AbhishekGhosh1998 commited on
Commit
7ea23fa
·
verified ·
1 Parent(s): 428629c

granularity results granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb @ 2026-09-21T08:43:49Z from sp-aiinfra-interactive-aghosh-8-mb99k-8xb200-moe-cg-0

Browse files
granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/OLMOE_1B_7B_L16__moonep__matched__seq4096__gbs8__capnone__async/rep3/mem_samples.csv CHANGED
@@ -62,3 +62,139 @@
62
  5, 4
63
  6, 4
64
  7, 4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
62
  5, 4
63
  6, 4
64
  7, 4
65
+ 0, 4
66
+ 1, 4
67
+ 2, 4
68
+ 3, 4
69
+ 4, 4
70
+ 5, 4
71
+ 6, 4
72
+ 7, 4
73
+ 0, 821
74
+ 1, 821
75
+ 2, 821
76
+ 3, 821
77
+ 4, 821
78
+ 5, 821
79
+ 6, 821
80
+ 7, 821
81
+ 0, 1537
82
+ 1, 1537
83
+ 2, 1537
84
+ 3, 1537
85
+ 4, 1537
86
+ 5, 1537
87
+ 6, 1537
88
+ 7, 1537
89
+ 0, 1823
90
+ 1, 1823
91
+ 2, 1875
92
+ 3, 1875
93
+ 4, 1823
94
+ 5, 1823
95
+ 6, 1823
96
+ 7, 1823
97
+ 0, 13391
98
+ 1, 13385
99
+ 2, 13409
100
+ 3, 13401
101
+ 4, 13397
102
+ 5, 13407
103
+ 6, 13387
104
+ 7, 13395
105
+ 0, 14363
106
+ 1, 14369
107
+ 2, 14383
108
+ 3, 14383
109
+ 4, 14387
110
+ 5, 14375
111
+ 6, 14333
112
+ 7, 14433
113
+ 0, 15063
114
+ 1, 15077
115
+ 2, 15053
116
+ 3, 15061
117
+ 4, 15069
118
+ 5, 15079
119
+ 6, 15057
120
+ 7, 15059
121
+ 0, 15599
122
+ 1, 15605
123
+ 2, 15607
124
+ 3, 15607
125
+ 4, 15597
126
+ 5, 15607
127
+ 6, 15607
128
+ 7, 15599
129
+ 0, 15877
130
+ 1, 15743
131
+ 2, 15883
132
+ 3, 15883
133
+ 4, 15875
134
+ 5, 15883
135
+ 6, 15609
136
+ 7, 15735
137
+ 0, 26339
138
+ 1, 26329
139
+ 2, 26337
140
+ 3, 26339
141
+ 4, 26345
142
+ 5, 26317
143
+ 6, 26329
144
+ 7, 26375
145
+ 0, 26775
146
+ 1, 26781
147
+ 2, 26801
148
+ 3, 26783
149
+ 4, 26793
150
+ 5, 26781
151
+ 6, 26781
152
+ 7, 26793
153
+ 0, 27127
154
+ 1, 27101
155
+ 2, 27131
156
+ 3, 27095
157
+ 4, 27109
158
+ 5, 27141
159
+ 6, 27163
160
+ 7, 27135
161
+ 0, 28063
162
+ 1, 28041
163
+ 2, 28077
164
+ 3, 28093
165
+ 4, 28087
166
+ 5, 28085
167
+ 6, 28093
168
+ 7, 28075
169
+ 0, 35739
170
+ 1, 35739
171
+ 2, 35727
172
+ 3, 35745
173
+ 4, 35751
174
+ 5, 35743
175
+ 6, 35745
176
+ 7, 35751
177
+ 0, 36405
178
+ 1, 36409
179
+ 2, 36429
180
+ 3, 36413
181
+ 4, 36421
182
+ 5, 36411
183
+ 6, 36413
184
+ 7, 36421
185
+ 0, 35071
186
+ 1, 35763
187
+ 2, 35031
188
+ 3, 34433
189
+ 4, 34703
190
+ 5, 34693
191
+ 6, 34759
192
+ 7, 35775
193
+ 0, 5889
194
+ 1, 4609
195
+ 2, 4609
196
+ 3, 4609
197
+ 4, 4612
198
+ 5, 4612
199
+ 6, 4868
200
+ 7, 35775
granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/OLMOE_1B_7B_L16__moonep__matched__seq4096__gbs8__capnone__async/rep3/train.log CHANGED
@@ -10,3 +10,1107 @@ W0921 08:42:53.732000 714978 torch/distributed/run.py:766]
10
  W0921 08:42:53.732000 714978 torch/distributed/run.py:766] *****************************************
11
  W0921 08:42:53.732000 714978 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
  W0921 08:42:53.732000 714978 torch/distributed/run.py:766] *****************************************
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  W0921 08:42:53.732000 714978 torch/distributed/run.py:766] *****************************************
11
  W0921 08:42:53.732000 714978 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
  W0921 08:42:53.732000 714978 torch/distributed/run.py:766] *****************************************
13
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
14
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
15
+ 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
16
+ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer
17
+ Number of virtual stages per pipeline stage: None
