AbhishekGhosh1998 commited on
Commit
a19ea07
·
verified ·
1 Parent(s): b9028f6

granularity results granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace @ 2026-09-21T09:05:09Z from sp-aiinfra-interactive-aghosh-8-mb99k-8xb200-moe-cg-0

Browse files
.gitattributes CHANGED
@@ -116,3 +116,4 @@ granularity_next_b200/tierb_models/traces/KIMI_K2_L2/echo/tierbtrace/tierbtrace_
116
  granularity_next_b200/tierb_models/traces/KIMI_K2_L2/ultraep/tierbtrace/tierbtrace_gbs8/KIMI_K2_L2__ultraep__matched__seq4096__gbs8__cap1p17578125/nsys/KIMI_K2_L2__ultraep__matched__seq4096__gbs8__cap1p17578125.nsys-rep filter=lfs diff=lfs merge=lfs -text
117
  granularity_next_b200/tierb_models/traces/KIMI_K2_L2/moonep/tierbtrace/tierbtrace_gbs8/KIMI_K2_L2__moonep__matched__seq4096__gbs8__capnone__async/nsys/KIMI_K2_L2__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
118
  granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/echo/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__echo__matched__seq4096__gbs8__cap1p08984375/nsys/OLMOE_1B_7B_L16__echo__matched__seq4096__gbs8__cap1p08984375.nsys-rep filter=lfs diff=lfs merge=lfs -text
 
 
116
  granularity_next_b200/tierb_models/traces/KIMI_K2_L2/ultraep/tierbtrace/tierbtrace_gbs8/KIMI_K2_L2__ultraep__matched__seq4096__gbs8__cap1p17578125/nsys/KIMI_K2_L2__ultraep__matched__seq4096__gbs8__cap1p17578125.nsys-rep filter=lfs diff=lfs merge=lfs -text
117
  granularity_next_b200/tierb_models/traces/KIMI_K2_L2/moonep/tierbtrace/tierbtrace_gbs8/KIMI_K2_L2__moonep__matched__seq4096__gbs8__capnone__async/nsys/KIMI_K2_L2__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
118
  granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/echo/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__echo__matched__seq4096__gbs8__cap1p08984375/nsys/OLMOE_1B_7B_L16__echo__matched__seq4096__gbs8__cap1p08984375.nsys-rep filter=lfs diff=lfs merge=lfs -text
119
+ granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375.nsys-rep filter=lfs diff=lfs merge=lfs -text
granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375.nsys-rep ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:582723e90e937e49dfa03fcddb80b083a17d95f7d18a576cb85e3833befea171
3
+ size 7612671
granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/run.env ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=ultraep
2
+ config=OLMOE_1B_7B_L16
3
+ family=tierB
4
+ source_model=OLMoE-1B-7B-0924
5
+ standard_num_layers=16
6
+ base_rung=
7
+ repo=/workspace/Megatron-UltraEP
8
+ branch=ultraep-bracing-regions-through-kernels
9
+ commit=03d57ef7
10
+ num_layers=16
11
+ hidden=2048
12
+ ffn_hidden=1024
13
+ moe_ffn_hidden=1024
14
+ num_experts=64
15
+ topk=8
16
+ experts_per_rank=8
17
+ expert_params=6291456
18
+ heads=16
19
+ kv_groups=16
20
+ kv_channels=128
21
+ qk_layernorm=0
22
+ shared_ffn=0
23
+ shared_expert_overlap=0
24
+ moe_token_dispatcher=flex
25
+ mla=0
26
+ q_lora_rank=
27
+ kv_lora_rank=512
28
+ qk_head_dim=128
29
+ qk_pos_emb_head_dim=64
30
+ v_head_dim=128
31
+ dense_first_layers=0
32
+ moe_layers=16
33
+ rotary_base=10000
34
+ rotary_percent=1.0
35
+ rotary_scaling=
36
+ mscale=
37
+ mscale_all_dim=
38
+ qkv_bias=0
39
+ router_balancing=aux_loss
40
+ aux_loss_coeff=1e-2
41
+ router_pre_softmax=1
42
+ router_fusion=0
43
+ router_dtype=fp32
44
+ router_groups=
45
+ router_group_topk=
46
+ router_score_function=softmax
47
+ router_expert_bias=0
48
+ router_bias_update_rate=
49
+ router_topk_scaling_factor=
50
+ shared_expert_gate=0
51
+ apply_layernorm_1p=0
52
+ attention_output_gate=0
53
+ linear_attention_type=
54
+ linear_attention_freq=
55
+ linear_conv_kernel_dim=
56
+ linear_key_head_dim=
57
+ linear_value_head_dim=
58
+ linear_num_key_heads=
59
+ linear_num_value_heads=
60
+ mtp_num_layers=0
61
+ mtp_loss_scaling_factor=0.1
62
+ no_weight_decay_cond_type=
63
+ max_position_embeddings=4096
64
+ init_method_std=0.02
65
+ device_initiated_grouped_gemm=1
66
+ norm_epsilon=1e-5
67
+ seq_length=4096
68
+ micro_batch_size=1
69
+ global_batch_size=8
70
+ num_microbatches=1
71
+ assignments_per_rank=32768
72
+ mean_tokens_per_expert=4096
73
+ slot_policy=matched
74
+ slots_per_rank=8
75
+ capacity_mode=static
76
+ capacity=1.08984375
77
+ cuda_graph=1
78
+ probe_mode=0
79
+ async_grad_reduce=0
80
+ branch_role=timing
81
+ memory_dump_mode=none
82
+ memory_snapshot_path=
83
+ cuda_graph_warmup_steps=3
84
+ recompute=none
85
+ force_balance=0
86
+ train_iters=10
87
+ ep_size=8
88
+ echo_dispatch_overlap=1
89
+ echo_dispatch_recompute=0
90
+ ultraep_quota_min_tokens=512
91
+ ultraep_weight_sync_join=
92
+ profile_enable=1
93
+ kernel_trace=1
94
+ fp8_mode=mxfp8
95
+ fp8_recipe=mxfp8
96
+ training_seed=1234
97
+ learning_rate=1e-4
98
+ min_learning_rate=1.0e-5
99
+ lr_warmup_iters=5
100
+ state_dump_enable=0
101
+ state_dump_dir=
102
+ moonep_assert_no_pending_reduce=0
granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/train.log ADDED
@@ -0,0 +1,1276 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ OLMOE_1B_7B_L16 / ultraep (slots/rank=8 policy=matched)
3
+ E=64 K=8 H'=1024 m=8
4
+ L=16 seq=4096 gbs=8 (1 microbatches)
5
+ cap_mode=static cap=1.08984375 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-UltraEP @ 03d57ef7
8
+ ==============================================================
9
+ W0921 09:02:53.628000 787046 torch/distributed/run.py:766]
10
+ W0921 09:02:53.628000 787046 torch/distributed/run.py:766] *****************************************
11
+ W0921 09:02:53.628000 787046 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
+ W0921 09:02:53.628000 787046 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
16
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
17
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
18
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
19
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
20
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
21
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
22
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
23
+ 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
24
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
25
+ Number of virtual stages per pipeline stage: None
26
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
27
+ using torch.bfloat16 for parameters ...
