AbhishekGhosh1998 commited on
Commit
f236428
·
verified ·
1 Parent(s): ba25f0d

granularity results tierb-fixed-moonep-full/echo/tierbtrace @ 2026-08-30T00:27:29Z from sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0

Browse files
tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/nsys/run.env ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backend=echo
2
+ config=DSV2_LITE_L24
3
+ family=tierB
4
+ repo=/workspace/Megatron-MoE-Echo
5
+ branch=bracing-regions-through-kernels
6
+ commit=9a236793
7
+ num_layers=24
8
+ hidden=2048
9
+ ffn_hidden=10944
10
+ moe_ffn_hidden=1408
11
+ num_experts=64
12
+ topk=6
13
+ experts_per_rank=8
14
+ expert_params=8650752
15
+ heads=16
16
+ kv_groups=0
17
+ kv_channels=128
18
+ qk_layernorm=1
19
+ shared_ffn=2816
20
+ shared_expert_overlap=0
21
+ moe_token_dispatcher=flex
22
+ mla=1
23
+ dense_first_layers=1
24
+ moe_layers=23
25
+ rotary_base=10000
26
+ rotary_scaling=40
27
+ qkv_bias=0
28
+ router_balancing=seq_aux_loss
29
+ aux_loss_coeff=1e-3
30
+ router_pre_softmax=1
31
+ router_fusion=0
32
+ router_dtype=fp32
33
+ device_initiated_grouped_gemm=1
34
+ norm_epsilon=1e-6
35
+ seq_length=8192
36
+ micro_batch_size=1
37
+ global_batch_size=8
38
+ num_microbatches=1
39
+ assignments_per_rank=49152
40
+ mean_tokens_per_expert=6144
41
+ slot_policy=matched
42
+ slots_per_rank=8
43
+ capacity_mode=static
44
+ capacity=1.0755208333333333
45
+ cuda_graph=1
46
+ probe_mode=0
47
+ async_grad_reduce=0
48
+ branch_role=timing
49
+ memory_dump_mode=none
50
+ memory_snapshot_path=
51
+ cuda_graph_warmup_steps=3
52
+ recompute=none
53
+ force_balance=0
54
+ train_iters=10
55
+ ep_size=8
56
+ echo_dispatch_overlap=1
57
+ echo_dispatch_recompute=0
58
+ ultraep_quota_min_tokens=
59
+ ultraep_weight_sync_join=
60
+ profile_enable=1
61
+ kernel_trace=1
62
+ fp8_mode=mxfp8
63
+ fp8_recipe=mxfp8
64
+ training_seed=1234
65
+ learning_rate=1e-4
66
+ min_learning_rate=1.0e-5
67
+ lr_warmup_iters=5
68
+ state_dump_enable=0
69
+ state_dump_dir=
70
+ moonep_assert_no_pending_reduce=0
tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/nsys/train.log ADDED
@@ -0,0 +1,1390 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ==============================================================
2
+ DSV2_LITE_L24 / echo (slots/rank=8 policy=matched)
3
+ E=64 K=6 H'=1408 m=8
4
+ L=24 seq=8192 gbs=8 (1 microbatches)
5
+ cap_mode=static cap=1.0755208333333333 cg=1 recompute=none
6
+ memory_dump=none snapshots=/workspace/tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/nsys/memory_snapshots/snapshot.pickle
7
+ repo=/workspace/Megatron-MoE-Echo @ 9a236793
8
+ ==============================================================
9
+ W0830 00:26:48.721000 311359 torch/distributed/run.py:766]
10
+ W0830 00:26:48.721000 311359 torch/distributed/run.py:766] *****************************************
11
+ W0830 00:26:48.721000 311359 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ W0830 00:26:48.721000 311359 torch/distributed/run.py:766] *****************************************
13
+ using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1
14
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
15
+ Number of virtual stages per pipeline stage: None
16
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
17
+ using torch.bfloat16 for parameters ...
18
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
19
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
20
+ ------------------------ arguments ------------------------
21
+ account_for_embedding_in_pipeline_split ......... False
22
+ account_for_loss_in_pipeline_split .............. False
23
+ accumulate_allreduce_grads_in_fp32 .............. True
24
+ adam_beta1 ...................................... 0.9
25
+ adam_beta2 ...................................... 0.999
26
+ adam_eps ........................................ 1e-08
27
+ add_bias_linear ................................. False
28
+ add_position_embedding .......................... False
29
+ add_qkv_bias .................................... False
30
+ adlr_autoresume ................................. False
31
+ adlr_autoresume_interval ........................ 1000
32
+ align_grad_reduce ............................... True
33
+ align_param_gather .............................. False
34
+ app_tag_run_name ................................ None
35
+ app_tag_run_version ............................. 0.0.0
36
+ apply_layernorm_1p .............................. False
37
+ apply_query_key_layer_scaling ................... False
38
+ apply_residual_connection_post_layernorm ........ False
39
+ apply_rope_fusion ............................... False
40
+ async_save ...................................... None
41
+ async_tensor_model_parallel_allreduce ........... True
42
+ attention_backend ............................... AttnBackend.auto
43
+ attention_dropout ............................... 0.0
44
+ attention_softmax_in_fp32 ....................... False
45
+ auto_detect_ckpt_format ......................... False
46
+ barrier_with_L1_time ............................ True
47
+ bert_binary_head ................................ True
48
+ bert_embedder_type .............................. megatron
49
+ bert_load ....................................... None
50
+ bf16 ............................................ True
51
+ bias_dropout_fusion ............................. True
52
+ bias_gelu_fusion ................................ False
53
+ bias_swiglu_fusion .............................. True
54
+ biencoder_projection_dim ........................ 0
55
+ biencoder_shared_query_context_model ............ False
56
+ block_data_path ................................. None
57
+ calc_ft_timeouts ................................ False
58
+ calculate_per_token_loss ........................ False
59
+ check_for_large_grads ........................... False
60
+ check_for_nan_in_loss_and_grad .................. False
61
+ check_for_spiky_loss ............................ False
62
+ check_weight_hash_across_dp_replicas_interval ... None
63
+ ckpt_assume_constant_structure .................. False
64
+ ckpt_convert_format ............................. None
65
+ ckpt_convert_save ............................... None
66
+ ckpt_convert_update_legacy_dist_opt_format ...... False
67
+ ckpt_format ..................................... torch_dist
68
+ ckpt_fully_parallel_load ........................ False
69
+ ckpt_fully_parallel_save ........................ True
70
+ ckpt_fully_parallel_save_deprecated ............. False
71
+ ckpt_step ....................................... None
72
+ classes_fraction ................................ 1.0
73
+ clip_grad ....................................... 1.0
74
+ clone_scatter_output_in_embedding ............... True
75
+ config_logger_dir ...............................
