.self_attention.linear_qkv.weight
692
+ Params for bucket 10 (42082304 elements, 42082304 padded size):
693
+ module.decoder.layers.7.self_attention.linear_qkv.weight
694
+ module.decoder.layers.7.pre_mlp_layernorm.weight
695
+ module.decoder.layers.7.mlp.router.weight
696
+ module.decoder.layers.7.self_attention.linear_proj.weight
697
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
698
+ Params for bucket 11 (42082304 elements, 42082304 padded size):
699
+ module.decoder.layers.6.mlp.router.weight
700
+ module.decoder.layers.6.pre_mlp_layernorm.weight
701
+ module.decoder.layers.6.self_attention.linear_qkv.weight
702
+ module.decoder.layers.6.self_attention.linear_proj.weight
703
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
704
+ Params for bucket 12 (42082304 elements, 42082304 padded size):
705
+ module.decoder.layers.5.self_attention.linear_qkv.weight
706
+ module.decoder.layers.5.self_attention.linear_proj.weight
707
+ module.decoder.layers.5.pre_mlp_layernorm.weight
708
+ module.decoder.layers.5.mlp.router.weight
709
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
710
+ Params for bucket 13 (42082304 elements, 42082304 padded size):
711
+ module.decoder.layers.4.mlp.router.weight
712
+ module.decoder.layers.4.self_attention.linear_proj.weight
713
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
714
+ module.decoder.layers.4.pre_mlp_layernorm.weight
715
+ module.decoder.layers.4.self_attention.linear_qkv.weight
716
+ Params for bucket 14 (42082304 elements, 42082304 padded size):
717
+ module.decoder.layers.3.self_attention.linear_qkv.weight
718
+ module.decoder.layers.3.pre_mlp_layernorm.weight
719
+ module.decoder.layers.3.mlp.router.weight
720
+ module.decoder.layers.3.self_attention.linear_proj.weight
721
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
722
+ Params for bucket 15 (42082304 elements, 42082304 padded size):
723
+ module.decoder.layers.2.mlp.router.weight
724
+ module.decoder.layers.2.self_attention.linear_proj.weight
725
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
726
+ module.decoder.layers.2.self_attention.linear_qkv.weight
727
+ module.decoder.layers.2.pre_mlp_layernorm.weight
728
+ Params for bucket 16 (42082304 elements, 42082304 padded size):
729
+ module.decoder.layers.1.pre_mlp_layernorm.weight
730
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
731
+ module.decoder.layers.1.self_attention.linear_proj.weight
732
+ module.decoder.layers.1.mlp.router.weight
733
+ module.decoder.layers.1.self_attention.linear_qkv.weight
734
+ Params for bucket 17 (42082304 elements, 42082304 padded size):
735
+ module.decoder.layers.0.mlp.router.weight
736
+ module.decoder.layers.0.self_attention.linear_proj.weight
737
+ module.decoder.layers.0.self_attention.linear_qkv.weight
738
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
739
+ module.decoder.layers.0.pre_mlp_layernorm.weight
740
+ Params for bucket 18 (206045184 elements, 206045184 padded size):
741
+ module.embedding.word_embeddings.weight
742
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56
743
+ Params for bucket 1 (44040192 elements, 44040192 padded size):
744
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
745
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
746
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
747
+ Params for bucket 2 (44040192 elements, 44040192 padded size):
748
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
750
+ Params for bucket 3 (58720256 elements, 58720256 padded size):
751
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
752
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
753
+ Params for bucket 4 (44040192 elements, 44040192 padded size):
754
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
755
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
756
+ Params for bucket 5 (44040192 elements, 44040192 padded size):
757
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
758
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
759
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
760
+ Params for bucket 6 (58720256 elements, 58720256 padded size):
761
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
762
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
763
+ Params for bucket 7 (58720256 elements, 58720256 padded size):
764
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
765
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
766
+ Params for bucket 8 (44040192 elements, 44040192 padded size):
767
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
768
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
769
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
770
+ Params for bucket 9 (44040192 elements, 44040192 padded size):
771
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
772
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
773
+ Params for bucket 10 (58720256 elements, 58720256 padded size):
774
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
775
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
776
+ Params for bucket 11 (44040192 elements, 44040192 padded size):
777
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
778
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
779
+ Params for bucket 12 (44040192 elements, 44040192 padded size):
780
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
781
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
782
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
783
+ Params for bucket 13 (58720256 elements, 58720256 padded size):
784
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
785
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
786
+ Params for bucket 14 (58720256 elements, 58720256 padded size):
787
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
788
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
789
+ Params for bucket 15 (44040192 elements, 44040192 padded size):
790
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
791
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
792
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
793
+ Params for bucket 16 (44040192 elements, 44040192 padded size):
794
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
795
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
796
+ Params for bucket 17 (58720256 elements, 58720256 padded size):
797
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
798
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
799
+ Params for bucket 18 (44040192 elements, 44040192 padded size):
800
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
801
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
802
+ Params for bucket 19 (44040192 elements, 44040192 padded size):
803
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
804
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
805
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
806
+ Params for bucket 20 (58720256 elements, 58720256 padded size):
807
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
808
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
809
+ Params for bucket 21 (58720256 elements, 58720256 padded size):
810
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
811
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
812
+ Params for bucket 22 (44040192 elements, 44040192 padded size):
813
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
814
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
815
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
816
+ Params for bucket 23 (44040192 elements, 44040192 padded size):
817
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
818
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
819
+ Params for bucket 24 (58720256 elements, 58720256 padded size):
820
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
821