28
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
29
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
30
+ ------------------------ arguments ------------------------
31
+ account_for_embedding_in_pipeline_split ......... False
32
+ account_for_loss_in_pipeline_split .............. False
33
+ accumulate_allreduce_grads_in_fp32 .............. True
34
+ adam_beta1 ...................................... 0.9
35
+ adam_beta2 ...................................... 0.999
36
+ adam_eps ........................................ 1e-08
37
+ add_bias_linear ................................. False
38
+ add_position_embedding .......................... False
39
+ add_qkv_bias .................................... False
40
+ adlr_autoresume ................................. False
41
+ adlr_autoresume_interval ........................ 1000
42
+ align_grad_reduce ............................... True
43
+ align_param_gather .............................. False
44
+ app_tag_run_name ................................ None
45
+ app_tag_run_version ............................. 0.0.0
46
+ apply_layernorm_1p .............................. False
47
+ apply_query_key_layer_scaling ................... False
48
+ apply_residual_connection_post_layernorm ........ False
49
+ apply_rope_fusion ............................... True
50
+ async_save ...................................... None
51
+ async_tensor_model_parallel_allreduce ........... True
52
+ attention_backend ............................... AttnBackend.auto
53
+ attention_dropout ............................... 0.0
54
+ attention_softmax_in_fp32 ....................... False
55
+ auto_detect_ckpt_format ......................... False
56
+ barrier_with_L1_time ............................ True
57
+ bert_binary_head ................................ True
58
+ bert_embedder_type .............................. megatron
59
+ bert_load ....................................... None
60
+ bf16 ............................................ True
61
+ bias_dropout_fusion ............................. True
62
+ bias_gelu_fusion ................................ False
63
+ bias_swiglu_fusion .............................. True
64
+ biencoder_projection_dim ........................ 0
65
+ biencoder_shared_query_context_model ............ False
66
+ block_data_path ................................. None
67
+ calc_ft_timeouts ................................ False
68
+ calculate_per_token_loss ........................ False
69
+ check_for_large_grads ........................... False
70
+ check_for_nan_in_loss_and_grad .................. False
71
+ check_for_spiky_loss ............................ False
72
+ check_weight_hash_across_dp_replicas_interval ... None
73
+ ckpt_assume_constant_structure .................. False
74
+ ckpt_convert_format ............................. None
75
+ ckpt_convert_save ............................... None
76
+ ckpt_convert_update_legacy_dist_opt_format ...... False
77
+ ckpt_format ..................................... torch_dist
78
+ ckpt_fully_parallel_load ........................ False
79
+ ckpt_fully_parallel_save ........................ True
80
+ ckpt_fully_parallel_save_deprecated ............. False
81
+ ckpt_step ....................................... None
82
+ classes_fraction ................................ 1.0
83
+ clip_grad ....................................... 1.0
84
+ clone_scatter_output_in_embedding ............... True
85
+ config_logger_dir ...............................
86
+ consumed_train_samples .......................... 0
87
+ consumed_valid_samples .......................... 0
88
+ context_parallel_size ........................... 1
89
+ cp_comm_type .................................... ['p2p']
90
+ create_attention_mask_in_dataloader ............. True
91
+ cross_entropy_fusion_impl ....................... native
92
+ cross_entropy_loss_fusion ....................... False
93
+ cuda_graph_scope ................................ full_iteration
94
+ cuda_graph_warmup_steps ......................... 3
95
+ data_args_path .................................. None
96
+ data_cache_path ................................. None
97
+ data_parallel_random_init ....................... False
98
+ data_parallel_sharding_strategy ................. no_shard
99
+ data_parallel_size .............................. 8
100
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
101
+ data_per_class_fraction ......................... 1.0
102
+ data_sharding ................................... True
103
+ dataloader_type ................................. single
104
+ ddp_average_in_collective ....................... False
105
+ ddp_bucket_size ................................. None
106
+ ddp_num_buckets ................................. None
107
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
108
+ decoder_first_pipeline_num_layers ............... None
109
+ decoder_last_pipeline_num_layers ................ None
110
+ decoder_num_layers .............................. None
111
+ decoder_seq_length .............................. None
112
+ decoupled_lr .................................... None
113
+ decoupled_min_lr ................................ None
114
+ decrease_batch_size_if_needed ................... False
115
+ defer_embedding_wgrad_compute ................... False
116
+ delay_wgrad_compute ............................. False
117
+ deprecated_use_mcore_models ..................... True
118
+ deterministic_mode .............................. False
119
+ dino_bottleneck_size ............................ 256
120
+ dino_freeze_last_layer .......................... 1
121
+ dino_head_hidden_size ........................... 2048
122
+ dino_local_crops_number ......................... 10
123
+ dino_local_img_size ............................. 96
124
+ dino_norm_last_layer ............................ False
125
+ dino_teacher_temp ............................... 0.07
126
+ dino_warmup_teacher_temp ........................ 0.04
127
+ dino_warmup_teacher_temp_epochs ................. 30
128
+ disable_bf16_reduced_precision_matmul ........... False
129
+ disable_mamba_mem_eff_path ...................... False
130
+ disable_straggler_on_startup .................... False
131
+ dist_ckpt_format_deprecated ..................... None
132
+ dist_ckpt_strictness ............................ assume_ok_unexpected
133
+ distribute_saved_activations .................... False
134
+ distributed_backend ............................. nccl
135
+ distributed_timeout_minutes ..................... 10
136
+ embedding_init_method_std ....................... None
137
+ embedding_path .................................. None
138
+ empty_unused_memory_level ....................... 0
139
+ enable_cuda_graph ............................... True
140
+ enable_experimental ............................. False
141
+ enable_ft_package ............................... False
142
+ enable_gloo_process_groups ...................... True
143
+ enable_msc ...................................... True
144
+ enable_one_logger ............................... True
145
+ encoder_num_layers .............................. 16
146
+ encoder_seq_length .............................. 4096
147
+ end_weight_decay ................................ 0.1
148
+ eod_mask_loss ................................... False
149
+ error_injection_rate ............................ 0
150
+ error_injection_type ............................ transient_error
151
+ eval_interval ................................... 1000000
152
+ eval_iters ...................................... 0
153
+ evidence_data_path .............................. None
154
+ exit_duration_in_mins ........................... None
155
+ exit_interval ................................... None
156
+ exit_on_missing_checkpoint ...................... False
157
+ exit_signal_handler ............................. False
158
+ exp_avg_dtype ................................... torch.float32
159
+ exp_avg_sq_dtype ................................ torch.float32
160
+ expert_model_parallel_size ...................... 8
161
+ expert_tensor_parallel_size ..................... 1
162
+ export_force_local_attention .................... False
163
+ export_kd_cfg ................................... None
164
+ export_kd_teacher_ckpt_format ................... None
165
+ export_kd_teacher_load .......................... None
166
+ export_kv_cache_quant ........................... False
167
+ export_legacy_megatron .......................... False
168
+ export_model_type ............................... GPTModel
169
+ export_moe_apply_probs_on_input ................. False
170
+ export_qk_l2_norm ............................... False
171
+ export_quant_cfg ................................ None
172
+ export_real_quant_cfg ........................... None
173
+ export_te_mcore_model ........................... False
174
+ external_cuda_graph ............................. False
175
+ ffn_hidden_size ................................. 1024
176
+ finetune ........................................ False
177
+ finetune_data_split ............................. train
178
+ finetune_hf_dataset ............................. None
179