76
+ consumed_train_samples .......................... 0
77
+ consumed_valid_samples .......................... 0
78
+ context_parallel_size ........................... 1
79
+ cp_comm_type .................................... ['p2p']
80
+ create_attention_mask_in_dataloader ............. True
81
+ cross_entropy_fusion_impl ....................... native
82
+ cross_entropy_loss_fusion ....................... False
83
+ cuda_graph_scope ................................ full_iteration
84
+ cuda_graph_warmup_steps ......................... 3
85
+ data_args_path .................................. None
86
+ data_cache_path ................................. None
87
+ data_parallel_random_init ....................... False
88
+ data_parallel_sharding_strategy ................. no_shard
89
+ data_parallel_size .............................. 8
90
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
91
+ data_per_class_fraction ......................... 1.0
92
+ data_sharding ................................... True
93
+ dataloader_type ................................. single
94
+ ddp_average_in_collective ....................... False
95
+ ddp_bucket_size ................................. None
96
+ ddp_num_buckets ................................. None
97
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
98
+ decoder_first_pipeline_num_layers ............... None
99
+ decoder_last_pipeline_num_layers ................ None
100
+ decoder_num_layers .............................. None
101
+ decoder_seq_length .............................. None
102
+ decoupled_lr .................................... None
103
+ decoupled_min_lr ................................ None
104
+ decrease_batch_size_if_needed ................... False
105
+ defer_embedding_wgrad_compute ................... False
106
+ delay_wgrad_compute ............................. False
107
+ deprecated_use_mcore_models ..................... True
108
+ deterministic_mode .............................. False
109
+ dino_bottleneck_size ............................ 256
110
+ dino_freeze_last_layer .......................... 1
111
+ dino_head_hidden_size ........................... 2048
112
+ dino_local_crops_number ......................... 10
113
+ dino_local_img_size ............................. 96
114
+ dino_norm_last_layer ............................ False
115
+ dino_teacher_temp ............................... 0.07
116
+ dino_warmup_teacher_temp ........................ 0.04
117
+ dino_warmup_teacher_temp_epochs ................. 30
118
+ disable_bf16_reduced_precision_matmul ........... False
119
+ disable_mamba_mem_eff_path ...................... False
120
+ disable_straggler_on_startup .................... False
121
+ dist_ckpt_format_deprecated ..................... None
122
+ dist_ckpt_strictness ............................ assume_ok_unexpected
123
+ distribute_saved_activations .................... False
124
+ distributed_backend ............................. nccl
125
+ distributed_timeout_minutes ..................... 10
126
+ embedding_init_method_std ....................... None
127
+ embedding_path .................................. None
128
+ empty_unused_memory_level ....................... 0
129
+ enable_cuda_graph ............................... True
130
+ enable_experimental ............................. False
131
+ enable_ft_package ............................... False
132
+ enable_gloo_process_groups ...................... True
133
+ enable_msc ...................................... True
134
+ enable_one_logger ............................... True
135
+ encoder_num_layers .............................. 24
136
+ encoder_seq_length .............................. 8192
137
+ end_weight_decay ................................ 0.1
138
+ eod_mask_loss ................................... False
139
+ error_injection_rate ............................ 0
140
+ error_injection_type ............................ transient_error
141
+ eval_interval ................................... 1000000
142
+ eval_iters ...................................... 0
143
+ evidence_data_path .............................. None
144
+ exit_duration_in_mins ........................... None
145
+ exit_interval ................................... None
146
+ exit_on_missing_checkpoint ...................... False
147
+ exit_signal_handler ............................. False
148
+ exp_avg_dtype ................................... torch.float32
149
+ exp_avg_sq_dtype ................................ torch.float32
150
+ expert_model_parallel_size ...................... 8
151
+ expert_tensor_parallel_size ..................... 1
152
+ export_force_local_attention .................... False
153
+ export_kd_cfg ................................... None
154
+ export_kd_teacher_ckpt_format ................... None
155
+ export_kd_teacher_load .......................... None
156
+ export_kv_cache_quant ........................... False
157
+ export_legacy_megatron .......................... False
158
+ export_model_type ............................... GPTModel
159
+ export_moe_apply_probs_on_input ................. False
160
+ export_qk_l2_norm ............................... False
161
+ export_quant_cfg ................................ None
162
+ export_real_quant_cfg ........................... None
163
+ export_te_mcore_model ........................... False
164
+ external_cuda_graph ............................. False
165
+ ffn_hidden_size ................................. 10944
166
+ finetune ........................................ False
167
+ finetune_data_split ............................. train
168
+ finetune_hf_dataset ............................. None
169
+ first_last_layers_bf16 .......................... False
170
+ flash_decode .................................... False
171
+ fp16 ............................................ False
172
+ fp16_lm_cross_entropy ........................... False
173
+ fp32_residual_connection ........................ False
174
+ fp8 ............................................. e4m3
175
+ fp8_amax_compute_algo ........................... most_recent
176
+ fp8_amax_history_len ............................ 1
177
+ fp8_interval .................................... 1
178
+ fp8_margin ...................................... 0
179
+ fp8_param_gather ................................ False
180
+ fp8_recipe ...................................... mxfp8
181
+ fp8_wgrad ....................................... True
182
+ fsdp_double_buffer .............................. False
183
+ global_batch_size ............................... 8
184
+ grad_reduce_in_bf16 ............................. False
185
+ gradient_accumulation_fusion .................... True
186
+ gradient_reduce_div_fusion ...................... True
187
+ group_query_attention ........................... False
188
+ head_lr_mult .................................... 1.0
189
+ heterogeneous_layers_config_encoded_json ........ None
190
+ heterogeneous_layers_config_path ................ None
191
+ hidden_dropout .................................. 0.0
192
+ hidden_size ..................................... 2048
193
+ hierarchical_context_parallel_sizes ............. None
194
+ high_priority_stream_groups ..................... []
195
+ hybrid_attention_ratio .......................... 0.0
196
+ hybrid_mlp_ratio ................................ 0.0
197
+ hybrid_override_pattern ......................... None
198
+ hysteresis ...................................... 2
199
+ ict_head_size ................................... None
200
+ ict_load ........................................ None
201
+ img_h ........................................... 224
202
+ img_w ........................................... 224
203
+ indexer_batch_size .............................. 128
204
+ indexer_log_interval ............................ 1000
205
+ inference_batch_times_seqlen_threshold .......... -1
206
+ inference_dynamic_batching ...................... False
207
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
208
+ inference_dynamic_batching_buffer_overflow_factor None
209
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
210
+ inference_dynamic_batching_chunk_size ........... 256
211
+ inference_dynamic_batching_max_requests_override None
212
+ inference_dynamic_batching_max_tokens_override .. None
213
+ inference_dynamic_batching_num_cuda_graphs ...... 16
214
+ inference_max_batch_size ........................ 8
215
+ inference_max_seq_length ........................ 2560
216
+ inference_rng_tracker ........................... False
217
+ init_method_std ................................. 0.01
218
+ init_method_xavier_uniform ...................... False
219
+ init_model_with_meta_device ..................... False
220
+ initial_loss_scale .............................. 4294967296
221
+ inprocess_active_world_size ..................... 8
222
+ inprocess_barrier_timeout ....................... 120
223
+ inprocess_completion_timeout .................... 120
224
+ inprocess_empty_cuda_cache ...................... False
225
+ inprocess_granularity ........................... node
226
+ inprocess_hard_timeout .......................... 90
227
+ inprocess_heartbeat_interval .................... 30
228
+ inprocess_heartbeat_timeout ..................... 60
229
+ inprocess_last_call_wait ........................ 1
230
+ inprocess_max_iterations ........................ None
231
+ inprocess_monitor_process_interval .............. 1.0
232
+ inprocess_monitor_thread_interval ............... 1.0
233
+ inprocess_progress_watchdog_interval ............ 1.0
234
+ inprocess_restart ............................... False
235
+ inprocess_soft_timeout .......................... 60
236
+ inprocess_termination_grace_time ................ 1
237
+ is_hybrid_model ................................. False
238
+ iter_per_epoch .................................. 1250
239
+ iterations_to_skip .............................. []
240
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
241
+ kitchen_config_file ............................. None
242
+ kitchen_recipe_number ........................... None
243
+ kv_channels ..................................... 128
244
+ kv_lora_rank .................................... 512
245
+ lazy_mpu_init ................................... None
246
+ load ............................................ None
247
+ load_main_params_from_ckpt ...................... None
248
+ load_model_opt_format ........................... False
249
+ local_rank ...................................... 0
250
+ log_energy ...................................... False
251
+ log_interval .................................... 1
252
+ log_loss_scale_to_tensorboard ................... True
253
+ log_memory_to_tensorboard ....................... False
254
+ log_num_zeros_in_grad ........................... False
255
+ log_params_norm ................................. False
256
+ log_progress .................................... False
257
+ log_straggler ................................... False
258
+ log_throughput .................................. True
259
+ log_timers_to_tensorboard ....................... False
260
+ log_validation_ppl_to_tensorboard ............... False
261
+ log_world_size_to_tensorboard ................... False
262
+ logging_level ................................... None
263