18
+ accumulate and all-reduce gradients in fp32 for bfloat16 data type.
19
+ using torch.bfloat16 for parameters ...
20
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
21
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
22
+ ------------------------ arguments ------------------------
23
+ account_for_embedding_in_pipeline_split ......... False
24
+ account_for_loss_in_pipeline_split .............. False
25
+ accumulate_allreduce_grads_in_fp32 .............. True
26
+ adam_beta1 ...................................... 0.9
27
+ adam_beta2 ...................................... 0.999
28
+ adam_eps ........................................ 1e-08
29
+ add_bias_linear ................................. False
30
+ add_position_embedding .......................... False
31
+ add_qkv_bias .................................... False
32
+ adlr_autoresume ................................. False
33
+ adlr_autoresume_interval ........................ 1000
34
+ align_grad_reduce ............................... True
35
+ align_param_gather .............................. False
36
+ app_tag_run_name ................................ None
37
+ app_tag_run_version ............................. 0.0.0
38
+ apply_layernorm_1p .............................. False
39
+ apply_query_key_layer_scaling ................... False
40
+ apply_residual_connection_post_layernorm ........ False
41
+ apply_rope_fusion ............................... True
42
+ async_save ...................................... None
43
+ async_tensor_model_parallel_allreduce ........... True
44
+ attention_backend ............................... AttnBackend.auto
45
+ attention_dropout ............................... 0.0
46
+ attention_softmax_in_fp32 ....................... False
47
+ auto_detect_ckpt_format ......................... False
48
+ barrier_with_L1_time ............................ True
49
+ bert_binary_head ................................ True
50
+ bert_embedder_type .............................. megatron
51
+ bert_load ....................................... None
52
+ bf16 ............................................ True
53
+ bias_dropout_fusion ............................. True
54
+ bias_gelu_fusion ................................ False
55
+ bias_swiglu_fusion .............................. True
56
+ biencoder_projection_dim ........................ 0
57
+ biencoder_shared_query_context_model ............ False
58
+ block_data_path ................................. None
59
+ calc_ft_timeouts ................................ False
60
+ calculate_per_token_loss ........................ False
61
+ check_for_large_grads ........................... False
62
+ check_for_nan_in_loss_and_grad .................. False
63
+ check_for_spiky_loss ............................ False
64
+ check_weight_hash_across_dp_replicas_interval ... None
65
+ ckpt_assume_constant_structure .................. False
66
+ ckpt_convert_format ............................. None
67
+ ckpt_convert_save ............................... None
68
+ ckpt_convert_update_legacy_dist_opt_format ...... False
69
+ ckpt_format ..................................... torch_dist
70
+ ckpt_fully_parallel_load ........................ False
71
+ ckpt_fully_parallel_save ........................ True
72
+ ckpt_fully_parallel_save_deprecated ............. False
73
+ ckpt_step ....................................... None
74
+ classes_fraction ................................ 1.0
75
+ clip_grad ....................................... 1.0
76
+ clone_scatter_output_in_embedding ............... True
77
+ config_logger_dir ...............................