+ first_last_layers_bf16 .......................... False
180
+ flash_decode .................................... False
181
+ fp16 ............................................ False
182
+ fp16_lm_cross_entropy ........................... False
183
+ fp32_residual_connection ........................ False
184
+ fp8 ............................................. e4m3
185
+ fp8_amax_compute_algo ........................... most_recent
186
+ fp8_amax_history_len ............................ 1
187
+ fp8_interval .................................... 1
188
+ fp8_margin ...................................... 0
189
+ fp8_param_gather ................................ False
190
+ fp8_recipe ...................................... mxfp8
191
+ fp8_wgrad ....................................... True
192
+ fsdp_double_buffer .............................. False
193
+ global_batch_size ............................... 8
194
+ grad_reduce_in_bf16 ............................. False
195
+ gradient_accumulation_fusion .................... True
196
+ gradient_reduce_div_fusion ...................... True
197
+ group_query_attention ........................... True
198
+ head_lr_mult .................................... 1.0
199
+ heterogeneous_layers_config_encoded_json ........ None
200
+ heterogeneous_layers_config_path ................ None
201
+ hidden_dropout .................................. 0.0
202
+ hidden_size ..................................... 2048
203
+ hierarchical_context_parallel_sizes ............. None
204
+ high_priority_stream_groups ..................... []
205
+ hybrid_attention_ratio .......................... 0.0
206
+ hybrid_mlp_ratio ................................ 0.0
207
+ hybrid_override_pattern ......................... None
208
+ hysteresis ...................................... 2
209
+ ict_head_size ................................... None
210
+ ict_load ........................................ None
211
+ img_h ........................................... 224
212
+ img_w ........................................... 224
213
+ indexer_batch_size .............................. 128
214
+ indexer_log_interval ............................ 1000
215
+ inference_batch_times_seqlen_threshold .......... -1
216
+ inference_dynamic_batching ...................... False
217
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
218
+ inference_dynamic_batching_buffer_overflow_factor None
219
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
220
+ inference_dynamic_batching_chunk_size ........... 256
221
+ inference_dynamic_batching_max_requests_override None
222
+ inference_dynamic_batching_max_tokens_override .. None
223
+ inference_dynamic_batching_num_cuda_graphs ...... 16
224
+ inference_max_batch_size ........................ 8
225
+ inference_max_seq_length ........................ 2560
226
+ inference_rng_tracker ........................... False
227
+ init_method_std ................................. 0.02
228
+ init_method_xavier_uniform ...................... False
229
+ init_model_with_meta_device ..................... False
230
+ initial_loss_scale .............................. 4294967296
231
+ inprocess_active_world_size ..................... 8
232
+ inprocess_barrier_timeout ....................... 120
233
+ inprocess_completion_timeout .................... 120
234
+ inprocess_empty_cuda_cache ...................... False
235
+ inprocess_granularity ........................... node
236
+ inprocess_hard_timeout .......................... 90
237
+ inprocess_heartbeat_interval .................... 30
238
+ inprocess_heartbeat_timeout ..................... 60
239
+ inprocess_last_call_wait ........................ 1
240
+ inprocess_max_iterations ........................ None
241
+ inprocess_monitor_process_interval .............. 1.0
242
+ inprocess_monitor_thread_interval ............... 1.0
243
+ inprocess_progress_watchdog_interval ............ 1.0
244
+ inprocess_restart ............................... False
245
+ inprocess_soft_timeout .......................... 60
246
+ inprocess_termination_grace_time ................ 1
247
+ is_hybrid_model ................................. False
248
+ iter_per_epoch .................................. 1250
249
+ iterations_to_skip .............................. []
250
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
251
+ kitchen_config_file ............................. None
252
+ kitchen_recipe_number ........................... None
253
+ kv_channels ..................................... 128
254
+ kv_lora_rank .................................... 32
255
+ lazy_mpu_init ................................... None
256
+ load ............................................ None
257
+ load_main_params_from_ckpt ...................... None
258
+ load_model_opt_format ........................... False
259
+ local_rank ...................................... 0
260
+ log_energy ...................................... False
261
+ log_interval .................................... 1
262
+ log_loss_scale_to_tensorboard ................... True
263
+ log_memory_to_tensorboard ....................... False
264
+ log_num_zeros_in_grad ........................... False
265
+ log_params_norm ................................. False
266
+ log_progress .................................... False
267
+ log_straggler ................................... False
268
+ log_throughput .................................. True
269
+ log_timers_to_tensorboard ....................... False
270
+ log_validation_ppl_to_tensorboard ............... False
271
+ log_world_size_to_tensorboard ................... False
272
+ logging_level ................................... None
273
+ loss_scale ...................................... None
274
+ loss_scale_window ............................... 1000
275
+ lr .............................................. 0.0001
276
+ lr_decay_iters .................................. 8
277
+ lr_decay_samples ................................ None
278
+ lr_decay_style .................................. cosine
279
+ lr_warmup_fraction .............................. None
280
+ lr_warmup_init .................................. 0.0
281
+ lr_warmup_iters ................................. 5
282
+ lr_warmup_samples ............................... 0
283
+ lr_wsd_decay_iters .............................. None
284
+ lr_wsd_decay_samples ............................ None
285
+ lr_wsd_decay_style .............................. exponential
286
+ main_grads_dtype ................................ torch.float32
287
+ main_params_dtype ............................... torch.float32
288
+ make_vocab_size_divisible_by .................... 128
289
+ mamba_head_dim .................................. 64
290
+ mamba_num_groups ................................ 8
291
+ mamba_num_heads ................................. None
292
+ mamba_state_dim ................................. 128
293
+ manual_gc ....................................... False
294
+ manual_gc_eval .................................. True
295
+ manual_gc_interval .............................. 0
296
+ mask_factor ..................................... 1.0
297
+ mask_prob ....................................... 0.15
298
+ mask_type ....................................... random
299
+ masked_softmax_fusion ........................... False
300
+ max_position_embeddings ......................... 4096
301
+ max_tokens_to_oom ............................... 12000
302
+ memory_snapshot_path ............................ snapshot.pickle
303
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
304
+ micro_batch_size ................................ 1
305
+ microbatch_group_size_per_vp_stage .............. None
306
+ mid_level_dataset_surplus ....................... 0.005
307
+ min_loss_scale .................................. 1.0
308
+ min_lr .......................................... 1e-05
309
+ mlp_chunks_for_prefill .......................... 1
310
+ mmap_bin_files .................................. True
311
+ mock_data ....................................... False
312
+ moe_apply_probs_on_input ........................ False
313
+ moe_aux_loss_coeff .............................. 0.01
314
+ moe_capacity_monitor ............................ True
315
+ moe_deepep_num_sms .............................. 20
316
+ moe_echo_enable_random_offloading ............... False
317
+ moe_echo_expert_dispatch_overlap ................ False
318
+ moe_echo_expert_dispatcher_type ................. hybridep
319
+ moe_echo_recompute_expert_dispatch .............. False
320
+ moe_enable_deepep ............................... False
321
+ moe_enable_echo ................................. False
322
+ moe_enable_hybridep ............................. True
323
+ moe_enable_ultraep .............................. True
324
+ moe_expert_capacity_factor ...................... None
325
+ moe_extended_tp ................................. False
326
+ moe_ffn_hidden_size ............................. 1024
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 ............................ None
333
+ moe_num_redundant_experts_per_rank .............. 8
334
+ moe_pad_expert_input_to_capacity ................ False
335