+ loss_scale ...................................... None
264
+ loss_scale_window ............................... 1000
265
+ lr .............................................. 0.0001
266
+ lr_decay_iters .................................. 8
267
+ lr_decay_samples ................................ None
268
+ lr_decay_style .................................. cosine
269
+ lr_warmup_fraction .............................. None
270
+ lr_warmup_init .................................. 0.0
271
+ lr_warmup_iters ................................. 5
272
+ lr_warmup_samples ............................... 0
273
+ lr_wsd_decay_iters .............................. None
274
+ lr_wsd_decay_samples ............................ None
275
+ lr_wsd_decay_style .............................. exponential
276
+ main_grads_dtype ................................ torch.float32
277
+ main_params_dtype ............................... torch.float32
278
+ make_vocab_size_divisible_by .................... 128
279
+ mamba_head_dim .................................. 64
280
+ mamba_num_groups ................................ 8
281
+ mamba_num_heads ................................. None
282
+ mamba_state_dim ................................. 128
283
+ manual_gc ....................................... False
284
+ manual_gc_eval .................................. True
285
+ manual_gc_interval .............................. 0
286
+ mask_factor ..................................... 1.0
287
+ mask_prob ....................................... 0.15
288
+ mask_type ....................................... random
289
+ masked_softmax_fusion ........................... False
290
+ max_position_embeddings ......................... 32768
291
+ max_tokens_to_oom ............................... 12000
292
+ memory_snapshot_path ............................ snapshot.pickle
293
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
294
+ micro_batch_size ................................ 1
295
+ microbatch_group_size_per_vp_stage .............. None
296
+ mid_level_dataset_surplus ....................... 0.005
297
+ min_loss_scale .................................. 1.0
298
+ min_lr .......................................... 1e-05
299
+ mlp_chunks_for_prefill .......................... 1
300
+ mmap_bin_files .................................. True
301
+ mock_data ....................................... False
302
+ moe_apply_probs_on_input ........................ False
303
+ moe_aux_loss_coeff .............................. 0.001
304
+ moe_deepep_num_sms .............................. 20
305
+ moe_echo_enable_random_offloading ............... False
306
+ moe_echo_expert_dispatch_overlap ................ True
307
+ moe_echo_expert_dispatcher_type ................. hybridep
308
+ moe_echo_recompute_expert_dispatch .............. False
309
+ moe_enable_deepep ............................... False
310
+ moe_enable_echo ................................. True
311
+ moe_enable_hybridep ............................. True
312
+ moe_expert_capacity_factor ...................... None
313
+ moe_extended_tp ................................. False
314
+ moe_ffn_hidden_size ............................. 1408
315
+ moe_grouped_gemm ................................ True
316
+ moe_hybridep_num_sms ............................ 16
317
+ moe_input_jitter_eps ............................ None
318
+ moe_layer_freq .................................. [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
319
+ moe_layer_recompute ............................. False
320
+ moe_num_echo_experts ............................ 64
321
+ moe_pad_expert_input_to_capacity ................ False
322
+ moe_per_layer_logging ........................... False
323
+ moe_permute_fusion .............................. False
324
+ moe_received_token_capacity ..................... 1.0755208333333333
325
+ moe_router_bias_update_rate ..................... 0.001
326
+ moe_router_dtype ................................ fp32
327
+ moe_router_enable_expert_bias ................... False
328
+ moe_router_force_load_balancing ................. False
329
+ moe_router_group_topk ........................... None
330
+ moe_router_load_balancing_type .................. seq_aux_loss
331
+ moe_router_num_groups ........................... None
332
+ moe_router_padding_for_fp8 ...................... False
333
+ moe_router_pre_softmax .......................... True
334
+ moe_router_score_function ....................... softmax
335
+ moe_router_topk ................................. 6
336
+ moe_router_topk_scaling_factor .................. None
337
+ moe_shared_expert_intermediate_size ............. 2816
338
+ moe_shared_expert_overlap ....................... False
339
+ moe_token_dispatcher_type ....................... flex
340
+ moe_token_drop_policy ........................... probs
341
+ moe_upcycling_granularity ....................... 1
342
+ moe_use_device_initiated_grouped_gemm ........... True
343
+ moe_use_legacy_grouped_gemm ..................... False
344
+ moe_use_upcycling ............................... False
345
+ moe_z_loss_coeff ................................ None
346
+ mrope_section ................................... None
347
+ mscale .......................................... 0.707
348
+ mscale_all_dim .................................. 0.707
349
+ mtp_loss_scaling_factor ......................... 0.1
350
+ mtp_num_layers .................................. None
351
+ multi_latent_attention .......................... True
352
+ nccl_all_reduce_for_prefill ..................... False
353
+ nccl_communicator_config_path ................... None
354
+ nccl_ub ......................................... False
355
+ no_load_optim ................................... None
356
+ no_load_rng ..................................... None
357
+ no_persist_layer_norm ........................... False
358
+ no_rope_freq .................................... None
359
+ no_save_optim ................................... None
360
+ no_save_rng ..................................... None
361
+ non_persistent_ckpt_type ........................ None
362
+ non_persistent_global_ckpt_dir .................. None
363
+ non_persistent_local_ckpt_algo .................. fully_parallel
364
+ non_persistent_local_ckpt_dir ................... None
365
+ non_persistent_save_interval .................... None
366
+ norm_epsilon .................................... 1e-06
367
+ normalization ................................... RMSNorm
368
+ num_attention_heads ............................. 16
369
+ num_channels .................................... 3
370
+ num_classes ..................................... 1000
371
+ num_dataset_builder_threads ..................... 1
372
+ num_distributed_optimizer_instances ............. 1
373
+ num_experts ..................................... 64
374
+ num_layers ...................................... 24
375
+ num_layers_at_end_in_bf16 ....................... 1
376
+ num_layers_at_start_in_bf16 ..................... 1
377
+ num_layers_per_virtual_pipeline_stage ........... None
378
+ num_query_groups ................................ 1
379
+ num_virtual_stages_per_pipeline_rank ............ None
380
+ num_workers ..................................... 2
381
+ object_storage_cache_path ....................... None
382
+ one_logger_async ................................ False
383
+ one_logger_project .............................. megatron-lm
384
+ one_logger_run_name ............................. None
385
+ onnx_safe ....................................... None
386
+ openai_gelu ..................................... False
387
+ optimizer ....................................... adam
388
+ optimizer_cpu_offload ........................... False
389
+ optimizer_offload_fraction ...................... 1.0
390
+ output_bert_embeddings .......................... False
391
+ overlap_cpu_optimizer_d2h_h2d ................... False
392
+ overlap_grad_reduce ............................. True
393
+ overlap_moe_expert_parallel_comm ................ False
394
+ overlap_p2p_comm ................................ False
395
+ overlap_p2p_comm_warmup_flush ................... False
396
+ overlap_param_gather ............................ True
397
+ overlap_param_gather_with_optimizer_step ........ False
398
+ override_opt_param_scheduler .................... False
399
+ params_dtype .................................... torch.bfloat16
400
+ patch_dim ....................................... 16
401
+ per_split_data_args_path ........................ None
402
+ perform_initialization .......................... True
403
+ pin_cpu_grads ................................... True
404
+ pin_cpu_params .................................. True
405
+ pipeline_model_parallel_comm_backend ............ None
406
+ pipeline_model_parallel_layout .................. None
407
+ pipeline_model_parallel_size .................... 1
408
+ position_embedding_type ......................... rope
409
+ pretrained_checkpoint ........................... None
410
+ profile ......................................... True
411
+ profile_ranks ................................... [0, 1, 2, 3, 4, 5, 6, 7]
412
+ profile_step_end ................................ 8
413
+ profile_step_start .............................. 6
414
+ q_lora_rank ..................................... None
415
+ qk_head_dim ..................................... 128
416
+ qk_l2_norm ...................................... False
417
+ qk_layernorm .................................... True
418
+ qk_pos_emb_head_dim ............................. 64
419
+ query_in_block_prob ............................. 0.1
420
+ rampup_batch_size ............................... None
421
+ rank ............................................ 0
422
+ recompute_granularity ........................... None
423
+ recompute_method ................................ None
424
+ recompute_modules ............................... None
425
+ recompute_num_layers ............................ None
426
+ record_memory_history ........................... False
427
+ relative_attention_max_distance ................. 128
428
+ relative_attention_num_buckets .................. 32
429
+ replication ..................................... False
430
+ replication_factor .............................. 2
431
+ replication_jump ................................ None
432
+ rerun_mode ...................................... validate_results
433
+ reset_attention_mask ............................ False
434
+ reset_position_ids .............................. False
435
+ result_rejected_tracker_filename ................ None
436
+ retriever_report_topk_accuracies ................ []
437
+ retriever_score_scaling ......................... False
438
+ retriever_seq_length ............................ 256
439
+ retro_add_retriever ............................. False
440
+ retro_attention_gate ............................ 1
441
+ retro_cyclic_train_iters ........................ None
442
+ retro_encoder_attention_dropout ................. 0.1
443
+ retro_encoder_hidden_dropout .................... 0.1
444
+ retro_encoder_layers ............................ 2
445
+ retro_num_neighbors ............................. 2
446
+ retro_num_retrieved_chunks ...................... 2
447
+ retro_project_dir ............................... None
448