78
+ consumed_train_samples .......................... 0
79
+ consumed_valid_samples .......................... 0
80
+ context_parallel_size ........................... 1
81
+ cp_comm_type .................................... ['p2p']
82
+ create_attention_mask_in_dataloader ............. True
83
+ cross_entropy_fusion_impl ....................... native
84
+ cross_entropy_loss_fusion ....................... False
85
+ cuda_graph_scope ................................ full_iteration
86
+ cuda_graph_warmup_steps ......................... 3
87
+ data_args_path .................................. None
88
+ data_cache_path ................................. None
89
+ data_parallel_random_init ....................... False
90
+ data_parallel_sharding_strategy ................. no_shard
91
+ data_parallel_size .............................. 8
92
+ data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document']
93
+ data_per_class_fraction ......................... 1.0
94
+ data_sharding ................................... True
95
+ dataloader_type ................................. single
96
+ ddp_average_in_collective ....................... False
97
+ ddp_bucket_size ................................. None
98
+ ddp_num_buckets ................................. None
99
+ ddp_pad_buckets_for_high_nccl_busbw ............. False
100
+ decoder_first_pipeline_num_layers ............... None
101
+ decoder_last_pipeline_num_layers ................ None
102
+ decoder_num_layers .............................. None
103
+ decoder_seq_length .............................. None
104
+ decoupled_lr .................................... None
105
+ decoupled_min_lr ................................ None
106
+ decrease_batch_size_if_needed ................... False
107
+ defer_embedding_wgrad_compute ................... False
108
+ delay_wgrad_compute ............................. False
109
+ deprecated_use_mcore_models ..................... True
110
+ deterministic_mode .............................. False
111
+ dino_bottleneck_size ............................ 256
112
+ dino_freeze_last_layer .......................... 1
113
+ dino_head_hidden_size ........................... 2048
114
+ dino_local_crops_number ......................... 10
115
+ dino_local_img_size ............................. 96
116
+ dino_norm_last_layer ............................ False
117
+ dino_teacher_temp ............................... 0.07
118
+ dino_warmup_teacher_temp ........................ 0.04
119
+ dino_warmup_teacher_temp_epochs ................. 30
120
+ disable_bf16_reduced_precision_matmul ........... False
121
+ disable_mamba_mem_eff_path ...................... False
122
+ disable_straggler_on_startup .................... False
123
+ dist_ckpt_format_deprecated ..................... None
124
+ dist_ckpt_strictness ............................ assume_ok_unexpected
125
+ distribute_saved_activations .................... False
126
+ distributed_backend ............................. nccl
127
+ distributed_timeout_minutes ..................... 10
128
+ embedding_init_method_std ....................... None
129
+ embedding_path .................................. None
130
+ empty_unused_memory_level ....................... 0
131
+ enable_cuda_graph ............................... True
132
+ enable_experimental ............................. False
133
+ enable_ft_package ............................... False
134
+ enable_gloo_process_groups ...................... True
135
+ enable_msc ...................................... True
136
+ enable_one_logger ............................... True
137
+ encoder_num_layers .............................. 16
138
+ encoder_seq_length .............................. 4096
139
+ end_weight_decay ................................ 0.1
140
+ eod_mask_loss ................................... False
141
+ error_injection_rate ............................ 0
142
+ error_injection_type ............................ transient_error
143
+ eval_interval ................................... 1000000
144
+ eval_iters ...................................... 0
145
+ evidence_data_path .............................. None
146
+ exit_duration_in_mins ........................... None
147
+ exit_interval ................................... None
148
+ exit_on_missing_checkpoint ...................... False