+ moe_per_layer_logging ........................... False
336
+ moe_permute_fusion .............................. False
337
+ moe_received_token_capacity ..................... 1.08984375
338
+ moe_router_bias_update_rate ..................... 0.001
339
+ moe_router_dtype ................................ fp32
340
+ moe_router_enable_expert_bias ................... False
341
+ moe_router_force_load_balancing ................. False
342
+ moe_router_group_topk ........................... None
343
+ moe_router_load_balancing_type .................. aux_loss
344
+ moe_router_num_groups ........................... None
345
+ moe_router_padding_for_fp8 ...................... False
346
+ moe_router_pre_softmax .......................... True
347
+ moe_router_score_function ....................... softmax
348
+ moe_router_topk ................................. 8
349
+ moe_router_topk_scaling_factor .................. None
350
+ moe_shared_expert_intermediate_size ............. None
351
+ moe_shared_expert_overlap ....................... False
352
+ moe_token_dispatcher_type ....................... flex
353
+ moe_token_drop_policy ........................... probs
354
+ moe_ultraep_max_microbatches .................... None
355
+ moe_ultraep_overlap_weight_sync ................. True
356
+ moe_ultraep_weight_sync_join .................... before_dispatch
357
+ moe_upcycling_granularity ....................... 1
358
+ moe_use_device_initiated_grouped_gemm ........... True
359
+ moe_use_legacy_grouped_gemm ..................... False
360
+ moe_use_upcycling ............................... False
361
+ moe_z_loss_coeff ................................ None
362
+ mrope_section ................................... None
363
+ mscale .......................................... 1.0
364
+ mscale_all_dim .................................. 0.0
365
+ mtp_loss_scaling_factor ......................... 0.1
366
+ mtp_num_layers .................................. None
367
+ multi_latent_attention .......................... False
368
+ nccl_all_reduce_for_prefill ..................... False
369
+ nccl_communicator_config_path ................... None
370
+ nccl_ub ......................................... False
371
+ no_load_optim ................................... None
372
+ no_load_rng ..................................... None
373
+ no_persist_layer_norm ........................... False
374
+ no_rope_freq .................................... None
375
+ no_save_optim ................................... None
376
+ no_save_rng ..................................... None
377
+ non_persistent_ckpt_type ........................ None
378
+ non_persistent_global_ckpt_dir .................. None
379
+ non_persistent_local_ckpt_algo .................. fully_parallel
380
+ non_persistent_local_ckpt_dir ................... None
381
+ non_persistent_save_interval .................... None
382
+ norm_epsilon .................................... 1e-05
383
+ normalization ................................... RMSNorm
384
+ num_attention_heads ............................. 16
385
+ num_channels .................................... 3
386
+ num_classes ..................................... 1000
387
+ num_dataset_builder_threads ..................... 1
388
+ num_distributed_optimizer_instances ............. 1
389
+ num_experts ..................................... 64
390
+ num_layers ...................................... 16
391
+ num_layers_at_end_in_bf16 ....................... 1
392
+ num_layers_at_start_in_bf16 ..................... 1
393
+ num_layers_per_virtual_pipeline_stage ........... None
394
+ num_query_groups ................................ 16
395
+ num_virtual_stages_per_pipeline_rank ............ None
396
+ num_workers ..................................... 2
397
+ object_storage_cache_path ....................... None
398
+ one_logger_async ................................ False
399
+ one_logger_project .............................. megatron-lm
400
+ one_logger_run_name ............................. None
401
+ onnx_safe ....................................... None
402
+ openai_gelu ..................................... False
403
+ optimizer ....................................... adam
404
+ optimizer_cpu_offload ........................... False
405
+ optimizer_offload_fraction ...................... 1.0
406
+ output_bert_embeddings .......................... False
407
+ overlap_cpu_optimizer_d2h_h2d ................... False
408
+ overlap_grad_reduce ............................. True
409
+ overlap_moe_expert_parallel_comm ................ False
410
+ overlap_p2p_comm ................................ False
411
+ overlap_p2p_comm_warmup_flush ................... False
412
+ overlap_param_gather ............................ True
413
+ overlap_param_gather_with_optimizer_step ........ False
414
+ override_opt_param_scheduler .................... False
415
+ params_dtype .................................... torch.bfloat16
416
+ patch_dim ....................................... 16
417
+ per_split_data_args_path ........................ None
418
+ perform_initialization .......................... True
419
+ pin_cpu_grads ................................... True
420
+ pin_cpu_params .................................. True
421
+ pipeline_model_parallel_comm_backend ............ None
422
+ pipeline_model_parallel_layout .................. None
423
+ pipeline_model_parallel_size .................... 1
424
+ position_embedding_type ......................... rope
425
+ pretrained_checkpoint ........................... None
426
+ profile ......................................... True
427
+ profile_ranks ................................... [0, 1, 2, 3, 4, 5, 6, 7]
428
+ profile_step_end ................................ 8
429
+ profile_step_start .............................. 6
430
+ q_lora_rank ..................................... None
431
+ qk_head_dim ..................................... 128
432
+ qk_l2_norm ...................................... False
433
+ qk_layernorm .................................... False
434
+ qk_pos_emb_head_dim ............................. 64
435
+ query_in_block_prob ............................. 0.1
436
+ rampup_batch_size ............................... None
437
+ rank ............................................ 0
438
+ recompute_granularity ........................... None
439
+ recompute_method ................................ None
440
+ recompute_modules ............................... None
441
+ recompute_num_layers ............................ None
442
+ record_memory_history ........................... False
443
+ relative_attention_max_distance ................. 128
444
+ relative_attention_num_buckets .................. 32
445
+ replication ..................................... False
446
+ replication_factor .............................. 2
447
+ replication_jump ................................ None
448
+ rerun_mode ...................................... validate_results
449
+ reset_attention_mask ............................ False
450
+ reset_position_ids .............................. False
451
+ result_rejected_tracker_filename ................ None
452
+ retriever_report_topk_accuracies ................ []
453
+ retriever_score_scaling ......................... False
454
+ retriever_seq_length ............................ 256
455
+ retro_add_retriever ............................. False
456
+ retro_attention_gate ............................ 1
457
+ retro_cyclic_train_iters ........................ None
458
+ retro_encoder_attention_dropout ................. 0.1
459
+ retro_encoder_hidden_dropout .................... 0.1
460
+ retro_encoder_layers ............................ 2
461
+ retro_num_neighbors ............................. 2
462
+ retro_num_retrieved_chunks ...................... 2
463
+ retro_project_dir ............................... None
464
+ retro_verify_neighbor_count ..................... True
465
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
466
+ rope_scaling_factor ............................. 8.0
467
+ rope_type ....................................... None
468
+ rotary_base ..................................... 10000
469
+ rotary_interleaved .............................. False
470
+ rotary_percent .................................. 1.0
471
+ rotary_scaling_factor ........................... 1.0
472
+ rotary_seq_len_interpolation_factor ............. None
473
+ run_workload_inspector_server ................... False
474
+ sample_rate ..................................... 1.0
475
+ save ............................................ None
476
+ save_interval ................................... None
477
+ save_retain_interval ............................ None
478
+ scatter_gather_tensors_in_pipeline .............. True
479
+ seed ............................................ 1234
480
+ seq_length ...................................... 4096
481
+ sequence_parallel ............................... False
482
+ sft ............................................. False
483
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
484
+ sgd_momentum .................................... 0.9
485
+ short_seq_prob .................................. 0.1
486