+ retro_verify_neighbor_count ..................... True
449
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
450
+ rope_scaling_factor ............................. 8.0
451
+ rope_type ....................................... None
452
+ rotary_base ..................................... 10000
453
+ rotary_interleaved .............................. False
454
+ rotary_percent .................................. 1.0
455
+ rotary_scaling_factor ........................... 40.0
456
+ rotary_seq_len_interpolation_factor ............. None
457
+ run_workload_inspector_server ................... False
458
+ sample_rate ..................................... 1.0
459
+ save ............................................ None
460
+ save_interval ................................... None
461
+ save_retain_interval ............................ None
462
+ scatter_gather_tensors_in_pipeline .............. True
463
+ seed ............................................ 1234
464
+ seq_length ...................................... 8192
465
+ sequence_parallel ............................... False
466
+ sft ............................................. False
467
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
468
+ sgd_momentum .................................... 0.9
469
+ short_seq_prob .................................. 0.1
470
+ skip_train ...................................... False
471
+ skipped_train_samples ........................... 0
472
+ spec ............................................ None
473
+ split ........................................... 100,0,0
474
+ squared_relu .................................... False
475
+ start_weight_decay .............................. 0.1
476
+ straggler_ctrlr_port ............................ 65535
477
+ straggler_minmax_count .......................... 1
478
+ suggested_communication_unit_size ............... None
479
+ swiglu .......................................... True
480
+ swin_backbone_type .............................. tiny
481
+ symmetric_ar_type ............................... None
482
+ te_rng_tracker .................................. True
483
+ tensor_model_parallel_size ...................... 1
484
+ tensorboard_dir ................................. None
485
+ tensorboard_log_interval ........................ 1
486
+ tensorboard_queue_size .......................... 1000
487
+ test_data_path .................................. None
488
+ test_mode ....................................... False
489
+ tiktoken_num_special_tokens ..................... 1000
490
+ tiktoken_pattern ................................ None
491
+ tiktoken_special_tokens ......................... None
492
+ timing_log_level ................................ 0
493
+ timing_log_option ............................... minmax
494
+ titles_data_path ................................ None
495
+ tokenizer_model ................................. None
496
+ tokenizer_type .................................. GPT2BPETokenizer
497
+ torch_fsdp2_reshard_after_forward ............... True
498
+ tp_comm_bootstrap_backend ....................... nccl
499
+ tp_comm_bulk_dgrad .............................. True
500
+ tp_comm_bulk_wgrad .............................. True
501
+ tp_comm_overlap ................................. False
502
+ tp_comm_overlap_ag .............................. True
503
+ tp_comm_overlap_cfg ............................. None
504
+ tp_comm_overlap_rs .............................. True
505
+ tp_comm_overlap_rs_dgrad ........................ False
506
+ tp_comm_split_ag ................................ True
507
+ tp_comm_split_rs ................................ True
508
+ train_data_path ................................. None
509
+ train_iters ..................................... 10
510
+ train_samples ................................... None
511
+ train_sync_interval ............................. None
512
+ transformer_impl ................................ transformer_engine
513
+ transformer_pipeline_model_parallel_size ........ 1
514
+ untie_embeddings_and_output_weights ............. True
515
+ use_checkpoint_args ............................. False
516
+ use_checkpoint_opt_param_scheduler .............. False
517
+ use_cpu_initialization .......................... None
518
+ use_custom_fsdp ................................. False
519
+ use_dist_ckpt ................................... True
520
+ use_dist_ckpt_deprecated ........................ False
521
+ use_distributed_optimizer ....................... True
522
+ use_flash_attn .................................. False
523
+ use_legacy_models ............................... False
524
+ use_mp_args_from_checkpoint_args ................ False
525
+ use_one_sent_docs ............................... False
526
+ use_persistent_ckpt_worker ...................... False
527
+ use_precision_aware_optimizer ................... False
528
+ use_pytorch_profiler ............................ False
529
+ use_ring_exchange_p2p ........................... False
530
+ use_rope_scaling ................................ False
531
+ use_rotary_position_embeddings .................. False
532
+ use_sharp ....................................... False
533
+ use_tokenizer_model_from_checkpoint_args ........ True
534
+ use_torch_fsdp2 ................................. False
535
+ use_torch_optimizer_for_cpu_offload ............. False
536
+ use_tp_pp_dp_mapping ............................ False
537
+ v_head_dim ...................................... 128
538
+ valid_data_path ................................. None
539
+ variable_seq_lengths ............................ False
540
+ virtual_pipeline_model_parallel_size ............ None
541
+ vision_backbone_type ............................ vit
542
+ vision_pretraining .............................. False
543
+ vision_pretraining_type ......................... classify
544
+ vocab_extra_ids ................................. 0
545
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
546
+ vocab_size ...................................... None
547
+ wandb_exp_name ..................................
548
+ wandb_project ...................................
549
+ wandb_save_dir ..................................
550
+ weight_decay .................................... 0.1
551
+ weight_decay_incr_style ......................... constant
552
+ wgrad_deferral_limit ............................ 0
553
+ world_size ...................................... 8
554
+ yaml_cfg ........................................ None
555
+ -------------------- end of arguments ---------------------
556
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1
557
+ > building GPT2BPETokenizer tokenizer ...
558
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
559
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
560
+ > initializing torch distributed ...
561
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
562
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
563
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
564
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
565
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
566
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
567
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
568
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
569
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
570
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
571
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
572
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
573
+ 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
574
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
575
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
576
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
577
+ [Gloo] Rank [Gloo] Rank 3[Gloo] Rank 2 is connected to is connected to 717 peer ranks. is connected to Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is : 77 peer ranks. Expected number of connected peer ranks is : 7
578
+
579
+
580
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
582
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
586
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
590
+ [Gloo] Rank [Gloo] Rank 67 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77
591
+
592
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
593
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
594
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
595
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
596
+ [Gloo] Rank 0[Gloo] Rank is connected to 00 peer ranks. is connected to Expected number of connected peer ranks is : 00 peer ranks. Expected number of connected peer ranks is :
597
+ 0
598
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
599
+ > initialized tensor model parallel with size 1
600
+ > initialized pipeline model parallel with size 1
601
+ > setting random seeds to 1234 ...
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [rank7]:[W830 00:27:02.652864778 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.
604
+ [rank6]:[W830 00:27:02.657980167 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.
605
+ [rank2]:[W830 00:27:02.658087555 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.
606
+ [rank5]:[W830 00:27:02.658268190 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
607
+ [rank3]:[W830 00:27:02.658953701 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.
608
+ [rank1]:[W830 00:27:02.659103105 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
609
+ [rank4]:[W830 00:27:02.659200438 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ > compiling dataset index builder ...
611
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
612
+ make: Nothing to be done for 'default'.
613
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
614
+ >>> done with dataset index builder. Compilation time: 0.430 seconds
615
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
616
+ > compiling and loading fused kernels ...
617
+ [rank0]:[W830 00:27:03.271744494 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
618
+ >>> done with compiling and loading fused kernels. Compilation time: 3.626 seconds
619
+ time to initialize megatron (seconds): 9.039
620
+ [after megatron is initialized] datetime: 2026-08-30 00:27:08
621
+ building GPT model ...
622
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
623
+ warnings.warn(
624
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
625
+ warnings.warn(
626
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
627
+ warnings.warn(
628
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
629
+ warnings.warn(
630
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
631
+ warnings.warn(
632
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
633
+ warnings.warn(
634
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
635
+ warnings.warn(
636
+ /workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value.
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
+ /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.
645
+ warnings.warn(
646
+ /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.
647
+ warnings.warn(
648
+ /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.
649
+ warnings.warn(
650
+ /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.
651
+ warnings.warn(
652
+ /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.