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
822
+ Params for bucket 25 (44040192 elements, 44040192 padded size):
823
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
824
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
825
+ Params for bucket 26 (44040192 elements, 44040192 padded size):
826
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
827
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
828
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
829
+ Params for bucket 27 (58720256 elements, 58720256 padded size):
830
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
831
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
832
+ Params for bucket 28 (58720256 elements, 58720256 padded size):
833
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
834
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
835
+ Params for bucket 29 (44040192 elements, 44040192 padded size):
836
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
837
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
838
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
839
+ Params for bucket 30 (44040192 elements, 44040192 padded size):
840
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
841
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
842
+ Params for bucket 31 (58720256 elements, 58720256 padded size):
843
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
844
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
845
+ Params for bucket 32 (44040192 elements, 44040192 padded size):
846
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
847
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
848
+ Params for bucket 33 (44040192 elements, 44040192 padded size):
849
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
850
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
851
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
852
+ Params for bucket 34 (58720256 elements, 58720256 padded size):
853
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
854
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
855
+ Params for bucket 35 (58720256 elements, 58720256 padded size):
856
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
857
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
858
+ Params for bucket 36 (44040192 elements, 44040192 padded size):
859
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
860
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
861
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
862
+ Params for bucket 37 (44040192 elements, 44040192 padded size):
863
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
864
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
865
+ Params for bucket 38 (58720256 elements, 58720256 padded size):
866
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
867
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
868
+ Params for bucket 39 (44040192 elements, 44040192 padded size):
869
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
870
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
871
+ Params for bucket 40 (44040192 elements, 44040192 padded size):
872
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
873
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
874
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
875
+ Params for bucket 41 (58720256 elements, 58720256 padded size):
876
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
877
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
878
+ Params for bucket 42 (58720256 elements, 58720256 padded size):
879
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
880
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
881
+ Params for bucket 43 (44040192 elements, 44040192 padded size):
882
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
883
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
884
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
885
+ Params for bucket 44 (44040192 elements, 44040192 padded size):
886
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
887
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
888
+ Params for bucket 45 (58720256 elements, 58720256 padded size):
889
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
890
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
891
+ Params for bucket 46 (44040192 elements, 44040192 padded size):
892
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
893
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
894
+ Params for bucket 47 (44040192 elements, 44040192 padded size):
895
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
896
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
897
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
898
+ Params for bucket 48 (58720256 elements, 58720256 padded size):
899
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
900
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
901
+ Params for bucket 49 (58720256 elements, 58720256 padded size):
902
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
903
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
904
+ Params for bucket 50 (44040192 elements, 44040192 padded size):
905
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
906
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
907
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
908
+ Params for bucket 51 (44040192 elements, 44040192 padded size):
909
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
910
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
911
+ Params for bucket 52 (58720256 elements, 58720256 padded size):
912
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
913
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
914
+ Params for bucket 53 (44040192 elements, 44040192 padded size):
915
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
916
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
917
+ Params for bucket 54 (44040192 elements, 44040192 padded size):
918
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
919
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
920
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
921
+ Params for bucket 55 (58720256 elements, 58720256 padded size):
922
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
923
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
924
+ Params for bucket 56 (58720256 elements, 58720256 padded size):
925
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
926
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
927
+ 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 0x76148178c410>, config_logger_dir='')
928
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
929
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 06:05:31
930
+ > building train, validation, and test datasets ...
931
+ > datasets target sizes (minimum size):
932
+ train: 192
933
+ validation: 0
934
+ test: 0
935
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
936
+ > building train, validation, and test datasets for GPT ...
937
+ 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 0x7614820e52b0>, 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)
938
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
939
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
940
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
941
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
942
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
943
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
944
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
945
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy
946
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy
947
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy
948
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
949
+ > finished creating GPT datasets ...