+ skip_train ...................................... False
487
+ skipped_train_samples ........................... 0
488
+ spec ............................................ None
489
+ split ........................................... 100,0,0
490
+ squared_relu .................................... False
491
+ start_weight_decay .............................. 0.1
492
+ straggler_ctrlr_port ............................ 65535
493
+ straggler_minmax_count .......................... 1
494
+ suggested_communication_unit_size ............... None
495
+ swiglu .......................................... True
496
+ swin_backbone_type .............................. tiny
497
+ symmetric_ar_type ............................... None
498
+ te_rng_tracker .................................. True
499
+ tensor_model_parallel_size ...................... 1
500
+ tensorboard_dir ................................. None
501
+ tensorboard_log_interval ........................ 1
502
+ tensorboard_queue_size .......................... 1000
503
+ test_data_path .................................. None
504
+ test_mode ....................................... False
505
+ tiktoken_num_special_tokens ..................... 1000
506
+ tiktoken_pattern ................................ None
507
+ tiktoken_special_tokens ......................... None
508
+ timing_log_level ................................ 0
509
+ timing_log_option ............................... minmax
510
+ titles_data_path ................................ None
511
+ tokenizer_model ................................. None
512
+ tokenizer_type .................................. GPT2BPETokenizer
513
+ torch_fsdp2_reshard_after_forward ............... True
514
+ tp_comm_bootstrap_backend ....................... nccl
515
+ tp_comm_bulk_dgrad .............................. True
516
+ tp_comm_bulk_wgrad .............................. True
517
+ tp_comm_overlap ................................. False
518
+ tp_comm_overlap_ag .............................. True
519
+ tp_comm_overlap_cfg ............................. None
520
+ tp_comm_overlap_rs .............................. True
521
+ tp_comm_overlap_rs_dgrad ........................ False
522
+ tp_comm_split_ag ................................ True
523
+ tp_comm_split_rs ................................ True
524
+ train_data_path ................................. None
525
+ train_iters ..................................... 10
526
+ train_samples ................................... None
527
+ train_sync_interval ............................. None
528
+ transformer_impl ................................ transformer_engine
529
+ transformer_pipeline_model_parallel_size ........ 1
530
+ untie_embeddings_and_output_weights ............. True
531
+ use_checkpoint_args ............................. False
532
+ use_checkpoint_opt_param_scheduler .............. False
533
+ use_cpu_initialization .......................... None
534
+ use_custom_fsdp ................................. False
535
+ use_dist_ckpt ................................... True
536
+ use_dist_ckpt_deprecated ........................ False
537
+ use_distributed_optimizer ....................... True
538
+ use_flash_attn .................................. False
539
+ use_legacy_models ............................... False
540
+ use_mp_args_from_checkpoint_args ................ False
541
+ use_one_sent_docs ............................... False
542
+ use_persistent_ckpt_worker ...................... False
543
+ use_precision_aware_optimizer ................... False
544
+ use_pytorch_profiler ............................ False
545
+ use_ring_exchange_p2p ........................... False
546
+ use_rope_scaling ................................ False
547
+ use_rotary_position_embeddings .................. False
548
+ use_sharp ....................................... False
549
+ use_tokenizer_model_from_checkpoint_args ........ True
550
+ use_torch_fsdp2 ................................. False
551
+ use_torch_optimizer_for_cpu_offload ............. False
552
+ use_tp_pp_dp_mapping ............................ False
553
+ v_head_dim ...................................... 128
554
+ valid_data_path ................................. None
555
+ variable_seq_lengths ............................ False
556
+ virtual_pipeline_model_parallel_size ............ None
557
+ vision_backbone_type ............................ vit
558
+ vision_pretraining .............................. False
559
+ vision_pretraining_type ......................... classify
560
+ vocab_extra_ids ................................. 0
561
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
562
+ vocab_size ...................................... None
563
+ wandb_exp_name ..................................
564
+ wandb_project ...................................
565
+ wandb_save_dir ..................................
566
+ weight_decay .................................... 0.1
567
+ weight_decay_incr_style ......................... constant
568
+ wgrad_deferral_limit ............................ 0
569
+ world_size ...................................... 8
570
+ yaml_cfg ........................................ None
571
+ -------------------- end of arguments ---------------------
572
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1
573
+ > building GPT2BPETokenizer tokenizer ...
574
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
575
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
576
+ /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
577
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
578
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
579
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
580
+ > initializing torch distributed ...
581
+ WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it
582
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
591
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
594
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
595
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
596
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
597
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ [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
+ > initialized tensor model parallel with size 1
604
+ > initialized pipeline model parallel with size 1
605
+ > setting random seeds to 1234 ...
606
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
607
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
608
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
609
+ [rank2]:[W921 09:03:06.520564662 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank4]:[W921 09:03:06.520922171 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank5]:[W921 09:03:06.520930907 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank6]:[W921 09:03:06.521137123 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.
613
+ [rank1]:[W921 09:03:06.521641859 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.
614
+ [rank7]:[W921 09:03:06.521719100 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
615
+ [rank3]:[W921 09:03:06.521866657 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.
616
+ > compiling dataset index builder ...
617
+ make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
618
+ make: Nothing to be done for 'default'.
619
+ make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets'
620
+ >>> done with dataset index builder. Compilation time: 0.431 seconds
621
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
622
+ > compiling and loading fused kernels ...
623
+ [rank0]:[W921 09:03:07.164659069 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
624
+ >>> done with compiling and loading fused kernels. Compilation time: 3.775 seconds
625
+ time to initialize megatron (seconds): 7.527
626
+ [after megatron is initialized] datetime: 2026-09-21 09:03:13
627
+ building GPT model ...
628
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
629
+ warnings.warn(
630
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
631
+ warnings.warn(
632
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
633
+ warnings.warn(
634
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
635
+ warnings.warn(
636
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
637
+ warnings.warn(
638
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
639
+ warnings.warn(
640
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
641
+ warnings.warn(
642
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
643
+ warnings.warn(
644
+ [INFO] Use MNNVL fabric: False
645
+ [WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8.
646
+ NVSHMEM v3.7.2
647
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
648
+
649
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
650
+
651
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
652
+
653
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
654
+
655
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
656
+
657
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
658
+
659
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
660
+
661
+ /dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed.
662
+
663
+ UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive).