653
+ warnings.warn(
654
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 2596386816
655
+ 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)
656
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 20
657
+ Params for bucket 1 (103022592 elements, 103022592 padded size):
658
+ module.output_layer.weight
659
+ Params for bucket 2 (48503296 elements, 48503296 padded size):
660
+ module.decoder.layers.23.pre_mlp_layernorm.weight
661
+ module.decoder.layers.23.self_attention.linear_kv_up_proj.layer_norm_weight
662
+ module.decoder.layers.23.self_attention.linear_kv_down_proj.weight
663
+ module.decoder.layers.23.self_attention.linear_proj.weight
664
+ module.decoder.layers.23.input_layernorm.weight
665
+ module.decoder.layers.22.mlp.shared_experts.linear_fc1.weight
666
+ module.decoder.final_layernorm.weight
667
+ module.decoder.layers.23.mlp.router.weight
668
+ module.decoder.layers.23.mlp.shared_experts.linear_fc2.weight
669
+ module.decoder.layers.23.mlp.shared_experts.linear_fc1.weight
670
+ module.decoder.layers.23.self_attention.linear_kv_up_proj.weight
671
+ module.decoder.layers.23.self_attention.linear_q_proj.weight
672
+ module.decoder.layers.22.mlp.shared_experts.linear_fc2.weight
673
+ Params for bucket 3 (40901632 elements, 40901632 padded size):
674
+ module.decoder.layers.22.mlp.router.weight
675
+ module.decoder.layers.22.self_attention.linear_kv_up_proj.weight
676
+ module.decoder.layers.22.self_attention.linear_proj.weight
677
+ module.decoder.layers.21.mlp.shared_experts.linear_fc1.weight
678
+ module.decoder.layers.21.mlp.router.weight
679
+ module.decoder.layers.21.self_attention.linear_kv_up_proj.weight
680
+ module.decoder.layers.22.pre_mlp_layernorm.weight
681
+ module.decoder.layers.22.self_attention.linear_kv_up_proj.layer_norm_weight
682
+ module.decoder.layers.22.self_attention.linear_kv_down_proj.weight
683
+ module.decoder.layers.22.self_attention.linear_q_proj.weight
684
+ module.decoder.layers.22.input_layernorm.weight
685
+ module.decoder.layers.21.mlp.shared_experts.linear_fc2.weight
686
+ module.decoder.layers.21.pre_mlp_layernorm.weight
687
+ module.decoder.layers.21.self_attention.linear_kv_up_proj.layer_norm_weight
688
+ module.decoder.layers.21.self_attention.linear_kv_down_proj.weight
689
+ module.decoder.layers.21.self_attention.linear_q_proj.weight
690
+ Params for bucket 4 (41163264 elements, 41163264 padded size):
691
+ module.decoder.layers.21.self_attention.linear_proj.weight
692
+ module.decoder.layers.20.mlp.shared_experts.linear_fc1.weight
693
+ module.decoder.layers.21.input_layernorm.weight
694
+ module.decoder.layers.20.pre_mlp_layernorm.weight
695
+ module.decoder.layers.20.self_attention.linear_kv_up_proj.layer_norm_weight
696
+ module.decoder.layers.20.self_attention.linear_kv_down_proj.weight
697
+ module.decoder.layers.20.self_attention.linear_q_proj.weight
698
+ module.decoder.layers.19.mlp.shared_experts.linear_fc2.weight
699
+ module.decoder.layers.20.mlp.shared_experts.linear_fc2.weight
700
+ module.decoder.layers.20.mlp.router.weight
701
+ module.decoder.layers.20.self_attention.linear_kv_up_proj.weight
702
+ module.decoder.layers.20.self_attention.linear_proj.weight
703
+ module.decoder.layers.20.input_layernorm.weight
704
+ Params for bucket 5 (42734080 elements, 42734080 padded size):
705
+ module.decoder.layers.19.self_attention.linear_kv_up_proj.weight
706
+ module.decoder.layers.19.mlp.router.weight
707
+ module.decoder.layers.19.self_attention.linear_q_proj.weight
708
+ module.decoder.layers.18.mlp.shared_experts.linear_fc2.weight
709
+ module.decoder.layers.19.mlp.shared_experts.linear_fc1.weight
710
+ module.decoder.layers.19.pre_mlp_layernorm.weight
711
+ module.decoder.layers.19.self_attention.linear_kv_up_proj.layer_norm_weight
712
+ module.decoder.layers.19.self_attention.linear_kv_down_proj.weight
713
+ module.decoder.layers.19.self_attention.linear_proj.weight
714
+ module.decoder.layers.19.input_layernorm.weight
715
+ module.decoder.layers.18.mlp.shared_experts.linear_fc1.weight
716
+ Params for bucket 6 (40901632 elements, 40901632 padded size):
717
+ module.decoder.layers.18.mlp.router.weight
718
+ module.decoder.layers.18.self_attention.linear_kv_up_proj.weight
719
+ module.decoder.layers.18.self_attention.linear_proj.weight
720
+ module.decoder.layers.18.input_layernorm.weight
721
+ module.decoder.layers.17.mlp.shared_experts.linear_fc1.weight
722
+ module.decoder.layers.17.pre_mlp_layernorm.weight
723
+ module.decoder.layers.17.self_attention.linear_kv_up_proj.layer_norm_weight
724
+ module.decoder.layers.17.self_attention.linear_kv_down_proj.weight
725
+ module.decoder.layers.17.self_attention.linear_q_proj.weight
726
+ module.decoder.layers.18.self_attention.linear_kv_down_proj.weight
727
+ module.decoder.layers.18.self_attention.linear_kv_up_proj.layer_norm_weight
728
+ module.decoder.layers.18.pre_mlp_layernorm.weight
729
+ module.decoder.layers.18.self_attention.linear_q_proj.weight
730
+ module.decoder.layers.17.mlp.shared_experts.linear_fc2.weight
731
+ module.decoder.layers.17.mlp.router.weight
732
+ module.decoder.layers.17.self_attention.linear_kv_up_proj.weight
733
+ Params for bucket 7 (41163264 elements, 41163264 padded size):
734
+ module.decoder.layers.17.self_attention.linear_proj.weight
735
+ module.decoder.layers.16.mlp.shared_experts.linear_fc1.weight
736
+ module.decoder.layers.16.mlp.router.weight
737
+ module.decoder.layers.17.input_layernorm.weight
738
+ module.decoder.layers.16.self_attention.linear_kv_up_proj.weight
739
+ module.decoder.layers.16.self_attention.linear_proj.weight
740
+ module.decoder.layers.16.input_layernorm.weight
741
+ module.decoder.layers.16.mlp.shared_experts.linear_fc2.weight
742
+ module.decoder.layers.16.pre_mlp_layernorm.weight
743
+ module.decoder.layers.16.self_attention.linear_kv_up_proj.layer_norm_weight
744
+ module.decoder.layers.16.self_attention.linear_kv_down_proj.weight
745
+ module.decoder.layers.16.self_attention.linear_q_proj.weight
746
+ module.decoder.layers.15.mlp.shared_experts.linear_fc2.weight
747
+ Params for bucket 8 (42734080 elements, 42734080 padded size):
748
+ module.decoder.layers.15.mlp.shared_experts.linear_fc1.weight
749
+ module.decoder.layers.15.self_attention.linear_kv_up_proj.layer_norm_weight
750
+ module.decoder.layers.15.pre_mlp_layernorm.weight
751
+ module.decoder.layers.15.self_attention.linear_kv_down_proj.weight
752
+ module.decoder.layers.15.self_attention.linear_proj.weight
753
+ module.decoder.layers.15.input_layernorm.weight
754
+ module.decoder.layers.14.mlp.shared_experts.linear_fc1.weight
755
+ module.decoder.layers.15.self_attention.linear_kv_up_proj.weight
756
+ module.decoder.layers.15.mlp.router.weight
757
+ module.decoder.layers.15.self_attention.linear_q_proj.weight
758
+ module.decoder.layers.14.mlp.shared_experts.linear_fc2.weight
759
+ Params for bucket 9 (40901632 elements, 40901632 padded size):
760
+ module.decoder.layers.14.mlp.router.weight
761
+ module.decoder.layers.14.self_attention.linear_kv_up_proj.layer_norm_weight
762
+ module.decoder.layers.14.self_attention.linear_kv_down_proj.weight
763
+ module.decoder.layers.14.pre_mlp_layernorm.weight
764
+ module.decoder.layers.14.self_attention.linear_q_proj.weight
765
+ module.decoder.layers.13.mlp.shared_experts.linear_fc2.weight
766
+ module.decoder.layers.13.mlp.router.weight
767
+ module.decoder.layers.13.self_attention.linear_kv_up_proj.weight
768
+ module.decoder.layers.14.self_attention.linear_kv_up_proj.weight
769
+ module.decoder.layers.14.self_attention.linear_proj.weight
770
+ module.decoder.layers.14.input_layernorm.weight
771
+ module.decoder.layers.13.mlp.shared_experts.linear_fc1.weight
772
+ module.decoder.layers.13.pre_mlp_layernorm.weight
773
+ module.decoder.layers.13.self_attention.linear_kv_up_proj.layer_norm_weight
774
+ module.decoder.layers.13.self_attention.linear_kv_down_proj.weight
775
+ module.decoder.layers.13.self_attention.linear_q_proj.weight
776
+ Params for bucket 10 (41163264 elements, 41163264 padded size):
777
+ module.decoder.layers.12.mlp.shared_experts.linear_fc2.weight
778
+ module.decoder.layers.12.pre_mlp_layernorm.weight
779
+ module.decoder.layers.12.self_attention.linear_kv_up_proj.layer_norm_weight
780
+ module.decoder.layers.12.self_attention.linear_kv_down_proj.weight
781
+ module.decoder.layers.12.self_attention.linear_q_proj.weight
782
+ module.decoder.layers.11.mlp.shared_experts.linear_fc2.weight
783
+ module.decoder.layers.13.self_attention.linear_proj.weight
784
+ module.decoder.layers.12.mlp.router.weight
785
+ module.decoder.layers.12.mlp.shared_experts.linear_fc1.weight
786
+ module.decoder.layers.13.input_layernorm.weight
787
+ module.decoder.layers.12.self_attention.linear_kv_up_proj.weight
788
+ module.decoder.layers.12.self_attention.linear_proj.weight
789
+ module.decoder.layers.12.input_layernorm.weight
790
+ Params for bucket 11 (42734080 elements, 42734080 padded size):
791
+ module.decoder.layers.11.self_attention.linear_kv_up_proj.weight
792
+ module.decoder.layers.11.mlp.router.weight
793
+ module.decoder.layers.11.self_attention.linear_q_proj.weight
794
+ module.decoder.layers.11.self_attention.linear_proj.weight
795
+ module.decoder.layers.11.input_layernorm.weight
796
+ module.decoder.layers.10.mlp.shared_experts.linear_fc1.weight
797
+ module.decoder.layers.11.mlp.shared_experts.linear_fc1.weight
798
+ module.decoder.layers.11.pre_mlp_layernorm.weight
799
+ module.decoder.layers.11.self_attention.linear_kv_up_proj.layer_norm_weight
800
+ module.decoder.layers.11.self_attention.linear_kv_down_proj.weight
801
+ module.decoder.layers.10.mlp.shared_experts.linear_fc2.weight
802
+ Params for bucket 12 (40901632 elements, 40901632 padded size):
803
+ module.decoder.layers.10.self_attention.linear_kv_down_proj.weight
804
+ module.decoder.layers.10.self_attention.linear_kv_up_proj.layer_norm_weight
805
+ module.decoder.layers.10.pre_mlp_layernorm.weight
806
+ module.decoder.layers.10.self_attention.linear_q_proj.weight
807
+ module.decoder.layers.9.mlp.shared_experts.linear_fc2.weight
808
+ module.decoder.layers.9.mlp.router.weight
809
+ module.decoder.layers.9.self_attention.linear_kv_up_proj.weight
810
+ module.decoder.layers.9.self_attention.linear_q_proj.weight
811
+ module.decoder.layers.10.mlp.router.weight
812
+ module.decoder.layers.10.self_attention.linear_kv_up_proj.weight
813
+ module.decoder.layers.10.self_attention.linear_proj.weight
814
+ module.decoder.layers.10.input_layernorm.weight
815
+ module.decoder.layers.9.mlp.shared_experts.linear_fc1.weight
816
+ module.decoder.layers.9.pre_mlp_layernorm.weight
817
+ module.decoder.layers.9.self_attention.linear_kv_up_proj.layer_norm_weight
818
+ module.decoder.layers.9.self_attention.linear_kv_down_proj.weight
819
+ Params for bucket 13 (41163264 elements, 41163264 padded size):
820
+ module.decoder.layers.9.input_layernorm.weight
821
+ module.decoder.layers.8.mlp.shared_experts.linear_fc2.weight
822
+ module.decoder.layers.8.mlp.shared_experts.linear_fc1.weight
823
+ module.decoder.layers.8.pre_mlp_layernorm.weight
824
+ module.decoder.layers.8.self_attention.linear_kv_up_proj.layer_norm_weight
825
+ module.decoder.layers.8.self_attention.linear_kv_down_proj.weight
826
+ module.decoder.layers.8.self_attention.linear_q_proj.weight
827
+ module.decoder.layers.8.self_attention.linear_proj.weight
828
+ module.decoder.layers.8.input_layernorm.weight
829
+ module.decoder.layers.9.self_attention.linear_proj.weight
830
+ module.decoder.layers.8.mlp.router.weight
831
+ module.decoder.layers.8.self_attention.linear_kv_up_proj.weight
832
+ module.decoder.layers.7.mlp.shared_experts.linear_fc2.weight
833
+ Params for bucket 14 (42734080 elements, 42734080 padded size):
834
+ module.decoder.layers.7.mlp.shared_experts.linear_fc1.weight
835
+ module.decoder.layers.7.pre_mlp_layernorm.weight
836
+ module.decoder.layers.7.self_attention.linear_kv_up_proj.layer_norm_weight
837
+ module.decoder.layers.7.self_attention.linear_kv_down_proj.weight
838
+ module.decoder.layers.7.self_attention.linear_proj.weight
839
+ module.decoder.layers.7.mlp.router.weight
840
+ module.decoder.layers.7.self_attention.linear_kv_up_proj.weight
841
+ module.decoder.layers.7.self_attention.linear_q_proj.weight
842
+ module.decoder.layers.7.input_layernorm.weight
843
+ module.decoder.layers.6.mlp.shared_experts.linear_fc2.weight
844
+ module.decoder.layers.6.mlp.shared_experts.linear_fc1.weight
845
+ Params for bucket 15 (40901632 elements, 40901632 padded size):
846
+ module.decoder.layers.6.self_attention.linear_kv_up_proj.weight
847
+ module.decoder.layers.6.self_attention.linear_q_proj.weight
848
+ module.decoder.layers.5.mlp.shared_experts.linear_fc1.weight
849
+ module.decoder.layers.5.pre_mlp_layernorm.weight
850
+ module.decoder.layers.5.self_attention.linear_kv_up_proj.layer_norm_weight