950
+ [after dataloaders are built] datetime: 2026-08-30 06:05:31
951
+ done with setup ...
952
+ (min, max) time across ranks (ms):
953
+ model-and-optimizer-setup ......................: (316.92, 320.58)
954
+ train/valid/test-data-iterators-setup ..........: (63.58, 75.23)
955
+ training ...
956
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
957
+ [before the start of training step] datetime: 2026-08-30 06:05:31
958
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
959
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
960
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
961
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
962
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
963
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
964
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
965
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
966
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
967
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
968
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
969
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
970
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
971
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
972
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
973
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
974
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
975
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
976
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
977
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
978
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
979
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
980
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
981
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
982
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
983
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
984
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
985
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
986
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
987
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
988
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
989
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
990
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
991
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
992
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
993
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
994
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
995
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
996
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
997
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
998
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
999
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1000
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1001
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1002
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1003
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1004
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1005
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1006
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1007
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1008
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1009
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1010
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1011
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1012
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1013
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1014
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1015
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1016
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1017
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1018
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1019
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1020
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1021
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1022
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1023
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1024
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1025
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1026
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1027
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1028
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1029
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1030
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1031
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1032
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1033
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1034
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1035
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1036
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1037
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1038
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1039
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1040
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1041
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1042
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1043
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1044
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1045
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1046
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1047
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1048
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1049
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1050
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1051
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1052
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1053
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1054
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1055
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1056
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1057
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1058
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1059
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1060
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1061
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1062
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1063
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1064
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1065
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1066
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1067
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1068
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1069
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1070
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1071
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1072
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1073
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1074
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1075
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1076
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1077
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1078
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1079
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1080
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1081
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1082
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1083
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1084