664
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 2087258112
665
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
666
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 7
667
+ Params for bucket 1 (103022592 elements, 103022592 padded size):
668
+ module.output_layer.weight
669
+ Params for bucket 2 (46542848 elements, 46542848 padded size):
670
+ module.decoder.layers.15.self_attention.linear_qkv.weight
671
+ module.decoder.layers.15.pre_mlp_layernorm.weight
672
+ module.decoder.layers.14.mlp.router.weight
673
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
674
+ module.decoder.layers.14.self_attention.linear_proj.weight
675
+ module.decoder.layers.13.pre_mlp_layernorm.weight
676
+ module.decoder.layers.13.self_attention.linear_qkv.weight
677
+ module.decoder.final_layernorm.weight
678
+ module.decoder.layers.15.mlp.router.weight
679
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
680
+ module.decoder.layers.15.self_attention.linear_proj.weight
681
+ module.decoder.layers.14.pre_mlp_layernorm.weight
682
+ module.decoder.layers.14.self_attention.linear_qkv.weight
683
+ module.decoder.layers.13.mlp.router.weight
684
+ Params for bucket 3 (50737152 elements, 50737152 padded size):
685
+ module.decoder.layers.12.self_attention.linear_qkv.weight
686
+ module.decoder.layers.12.pre_mlp_layernorm.weight
687
+ module.decoder.layers.11.pre_mlp_layernorm.weight
688
+ module.decoder.layers.11.self_attention.linear_qkv.weight
689
+ module.decoder.layers.10.pre_mlp_layernorm.weight
690
+ module.decoder.layers.10.self_attention.linear_qkv.weight
691
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
692
+ module.decoder.layers.13.self_attention.linear_proj.weight
693
+ module.decoder.layers.12.mlp.router.weight
694
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
695
+ module.decoder.layers.12.self_attention.linear_proj.weight
696
+ module.decoder.layers.11.mlp.router.weight
697
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
698
+ module.decoder.layers.11.self_attention.linear_proj.weight
699
+ module.decoder.layers.10.mlp.router.weight
700
+ Params for bucket 4 (50737152 elements, 50737152 padded size):
701
+ module.decoder.layers.9.pre_mlp_layernorm.weight
702
+ module.decoder.layers.9.self_attention.linear_qkv.weight
703
+ module.decoder.layers.8.pre_mlp_layernorm.weight
704
+ module.decoder.layers.8.self_attention.linear_qkv.weight
705
+ module.decoder.layers.7.pre_mlp_layernorm.weight
706
+ module.decoder.layers.7.self_attention.linear_qkv.weight
707
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
708
+ module.decoder.layers.9.mlp.router.weight
709
+ module.decoder.layers.10.self_attention.linear_proj.weight
710
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
711
+ module.decoder.layers.9.self_attention.linear_proj.weight
712
+ module.decoder.layers.8.mlp.router.weight
713
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
714
+ module.decoder.layers.8.self_attention.linear_proj.weight
715
+ module.decoder.layers.7.mlp.router.weight
716
+ Params for bucket 5 (50737152 elements, 50737152 padded size):
717
+ module.decoder.layers.6.self_attention.linear_qkv.weight
718
+ module.decoder.layers.6.pre_mlp_layernorm.weight
719
+ module.decoder.layers.5.pre_mlp_layernorm.weight
720
+ module.decoder.layers.5.self_attention.linear_qkv.weight
721
+ module.decoder.layers.5.self_attention.linear_proj.weight
722
+ module.decoder.layers.4.mlp.router.weight
723
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
724
+ module.decoder.layers.7.self_attention.linear_proj.weight
725
+ module.decoder.layers.6.mlp.router.weight
726
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
727
+ module.decoder.layers.6.self_attention.linear_proj.weight
728
+ module.decoder.layers.5.mlp.router.weight
729
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
730
+ module.decoder.layers.4.pre_mlp_layernorm.weight
731
+ module.decoder.layers.4.self_attention.linear_qkv.weight
732
+ Params for bucket 6 (50737152 elements, 50737152 padded size):
733
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
734
+ module.decoder.layers.3.mlp.router.weight
735
+ module.decoder.layers.4.self_attention.linear_proj.weight
736
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
737
+ module.decoder.layers.3.self_attention.linear_proj.weight
738
+ module.decoder.layers.2.mlp.router.weight
739
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
740
+ module.decoder.layers.2.self_attention.linear_proj.weight
741
+ module.decoder.layers.1.mlp.router.weight
742
+ module.decoder.layers.3.self_attention.linear_qkv.weight
743
+ module.decoder.layers.3.pre_mlp_layernorm.weight
744
+ module.decoder.layers.2.pre_mlp_layernorm.weight
745
+ module.decoder.layers.2.self_attention.linear_qkv.weight
746
+ module.decoder.layers.1.pre_mlp_layernorm.weight
747
+ module.decoder.layers.1.self_attention.linear_qkv.weight
748
+ Params for bucket 7 (124131328 elements, 124131328 padded size):
749
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
750
+ module.decoder.layers.0.pre_mlp_layernorm.weight
751
+ module.decoder.layers.1.self_attention.linear_proj.weight
752
+ module.decoder.layers.0.mlp.router.weight
753
+ module.decoder.layers.0.self_attention.linear_qkv.weight
754
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
755
+ module.decoder.layers.0.self_attention.linear_proj.weight
756
+ module.embedding.word_embeddings.weight
757
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 20
758
+ Params for bucket 1 (41943040 elements, 41943040 padded size):
759
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight6
760
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight4
761
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
762
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
763
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight7
764
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight5
765
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
766
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight7
767
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight5
768
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
769
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
770
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight6
771
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight4
772
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
773
+ Params for bucket 2 (41943040 elements, 41943040 padded size):
774
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
775
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight6
776
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight4
777
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
778
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
779
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight6
780
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight4
781
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight7
782
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
783
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight5
784
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
785
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
786
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight7
787
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight5
788
+ Params for bucket 3 (41943040 elements, 41943040 padded size):
789
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
790
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
791
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight6
792
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight4
793
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
794
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
795
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight7
796
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
797
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight7
798
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
799
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight5
800
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
801
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
802
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight6
803
+ Params for bucket 4 (41943040 elements, 41943040 padded size):
804
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight5
805
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
806
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
807
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight6
808
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight4
809
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
810
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
811
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight4
812
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
813
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
814
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight7
815
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight5
816
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
817
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
818
+ Params for bucket 5 (41943040 elements, 41943040 padded size):
819
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight7
820
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight5
821
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
822
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
823
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight6
824
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight4
825
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight4
826
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight6
827
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
828
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
829
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight7
830
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight5
831
+ Params for bucket 6 (41943040 elements, 41943040 padded size):
832
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
833
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight7
834
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
835
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight5
836
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
837
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
838
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
839
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
840
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight6
841
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight4
842
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
843
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
844
+ Params for bucket 7 (41943040 elements, 41943040 padded size):
845
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight7
846
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
847
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight5
848
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
849
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight6
850
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight4
851
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
852
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight6
853
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight4
854
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
855
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
856
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight7
857
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight5
858
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
859
+ Params for bucket 8 (41943040 elements, 41943040 padded size):
860
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight6
861
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
862
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight4
863
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
864
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
865
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight7
866
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight5
867
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
868
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight7
869
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight5
870
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
871
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
872
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight6
873
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight4
874
+ Params for bucket 9 (41943040 elements, 41943040 padded size):
875
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
876
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
877
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight6
878
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight4
879
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
880
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
881
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight7
882
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
883
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight7
884
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
885
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight5
886
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
887
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
888
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight6
889
+ Params for bucket 10 (41943040 elements, 41943040 padded size):
890
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight5
891
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
892
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
893
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight6
894
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight4
895
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
896
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
897
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight4
898
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
899
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
900
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight7
901
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight5
902
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
903
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
904
+ Params for bucket 11 (41943040 elements, 41943040 padded size):
905
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight7
906
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight5
907
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
908
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
909
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight6
910
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight4
911
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight6
912
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight4
913
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
914
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
915
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight7
916
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight5
917
+ Params for bucket 12 (41943040 elements, 41943040 padded size):
918
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
919
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight7
920
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
921
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight5
922
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
923
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
924
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
925
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
926
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight6
927
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight4
928
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
929
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
930
+ Params for bucket 13 (41943040 elements, 41943040 padded size):
931
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight6
932
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight4
933
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
934
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
935
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight6
936
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight4
937
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
938
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight7
939
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight5
940
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
941
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
942
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight7
943
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight5
944
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
945
+ Params for bucket 14 (41943040 elements, 41943040 padded size):
946
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight7
947
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight5
948
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
949
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
950
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
951
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight6
952
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight4
953
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
954
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight6
955
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight4
956
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
957
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
958
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight7
959
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight5
960
+ Params for bucket 15 (41943040 elements, 41943040 padded size):
961
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight7
962
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
963
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
964
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight5
965
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
966
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
967
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight6
968
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
969
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
970
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight6
971
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight4
972
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
973
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
974
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight7
975
+ Params for bucket 16 (41943040 elements, 41943040 padded size):
976
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
977
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight4
978
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
979
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight7
980
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight5
981
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
982
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
983
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight5
984
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
985
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
986
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight6
987
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight4
988
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
989
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
990
+ Params for bucket 17 (41943040 elements, 41943040 padded size):
991
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight4
992
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight6
993
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
994
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
995
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight7
996
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight5
997
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight7
998
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
999
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight5
1000
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
1001
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight6
1002
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight4
1003
+ Params for bucket 18 (41943040 elements, 41943040 padded size):
1004
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
1005
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
1006
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight6
1007
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight4
1008
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
1009
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
1010
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
1011
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
1012
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight7
1013
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight5
1014
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
1015
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
1016
+ Params for bucket 19 (41943040 elements, 41943040 padded size):
1017
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight7
1018
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight5
1019
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
1020
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
1021
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight7
1022
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight5
1023
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
1024
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight4
1025
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight6
1026
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
1027
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
1028
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight6
1029
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight4
1030
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
1031
+ Params for bucket 20 (8388608 elements, 8388608 padded size):
1032
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
1033
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
1034
+ 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 0x7d557fc4a540>, config_logger_dir='')
1035
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
1036
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-09-21 09:03:16
1037
+ > building train, validation, and test datasets ...