851
+ module.decoder.layers.5.self_attention.linear_kv_down_proj.weight
852
+ module.decoder.layers.5.self_attention.linear_q_proj.weight
853
+ module.decoder.layers.6.mlp.router.weight
854
+ module.decoder.layers.6.self_attention.linear_kv_up_proj.layer_norm_weight
855
+ module.decoder.layers.6.pre_mlp_layernorm.weight
856
+ module.decoder.layers.6.self_attention.linear_kv_down_proj.weight
857
+ module.decoder.layers.6.self_attention.linear_proj.weight
858
+ module.decoder.layers.6.input_layernorm.weight
859
+ module.decoder.layers.5.mlp.shared_experts.linear_fc2.weight
860
+ module.decoder.layers.5.mlp.router.weight
861
+ module.decoder.layers.5.self_attention.linear_kv_up_proj.weight
862
+ Params for bucket 16 (41163264 elements, 41163264 padded size):
863
+ module.decoder.layers.4.mlp.router.weight
864
+ module.decoder.layers.4.mlp.shared_experts.linear_fc1.weight
865
+ module.decoder.layers.4.self_attention.linear_kv_up_proj.weight
866
+ module.decoder.layers.4.self_attention.linear_proj.weight
867
+ module.decoder.layers.4.input_layernorm.weight
868
+ module.decoder.layers.5.self_attention.linear_proj.weight
869
+ module.decoder.layers.5.input_layernorm.weight
870
+ module.decoder.layers.4.mlp.shared_experts.linear_fc2.weight
871
+ module.decoder.layers.4.pre_mlp_layernorm.weight
872
+ module.decoder.layers.4.self_attention.linear_kv_up_proj.layer_norm_weight
873
+ module.decoder.layers.4.self_attention.linear_kv_down_proj.weight
874
+ module.decoder.layers.4.self_attention.linear_q_proj.weight
875
+ module.decoder.layers.3.mlp.shared_experts.linear_fc2.weight
876
+ Params for bucket 17 (42734080 elements, 42734080 padded size):
877
+ module.decoder.layers.3.mlp.shared_experts.linear_fc1.weight
878
+ module.decoder.layers.3.pre_mlp_layernorm.weight
879
+ module.decoder.layers.3.self_attention.linear_kv_up_proj.layer_norm_weight
880
+ module.decoder.layers.3.self_attention.linear_kv_down_proj.weight
881
+ module.decoder.layers.2.mlp.shared_experts.linear_fc2.weight
882
+ module.decoder.layers.3.mlp.router.weight
883
+ module.decoder.layers.3.self_attention.linear_kv_up_proj.weight
884
+ module.decoder.layers.3.self_attention.linear_q_proj.weight
885
+ module.decoder.layers.3.self_attention.linear_proj.weight
886
+ module.decoder.layers.3.input_layernorm.weight
887
+ module.decoder.layers.2.mlp.shared_experts.linear_fc1.weight
888
+ Params for bucket 18 (40901632 elements, 40901632 padded size):
889
+ module.decoder.layers.2.mlp.router.weight
890
+ module.decoder.layers.2.self_attention.linear_kv_up_proj.weight
891
+ module.decoder.layers.1.mlp.shared_experts.linear_fc2.weight
892
+ module.decoder.layers.1.pre_mlp_layernorm.weight
893
+ module.decoder.layers.1.self_attention.linear_kv_up_proj.weight
894
+ module.decoder.layers.2.self_attention.linear_kv_up_proj.layer_norm_weight
895
+ module.decoder.layers.2.self_attention.linear_kv_down_proj.weight
896
+ module.decoder.layers.2.pre_mlp_layernorm.weight
897
+ module.decoder.layers.2.self_attention.linear_q_proj.weight
898
+ module.decoder.layers.2.self_attention.linear_proj.weight
899
+ module.decoder.layers.2.input_layernorm.weight
900
+ module.decoder.layers.1.mlp.shared_experts.linear_fc1.weight
901
+ module.decoder.layers.1.mlp.router.weight
902
+ module.decoder.layers.1.self_attention.linear_kv_up_proj.layer_norm_weight
903
+ module.decoder.layers.1.self_attention.linear_kv_down_proj.weight
904
+ module.decoder.layers.1.self_attention.linear_q_proj.weight
905
+ Params for bucket 19 (71436288 elements, 71436288 padded size):
906
+ module.decoder.layers.1.self_attention.linear_proj.weight
907
+ module.decoder.layers.0.mlp.linear_fc2.weight
908
+ module.decoder.layers.1.input_layernorm.weight
909
+ module.decoder.layers.0.mlp.linear_fc1.weight
910
+ Params for bucket 20 (116789760 elements, 116789760 padded size):
911
+ module.decoder.layers.0.self_attention.linear_kv_up_proj.layer_norm_weight
912
+ module.decoder.layers.0.self_attention.linear_proj.weight
913
+ module.decoder.layers.0.input_layernorm.weight
914
+ module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight
915
+ module.decoder.layers.0.self_attention.linear_kv_up_proj.weight
916
+ module.decoder.layers.0.self_attention.linear_kv_down_proj.weight
917
+ module.decoder.layers.0.self_attention.linear_q_proj.weight
918
+ module.embedding.word_embeddings.weight
919
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 40
920
+ Params for bucket 1 (40370176 elements, 40370176 padded size):
921
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight7
922
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight3
923
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight5
924
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight1
925
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight7
926
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight5
927
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight4
928
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight6
929
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight2
930
+ module.decoder.layers.23.mlp.experts.linear_fc2.weight0
931
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight6
932
+ Params for bucket 2 (40370176 elements, 40370176 padded size):
933
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight3
934
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight1
935
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight7
936
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight5
937
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight4
938
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight0
939
+ module.decoder.layers.23.mlp.experts.linear_fc1.weight2
940
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight6
941
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight4
942
+ Params for bucket 3 (40370176 elements, 40370176 padded size):
943
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight3
944
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight1
945
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight6
946
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight4
947
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight7
948
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight0
949
+ module.decoder.layers.22.mlp.experts.linear_fc2.weight2
950
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight5
951
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight3
952
+ Params for bucket 4 (40370176 elements, 40370176 padded size):
953
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight2
954
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight0
955
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight6
956
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight4
957
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight2
958
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight0
959
+ module.decoder.layers.22.mlp.experts.linear_fc1.weight1
960
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight7
961
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight5
962
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight3
963
+ module.decoder.layers.21.mlp.experts.linear_fc2.weight1
964
+ Params for bucket 5 (40370176 elements, 40370176 padded size):
965
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight7
966
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight5
967
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight3
968
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight1
969
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight6
970
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight4
971
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight2
972
+ Params for bucket 6 (40370176 elements, 40370176 padded size):
973
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight5
974
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight7
975
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight3
976
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight1
977
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight6
978
+ module.decoder.layers.21.mlp.experts.linear_fc1.weight0
979
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight6
980
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight4
981
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight2
982
+ module.decoder.layers.20.mlp.experts.linear_fc2.weight0
983
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight7
984
+ Params for bucket 7 (40370176 elements, 40370176 padded size):
985
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight2
986
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight4
987
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight0
988
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight6
989
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight5
990
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight1
991
+ module.decoder.layers.20.mlp.experts.linear_fc1.weight3
992
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight7
993
+ Params for bucket 8 (40370176 elements, 40370176 padded size):
994
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight4
995
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight2
996
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight0
997
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight7
998
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight5
999
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight5
1000
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight3
1001
+ module.decoder.layers.19.mlp.experts.linear_fc2.weight1
1002
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight6
1003
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight4
1004
+ Params for bucket 9 (40370176 elements, 40370176 padded size):
1005
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight3
1006
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight1
1007
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight6
1008
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight4
1009
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight2
1010
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight0
1011
+ module.decoder.layers.19.mlp.experts.linear_fc1.weight2
1012
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight7
1013
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight5
1014
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight3
1015
+ Params for bucket 10 (40370176 elements, 40370176 padded size):
1016
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight5
1017
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight0
1018
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight7
1019
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight3
1020
+ module.decoder.layers.18.mlp.experts.linear_fc2.weight1
1021
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight6
1022
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight4
1023
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight2
1024
+ Params for bucket 11 (40370176 elements, 40370176 padded size):
1025
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight1
1026
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight7
1027
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight5
1028
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight3
1029
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight1
1030
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight6
1031
+ module.decoder.layers.18.mlp.experts.linear_fc1.weight0
1032
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight4
1033
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight2
1034
+ module.decoder.layers.17.mlp.experts.linear_fc2.weight0
1035
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight7
1036
+ Params for bucket 12 (40370176 elements, 40370176 padded size):
1037
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight6
1038
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight2
1039
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight4
1040
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight0
1041
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight3
1042
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight5
1043
+ module.decoder.layers.17.mlp.experts.linear_fc1.weight1
1044
+ Params for bucket 13 (40370176 elements, 40370176 padded size):
1045
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight6
1046
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight4
1047
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight2
1048
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight0
1049
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight6
1050
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight7
1051