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1085
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1086
+ [2026-08-30 06:06:09] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38419.5 | throughput per GPU (TFLOP/s/GPU): 8.7 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103012E+01 | loss scale: 1.0 | grad norm: 44.244 | number of skipped iterations: 0 | number of nan iterations: 0 |
1087
+ Number of parameters in transformer block in billions: 23.22
1088
+ Number of parameters in embedding layers in billions: 0.41
1089
+ Total number of parameters in billions: 23.63
1090
+ Number of parameters in most loaded shard in billions: 23.6320
1091
+ Theoretical memory footprints: weight and optimizer=169029.43 MB
1092
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 59522.912109375 | max allocated: 68963.6640625 | reserved: 73392.0 | max reserved: 73392.0
1093
+ [2026-08-30 06:06:10] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 638.2 | throughput per GPU (TFLOP/s/GPU): 522.4 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101632E+01 | loss scale: 1.0 | grad norm: 45.760 | number of skipped iterations: 0 | number of nan iterations: 0 |
1094
+ [2026-08-30 06:06:10] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 483.5 | throughput per GPU (TFLOP/s/GPU): 689.6 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.022623E+01 | loss scale: 1.0 | grad norm: 192.206 | number of skipped iterations: 0 | number of nan iterations: 0 |
1095
+ Capture CUDA graph for training!!!
1096
+ CUDA graph capture done!!!
1097
+ [2026-08-30 06:06:13] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2111.4 | throughput per GPU (TFLOP/s/GPU): 157.9 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.325131E+01 | loss scale: 1.0 | grad norm: 30.070 | number of skipped iterations: 0 | number of nan iterations: 0 |
1098
+ [2026-08-30 06:06:13] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 393.8 | throughput per GPU (TFLOP/s/GPU): 846.5 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.038896E+01 | loss scale: 1.0 | grad norm: 16.544 | number of skipped iterations: 0 | number of nan iterations: 0 |
1099
+ [2026-08-30 06:06:13] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 393.3 | throughput per GPU (TFLOP/s/GPU): 847.6 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.049941E+01 | loss scale: 1.0 | grad norm: 41.215 | number of skipped iterations: 0 | number of nan iterations: 0 |
1100
+ [2026-08-30 06:06:14] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 397.0 | throughput per GPU (TFLOP/s/GPU): 839.8 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.011006E+01 | loss scale: 1.0 | grad norm: 8.120 | number of skipped iterations: 0 | number of nan iterations: 0 |
1101
+ [2026-08-30 06:06:14] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 393.7 | throughput per GPU (TFLOP/s/GPU): 846.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.570474E+00 | loss scale: 1.0 | grad norm: 5.905 | number of skipped iterations: 0 | number of nan iterations: 0 |
1102
+ [2026-08-30 06:06:15] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 395.6 | throughput per GPU (TFLOP/s/GPU): 842.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.313912E+00 | loss scale: 1.0 | grad norm: 5.773 | number of skipped iterations: 0 | number of nan iterations: 0 |
1103
+ [2026-08-30 06:06:15] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 397.0 | throughput per GPU (TFLOP/s/GPU): 839.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.136040E+00 | loss scale: 1.0 | grad norm: 5.564 | number of skipped iterations: 0 | number of nan iterations: 0 |
1104
+ [2026-08-30 06:06:15] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 392.3 | throughput per GPU (TFLOP/s/GPU): 849.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.999651E+00 | loss scale: 1.0 | grad norm: 5.249 | number of skipped iterations: 0 | number of nan iterations: 0 |
1105
+ [2026-08-30 06:06:16] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 393.6 | throughput per GPU (TFLOP/s/GPU): 847.1 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.877594E+00 | loss scale: 1.0 | grad norm: 5.007 | number of skipped iterations: 0 | number of nan iterations: 0 |
1106
+ [after training is done] datetime: 2026-08-30 06:06:16
1107
+ [rank0]:[W830 06:06:17.710658132 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())
tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/results.tsv CHANGED
@@ -2,3 +2,6 @@ config backend policy slots seq gbs cap recompute rep status iter_ms_mean iter_m
2
  G1 echo one 1 4096 16 2.40625 none 1 OK 467.30 463.50 161947 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/rep1
3
  G1 echo one 1 4096 16 2.40625 none 2 OK 466.99 460.60 161723 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/rep2
4
  G1 echo one 1 4096 16 2.40625 none 3 OK 466.74 460.50 161749 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/rep3
 
 
 
 
2
  G1 echo one 1 4096 16 2.40625 none 1 OK 467.30 463.50 161947 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/rep1
3
  G1 echo one 1 4096 16 2.40625 none 2 OK 466.99 460.60 161723 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/rep2
4
  G1 echo one 1 4096 16 2.40625 none 3 OK 466.74 460.50 161749 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G1__echo__one__seq4096__gbs16__cap2p40625/rep3
5
+ G2 echo one 1 4096 16 1.08203125 none 1 OK 393.91 392.80 120227 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep1
6
+ G2 echo one 1 4096 16 1.08203125 none 2 OK 394.41 392.80 120225 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2
7
+ G2 echo one 1 4096 16 1.08203125 none 3 OK 394.49 392.30 120337 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3
tiera-rerun-fixed-moonep/hf_watcher.log CHANGED
@@ -15,3 +15,6 @@ Watching /workspace/tiera-rerun-fixed-moonep -> HF every 600s (kill to stop)
15
  uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
16
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
17
  [2026-08-30T05:59:19+00:00] backup pass
 
 
 
 
15
  uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
16
  https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
17
  [2026-08-30T05:59:19+00:00] backup pass
18
+ uploaded: /workspace/tiera-rerun-fixed-moonep -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tiera-rerun-fixed-moonep
19
+ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
20
+ [2026-08-30T06:09:22+00:00] backup pass
tiera-rerun-fixed-moonep/study.log CHANGED
@@ -410,3 +410,18 @@ driver_repo: /workspace/blackwell_run_scripts @ 9925dcc
410
  [guard] tearing down process group (outcome=DONE)
411
  [05:53:17] tracing G1__echo__one__seq4096__gbs16__cap2p40625 (ranks: 0 1 2 3 4 5 6 7)
412
  [guard] iter 8/10 (80s)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
410
  [guard] tearing down process group (outcome=DONE)
411
  [05:53:17] tracing G1__echo__one__seq4096__gbs16__cap2p40625 (ranks: 0 1 2 3 4 5 6 7)
412
  [guard] iter 8/10 (80s)
413
+ [guard] no progress for 420s at iter 8/10; treating as done
414
+ [guard] nsys stop (flushing report)
415
+ [guard] tearing down process group (outcome=STALLED)
416
+ [06:02:23] trace -> G1__echo__one__seq4096__gbs16__cap2p40625.nsys-rep (14M)
417
+ --------------------------------------------------------------------------
418
+ [06:02:23] CELL G2__echo__one__seq4096__gbs16__cap1p08203125 (slots=1)
419
+ [guard] iter 11/12 (70s)
420
+ [guard] iter 12/12 (80s)
421
+ [guard] iter 12/12 (70s)
422
+ [guard] tearing down process group (outcome=DONE)
423
+ [06:04:55] waiting for stale ranks to exit ...
424
+ [guard] iter 11/12 (70s)
425
+ [guard] iter 12/12 (80s)
426
+ [06:06:26] tracing G2__echo__one__seq4096__gbs16__cap1p08203125 (ranks: 0 1 2 3 4 5 6 7)
427
+ [guard] iter 8/10 (80s)