1038
+ > datasets target sizes (minimum size):
1039
+ train: 80
1040
+ validation: 0
1041
+ test: 0
1042
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
1043
+ > building train, validation, and test datasets for GPT ...
1044
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(80, 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 0x7d557fda9100>, 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)
1045
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
1046
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
1047
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
1048
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
1049
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
1050
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
1051
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
1052
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 71e99bc7c46bc3412b339a272e0f9cf3-GPTDataset-train-document_index.npy
1053
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 71e99bc7c46bc3412b339a272e0f9cf3-GPTDataset-train-sample_index.npy
1054
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 71e99bc7c46bc3412b339a272e0f9cf3-GPTDataset-train-shuffle_index.npy
1055
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
1056
+ > finished creating GPT datasets ...
1057
+ [after dataloaders are built] datetime: 2026-09-21 09:03:16
1058
+ done with setup ...
1059
+ training ...
1060
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
1061
+ (min, max) time across ranks (ms):
1062
+ model-and-optimizer-setup ......................: (3284.29, 3290.71)
1063
+ train/valid/test-data-iterators-setup ..........: (101.05, 119.60)
1064
+ [before the start of training step] datetime: 2026-09-21 09:03:16
1065
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1066
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1067
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1068
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1069
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1070
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1071
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1072
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1073
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1074
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1075
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1076
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1077
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1078
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1079
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1080
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1081
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1082
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1083
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1084
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1085
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1086
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1087
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1088
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1089
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1090
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1091
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1092
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1093
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1094
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1095
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1096
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1097
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1098
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1099
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1100
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1101
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1102
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1103
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1104
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1105
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1106
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1107
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1108
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1109
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1110
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1111
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1112
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1113
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1114
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1115
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1116
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1117
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1118
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1119
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1120
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1121
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1122
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1123
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1124
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1125
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1126
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1127
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1128
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1129
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1130
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1131
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1132
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1133
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1134
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1135
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1136
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1137
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1138
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1139
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1140
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1141
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1142
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1143
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1144
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1145
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1146
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1147
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1148
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1149
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1150
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1151
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1152
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1153
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1154
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1155
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1156
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1157
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1158
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1159
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1160
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1161
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1162
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1163
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1164
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1165
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1166
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1167
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1168
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1169
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1170
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1171
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1172
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1173
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1174
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1175
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1176
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1177
+ [2026-09-21 09:03:45] iteration 1/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 28663.1 | throughput per GPU (TFLOP/s/GPU): 1.1 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.118507E+01 | loss scale: 1.0 | grad norm: 35.483 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1178
+ Number of parameters in transformer block in billions: 6.71
1179
+ Number of parameters in embedding layers in billions: 0.21
1180
+ Total number of parameters in billions: 6.92
1181
+ Number of parameters in most loaded shard in billions: 6.9171
1182
+ Theoretical memory footprints: weight and optimizer=49474.72 MB
1183
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 18080.59228515625 | max allocated: 21748.87451171875 | reserved: 22168.0 | max reserved: 22168.0
1184
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1185
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1186
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1187
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1188
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1189
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1190
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1191
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1192
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1193
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1194
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1195
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1196
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1197
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1198
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1199
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1200
+ [2026-09-21 09:03:45] iteration 2/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 262.0 | throughput per GPU (TFLOP/s/GPU): 123.0 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.120943E+01 | loss scale: 1.0 | grad norm: 36.300 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1201
+ [2026-09-21 09:03:46] iteration 3/ 10 | consumed samples: 24 | elapsed time per iteration (ms): 592.1 | throughput per GPU (TFLOP/s/GPU): 54.4 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 1.006740E+01 | loss scale: 1.0 | grad norm: 101.586 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1202
+ Capture CUDA graph for training!!!
1203
+ CUDA graph capture done!!!
1204
+ [2026-09-21 09:03:47] iteration 4/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 841.8 | throughput per GPU (TFLOP/s/GPU): 38.3 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 1.023839E+01 | loss scale: 1.0 | grad norm: 28.929 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1205
+ [2026-09-21 09:03:47] iteration 5/ 10 | consumed samples: 40 | elapsed time per iteration (ms): 105.6 | throughput per GPU (TFLOP/s/GPU): 305.1 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 1.074156E+01 | loss scale: 1.0 | grad norm: 20.869 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1206
+ [2026-09-21 09:03:47] iteration 6/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 105.8 | throughput per GPU (TFLOP/s/GPU): 304.6 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 9.630577E+00 | loss scale: 1.0 | grad norm: 8.187 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1207
+ [2026-09-21 09:03:50] iteration 7/ 10 | consumed samples: 56 | elapsed time per iteration (ms): 2852.0 | throughput per GPU (TFLOP/s/GPU): 11.3 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 9.081074E+00 | loss scale: 1.0 | grad norm: 16.320 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1208
+ [2026-09-21 09:03:50] iteration 8/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 117.0 | throughput per GPU (TFLOP/s/GPU): 275.4 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.819430E+00 | loss scale: 1.0 | grad norm: 8.594 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 |
1209
+ Capture range started in the application.
1210
+ Capture range ended in the application.