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight3
1052
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight5
1053
+ module.decoder.layers.16.mlp.experts.linear_fc2.weight1
1054
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight7
1055
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight5
1056
+ Params for bucket 14 (40370176 elements, 40370176 padded size):
1057
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight4
1058
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight2
1059
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight0
1060
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight7
1061
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight5
1062
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight3
1063
+ module.decoder.layers.16.mlp.experts.linear_fc1.weight1
1064
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight6
1065
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight4
1066
+ Params for bucket 15 (40370176 elements, 40370176 padded size):
1067
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
1068
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
1069
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight6
1070
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight4
1071
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
1072
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight7
1073
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
1074
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight5
1075
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
1076
+ Params for bucket 16 (40370176 elements, 40370176 padded size):
1077
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
1078
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
1079
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight6
1080
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight4
1081
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
1082
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
1083
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
1084
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight7
1085
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight5
1086
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
1087
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
1088
+ Params for bucket 17 (40370176 elements, 40370176 padded size):
1089
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight7
1090
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
1091
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight5
1092
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
1093
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight6
1094
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight4
1095
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
1096
+ Params for bucket 18 (40370176 elements, 40370176 padded size):
1097
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight4
1098
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight6
1099
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
1100
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
1101
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight7
1102
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
1103
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight7
1104
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight5
1105
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
1106
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
1107
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight6
1108
+ Params for bucket 19 (40370176 elements, 40370176 padded size):
1109
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight5
1110
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
1111
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
1112
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight7
1113
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight4
1114
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
1115
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
1116
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight6
1117
+ Params for bucket 20 (40370176 elements, 40370176 padded size):
1118
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight5
1119
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
1120
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
1121
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight7
1122
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight5
1123
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight4
1124
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
1125
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
1126
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight6
1127
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight4
1128
+ Params for bucket 21 (40370176 elements, 40370176 padded size):
1129
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
1130
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
1131
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight6
1132
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight4
1133
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
1134
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
1135
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
1136
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight7
1137
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight5
1138
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
1139
+ Params for bucket 22 (40370176 elements, 40370176 padded size):
1140
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
1141
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight7
1142
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight5
1143
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
1144
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
1145
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight6
1146
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight4
1147
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
1148
+ Params for bucket 23 (40370176 elements, 40370176 padded size):
1149
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
1150
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight7
1151
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight5
1152
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
1153
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
1154
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight6
1155
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
1156
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight4
1157
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
1158
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
1159
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight7
1160
+ Params for bucket 24 (40370176 elements, 40370176 padded size):
1161
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight6
1162
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight4
1163
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
1164
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
1165
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight5
1166
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
1167
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
1168
+ Params for bucket 25 (40370176 elements, 40370176 padded size):
1169
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight4
1170
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight6
1171
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
1172
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
1173
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight7
1174
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight5
1175
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight7
1176
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight5
1177
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
1178
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
1179
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight6
1180
+ Params for bucket 26 (40370176 elements, 40370176 padded size):
1181
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
1182
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
1183
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight6
1184
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight4
1185
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight4
1186
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
1187
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
1188
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight7
1189
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight5
1190
+ Params for bucket 27 (40370176 elements, 40370176 padded size):
1191
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight7
1192
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
1193
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
1194
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight5
1195
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
1196
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
1197
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
1198
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight6
1199
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight4
1200
+ Params for bucket 28 (40370176 elements, 40370176 padded size):
1201
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight6
1202
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
1203
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight4
1204
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
1205
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
1206
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
1207
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight7
1208
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
1209
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight5
1210
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
1211
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
1212
+ Params for bucket 29 (40370176 elements, 40370176 padded size):
1213
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight6
1214
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight4
1215
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
1216
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight7
1217
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight5
1218
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
1219
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
1220
+ Params for bucket 30 (40370176 elements, 40370176 padded size):
1221
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
1222
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight7
1223
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight5
1224
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
1225
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
1226
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight6
1227
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight6
1228
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight4
1229
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
1230
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
1231
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight7
1232
+ Params for bucket 31 (40370176 elements, 40370176 padded size):
1233
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
1234
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight4
1235
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
1236
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight6
1237
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight5
1238
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
1239
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
1240
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight7
1241
+ Params for bucket 32 (40370176 elements, 40370176 padded size):
1242
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight4
1243
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
1244
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
1245
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight6
1246
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight4
1247
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight5
1248
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
1249
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
1250
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight7
1251
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight5
1252
+ Params for bucket 33 (40370176 elements, 40370176 padded size):
1253
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
1254
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
1255
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight6
1256
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight4