1211
+ Generating '/tmp/nsys-report-fff2.qdstrm'
1212
+
1213
+
1214
+ [2026-09-21 09:04:29] iteration 9/ 10 | consumed samples: 72 | elapsed time per iteration (ms): 39721.7 | throughput per GPU (TFLOP/s/GPU): 0.8 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.710793E+00 | loss scale: 1.0 | grad norm: 3.806 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 34176 | min safe capacity: 1.0430 |
1215
+ [2026-09-21 09:04:30] iteration 10/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 114.2 | throughput per GPU (TFLOP/s/GPU): 282.2 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.653622E+00 | loss scale: 1.0 | grad norm: 3.502 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 34176 | min safe capacity: 1.0430 |
1216
+ [after training is done] datetime: 2026-09-21 09:04:30
1217
+ [rank0]:[W921 09:04:31.344513429 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())
1218
+ [rank7]:[W921 09:04:32.916538530 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())
1219
+ [rank1]:[W921 09:04:32.921672280 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())
1220
+ [rank3]:[W921 09:04:32.923572406 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())
1221
+ [rank4]:[W921 09:04:33.170544613 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())
1222
+ [rank2]:[W921 09:04:33.236412864 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())
1223
+ [rank6]:[W921 09:04:33.236751282 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())
1224
+ [rank5]:[W921 09:04:33.454661850 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())
1225
+ [rank0]:[W921 09:04:35.896605275 observer.cpp:116] Warning: Leaked callback handle: 1 (function operator())
1226
+ pure virtual method called
1227
+ terminate called without an active exception
1228
+ [rank4]:[W921 09:04:36.073109527 observer.cpp:116] Warning: Leaked callback handle: 1 (function operator())
1229
+ pure virtual method called
1230
+ terminate called without an active exception
1231
+ [rank2]:[W921 09:04:36.205705618 observer.cpp:116] Warning: Leaked callback handle: 1 (function operator())
1232
+ pure virtual method called
1233
+ terminate called without an active exception
1234
+ W0921 09:04:36.524000 787046 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787388 closing signal SIGTERM
1235
+ W0921 09:04:36.525000 787046 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787389 closing signal SIGTERM
1236
+ W0921 09:04:36.525000 787046 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787390 closing signal SIGTERM
1237
+ W0921 09:04:36.526000 787046 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787391 closing signal SIGTERM
1238
+ W0921 09:04:36.526000 787046 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787392 closing signal SIGTERM
1239
+ W0921 09:04:36.527000 787046 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787393 closing signal SIGTERM
1240
+ W0921 09:04:36.527000 787046 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 787395 closing signal SIGTERM
1241
+ Collection stop is not allowed in this state.
1242
+ E0921 09:04:38.009000 787046 torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: -6) local_rank: 0 (pid: 787387) of binary: /usr/bin/python
1243
+ Traceback (most recent call last):
1244
+ File "/usr/local/bin/torchrun", line 8, in <module>
1245
+ sys.exit(main())
1246
+ ^^^^^^
1247
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1248
+ return f(*args, **kwargs)
1249
+ ^^^^^^^^^^^^^^^^^^
1250
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1251
+ run(args)
1252
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1253
+ elastic_launch(
1254
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1255
+ return launch_agent(self._config, self._entrypoint, list(args))
1256
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1257
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
1258
+ raise ChildFailedError(
1259
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
1260
+ ============================================================
1261
+ pretrain_gpt.py FAILED
1262
+ ------------------------------------------------------------
1263
+ Failures:
1264
+ <NO_OTHER_FAILURES>
1265
+ ------------------------------------------------------------
1266
+ Root Cause (first observed failure):
1267
+ [0]:
1268
+ time : 2026-09-21_09:04:36
1269
+ host : sp-aiinfra-interactive-aghosh-8-mb99k-8xb200-moe-cg-0
1270
+ rank : 0 (local_rank: 0)
1271
+ exitcode : -6 (pid: 787387)
1272
+ error_file: <N/A>
1273
+ traceback : Signal 6 (SIGABRT) received by PID 787387
1274
+ ============================================================
1275
+ Generated:
1276
+ /workspace/granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375.nsys-rep
granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/verify_trace_ranks.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ report : /workspace/granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375.nsys-rep
2
+ kernels : 79722
3
+ devices : 8 of 8 -> [0, 1, 2, 3, 4, 5, 6, 7]
4
+ processes: 1
5
+ markers : 13312 ultraep_trace__ kernels on devices [0, 1, 2, 3, 4, 5, 6, 7]
6
+ PASS: all ranks present and backend markers emitted
granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/results.tsv ADDED
@@ -0,0 +1 @@
 
 
1
+ config backend policy slots seq gbs cap recompute rep status iter_ms_mean iter_ms_min peak_mib run_dir
granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/trace_extension_build.log CHANGED
@@ -51,3 +51,8 @@ Emitting ninja build file /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cp
51
  Compiling objects...
52
  Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
53
  [1/2] c++ -MMD -MF /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_ops.o.d -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -I/usr/local/lib/python3.12/dist-packages/torch/include -I/usr/local/lib/python3.12/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/cuda/include -I/usr/include/python3.12 -c -c /workspace/Megatron-UltraEP/megatron/core/transformer/moe/ultraep_trace_ops.cpp -o /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_ops.o -O3 -std=c++17 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1016"' -DTORCH_EXTENSION_NAME=_ultraep_trace_C
 
 
 
 
 
 
51
  Compiling objects...
52
  Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
53
  [1/2] c++ -MMD -MF /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_ops.o.d -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -I/usr/local/lib/python3.12/dist-packages/torch/include -I/usr/local/lib/python3.12/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/cuda/include -I/usr/include/python3.12 -c -c /workspace/Megatron-UltraEP/megatron/core/transformer/moe/ultraep_trace_ops.cpp -o /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_ops.o -O3 -std=c++17 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1016"' -DTORCH_EXTENSION_NAME=_ultraep_trace_C
54
+ [2/2] /usr/local/cuda/bin/nvcc --generate-dependencies-with-compile --dependency-output /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_kernels.o.d -I/usr/local/lib/python3.12/dist-packages/torch/include -I/usr/local/lib/python3.12/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/cuda/include -I/usr/include/python3.12 -c -c /workspace/Megatron-UltraEP/megatron/core/transformer/moe/ultraep_trace_kernels.cu -o /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_kernels.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options ''"'"'-fPIC'"'"'' -O3 -std=c++17 -lineinfo -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1016"' -DTORCH_EXTENSION_NAME=_ultraep_trace_C -gencode=arch=compute_100,code=sm_100 -gencode=arch=compute_120,code=compute_120 -gencode=arch=compute_120,code=sm_120 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_86,code=sm_86 -gencode=arch=compute_90,code=sm_90
55
+ x86_64-linux-gnu-g++ -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -shared -Wl,-O1 -Wl,-Bsymbolic-functions /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_kernels.o /workspace/Megatron-UltraEP/build/temp.linux-x86_64-cpython-312/megatron/core/transformer/moe/ultraep_trace_ops.o -L/usr/local/lib/python3.12/dist-packages/torch/lib -L/usr/local/cuda/lib64 -L/usr/lib/x86_64-linux-gnu -lc10 -ltorch -ltorch_cpu -ltorch_python -lcudart -lc10_cuda -ltorch_cuda -o build/lib.linux-x86_64-cpython-312/megatron/core/transformer/moe/_ultraep_trace_C.cpython-312-x86_64-linux-gnu.so
56
+ copying build/lib.linux-x86_64-cpython-312/megatron/core/datasets/helpers_cpp.cpython-312-x86_64-linux-gnu.so -> megatron/core/datasets
57
+ copying build/lib.linux-x86_64-cpython-312/megatron/core/transformer/moe/_ultraep_trace_C.cpython-312-x86_64-linux-gnu.so -> megatron/core/transformer/moe
58
+ load check passed: /workspace/Megatron-UltraEP/megatron/core/transformer/moe/_ultraep_trace_C.cpython-312-x86_64-linux-gnu.so -> torch.ops.megatron_ultraep_trace.fwd_ultraep_forward_begin
granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/trace_results.tsv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ model backend capacity status report async_overlap_json
2
+ OLMOE_1B_7B_L16 ultraep 1.08984375 OK /workspace/granularity_next_b200/tierb_models/traces/OLMOE_1B_7B_L16/ultraep/tierbtrace/tierbtrace_gbs8/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375/nsys/OLMOE_1B_7B_L16__ultraep__matched__seq4096__gbs8__cap1p08984375.nsys-rep