1257
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
1258
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
1259
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight7
1260
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
1261
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight5
1262
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
1263
+ Params for bucket 34 (40370176 elements, 40370176 padded size):
1264
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
1265
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight7
1266
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight5
1267
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
1268
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
1269
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight6
1270
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight4
1271
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
1272
+ Params for bucket 35 (40370176 elements, 40370176 padded size):
1273
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
1274
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight5
1275
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight7
1276
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
1277
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
1278
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
1279
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight6
1280
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight4
1281
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
1282
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
1283
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight7
1284
+ Params for bucket 36 (40370176 elements, 40370176 padded size):
1285
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight6
1286
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
1287
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight4
1288
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
1289
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight5
1290
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
1291
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
1292
+ Params for bucket 37 (40370176 elements, 40370176 padded size):
1293
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight7
1294
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight5
1295
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
1296
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
1297
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight7
1298
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight5
1299
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight6
1300
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight4
1301
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
1302
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
1303
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight6
1304
+ Params for bucket 38 (40370176 elements, 40370176 padded size):
1305
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
1306
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
1307
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight6
1308
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight4
1309
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight4
1310
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
1311
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
1312
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight7
1313
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight5
1314
+ Params for bucket 39 (40370176 elements, 40370176 padded size):
1315
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
1316
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
1317
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight6
1318
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight4
1319
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
1320
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
1321
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight7
1322
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight5
1323
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
1324
+ Params for bucket 40 (17301504 elements, 17301504 padded size):
1325
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
1326
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
1327
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
1328
+ 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 0x74c7a6e7c800>, config_logger_dir='')
1329
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
1330
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 00:27:09
1331
+ > building train, validation, and test datasets ...
1332
+ > datasets target sizes (minimum size):
1333
+ train: 80
1334
+ validation: 0
1335
+ test: 0
1336
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
1337
+ > building train, validation, and test datasets for GPT ...
1338
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(80, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=8192, 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 0x74c7a6e13410>, 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)
1339
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
1340
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
1341
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
1342
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
1343
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
1344
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
1345
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
1346
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 715d60356d3f5be048391704c2aaf5b2-GPTDataset-train-document_index.npy
1347
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 715d60356d3f5be048391704c2aaf5b2-GPTDataset-train-sample_index.npy
1348
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 715d60356d3f5be048391704c2aaf5b2-GPTDataset-train-shuffle_index.npy
1349
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 14536
1350
+ > finished creating GPT datasets ...
1351
+ [after dataloaders are built] datetime: 2026-08-30 00:27:09
1352
+ done with setup ...
1353
+ (min, max) time across ranks (ms):
1354
+ model-and-optimizer-setup ......................: (635.62, 655.51)
1355
+ train/valid/test-data-iterators-setup ..........: (88.44, 107.74)
1356
+ training ...
1357
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
1358
+ [before the start of training step] datetime: 2026-08-30 00:27:09
1359
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1360
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1361
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1362
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1363
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1364
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1365
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1366
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1367
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1368
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1369
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1370
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1371
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1372
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1373
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1374
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1375
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1376
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1377
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1378
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1379
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1380
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1381
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1382
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1383
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1384
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1385
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1386
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
1387
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future.
1388
+ warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.")
1389
+ /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future.
1390
+ warnings.warn("The use_host_meta is deprecated, it will be removed in the future.")
tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/nsys/train.log CHANGED
@@ -2803,3 +2803,47 @@ Capture range ended in the application.
2803
  Generating '/tmp/nsys-report-b003.qdstrm'
2804
 
2805
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2803
  Generating '/tmp/nsys-report-b003.qdstrm'
2804
 
2805
 
2806
+ Collection stop is not allowed in this state.
2807
+ W0830 00:26:21.094000 301137 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
2808
+ W0830 00:26:21.094000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301292 closing signal SIGTERM
2809
+ Generated:
2810
+ /workspace/tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/nsys/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125.nsys-rep
2811
+ W0830 00:26:21.097000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301293 closing signal SIGTERM
2812
+ W0830 00:26:21.099000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301294 closing signal SIGTERM
2813
+ W0830 00:26:21.101000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301295 closing signal SIGTERM
2814
+ W0830 00:26:21.109000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301296 closing signal SIGTERM
2815
+ W0830 00:26:21.111000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301297 closing signal SIGTERM
2816
+ W0830 00:26:21.124000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301298 closing signal SIGTERM
2817
+ W0830 00:26:21.125000 301137 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 301299 closing signal SIGTERM
2818
+ Traceback (most recent call last):
2819
+ File "/usr/local/bin/torchrun", line 8, in <module>
2820
+ sys.exit(main())
2821
+ ^^^^^^
2822
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
2823
+ return f(*args, **kwargs)
2824
+ ^^^^^^^^^^^^^^^^^^
2825
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
2826
+ run(args)
2827
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
2828
+ elastic_launch(
2829
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
2830
+ return launch_agent(self._config, self._entrypoint, list(args))
2831
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2832
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
2833
+ result = agent.run()
2834
+ ^^^^^^^^^^^
2835
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
2836
+ result = f(*args, **kwargs)
2837
+ ^^^^^^^^^^^^^^^^^^
2838
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
2839
+ result = self._invoke_run(role)
2840
+ ^^^^^^^^^^^^^^^^^^^^^^
2841
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
2842
+ time.sleep(monitor_interval)
2843
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
2844
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
2845
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 301137 got signal: 15
2846
+ The target application terminated. One or more process it created re-parented.
2847
+ Waiting for termination of re-parented processes.
2848
+ Use the `--wait` option to modify this behavior.
2849
+ There are no active sessions.
tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/nsys/verify_trace_ranks.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ report : /workspace/tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/nsys/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125.nsys-rep
2
+ kernels : 436880
3
+ devices : 8 of 8 -> [0, 1, 2, 3, 4, 5, 6, 7]
4
+ processes: 1
5
+ markers : 35328 echo_trace__ kernels on devices [0, 1, 2, 3, 4, 5, 6, 7]
6
+ PASS: all ranks present and backend markers emitted
tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/trace_results.tsv CHANGED
@@ -1,2 +1,3 @@
1
  model backend capacity status report async_overlap_json
2
  QWEN3_235B echo 1.083984375 OK /workspace/tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/QWEN3_235B__echo__matched__seq8192__gbs8__cap1p083984375/nsys/QWEN3_235B__echo__matched__seq8192__gbs8__cap1p083984375.nsys-rep
 
 
1
  model backend capacity status report async_overlap_json
2
  QWEN3_235B echo 1.083984375 OK /workspace/tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/QWEN3_235B__echo__matched__seq8192__gbs8__cap1p083984375/nsys/QWEN3_235B__echo__matched__seq8192__gbs8__cap1p083984375.nsys-rep
3
+ QWEN3_30B echo 1.11328125 OK /workspace/tierb-fixed-moonep-full/echo/tierbtrace/tierbtrace_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/nsys/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125.nsys-rep