149
+ exit_signal_handler ............................. False
150
+ exp_avg_dtype ................................... torch.float32
151
+ exp_avg_sq_dtype ................................ torch.float32
152
+ expert_model_parallel_size ...................... 8
153
+ expert_tensor_parallel_size ..................... 1
154
+ export_force_local_attention .................... False
155
+ export_kd_cfg ................................... None
156
+ export_kd_teacher_ckpt_format ................... None
157
+ export_kd_teacher_load .......................... None
158
+ export_kv_cache_quant ........................... False
159
+ export_legacy_megatron .......................... False
160
+ export_model_type ............................... GPTModel
161
+ export_moe_apply_probs_on_input ................. False
162
+ export_qk_l2_norm ............................... False
163
+ export_quant_cfg ................................ None
164
+ export_real_quant_cfg ........................... None
165
+ export_te_mcore_model ........................... False
166
+ external_cuda_graph ............................. False
167
+ ffn_hidden_size ................................. 1024
168
+ finetune ........................................ False
169
+ finetune_data_split ............................. train
170
+ finetune_hf_dataset ............................. None
171
+ first_last_layers_bf16 .......................... False
172
+ flash_decode .................................... False
173
+ fp16 ............................................ False
174
+ fp16_lm_cross_entropy ........................... False
175
+ fp32_residual_connection ........................ False
176
+ fp8 ............................................. e4m3
177
+ fp8_amax_compute_algo ........................... most_recent
178
+ fp8_amax_history_len ............................ 1
179
+ fp8_interval .................................... 1
180
+ fp8_margin ...................................... 0
181
+ fp8_param_gather ................................ False
182
+ fp8_recipe ...................................... mxfp8
183
+ fp8_wgrad ....................................... True
184
+ fsdp_double_buffer .............................. False
185
+ global_batch_size ............................... 8
186
+ grad_reduce_in_bf16 ............................. False
187
+ gradient_accumulation_fusion .................... True
188
+ gradient_reduce_div_fusion ...................... True
189
+ group_query_attention ........................... True
190
+ head_lr_mult .................................... 1.0
191
+ heterogeneous_layers_config_encoded_json ........ None
192
+ heterogeneous_layers_config_path ................ None
193
+ hidden_dropout .................................. 0.0
194
+ hidden_size ..................................... 2048
195
+ hierarchical_context_parallel_sizes ............. None
196
+ high_priority_stream_groups ..................... []
197
+ hybrid_attention_ratio .......................... 0.0
198
+ hybrid_mlp_ratio ................................ 0.0
199
+ hybrid_override_pattern ......................... None
200
+ hysteresis ...................................... 2
201
+ ict_head_size ................................... None
202
+ ict_load ........................................ None
203
+ img_h ........................................... 224
204
+ img_w ........................................... 224
205
+ indexer_batch_size .............................. 128
206
+ indexer_log_interval ............................ 1000
207
+ inference_batch_times_seqlen_threshold .......... -1
208
+ inference_dynamic_batching ...................... False
209
+ inference_dynamic_batching_buffer_guaranteed_fraction 0.2
210
+ inference_dynamic_batching_buffer_overflow_factor None
211
+ inference_dynamic_batching_buffer_size_gb ....... 40.0
212
+ inference_dynamic_batching_chunk_size ........... 256
213
+ inference_dynamic_batching_max_requests_override None
214
+ inference_dynamic_batching_max_tokens_override .. None
215
+ inference_dynamic_batching_num_cuda_graphs ...... 16
216
+ inference_max_batch_size ........................ 8
217
+ inference_max_seq_length ........................ 2560
218
+ inference_rng_tracker ........................... False
219
+ init_method_std ................................. 0.02
220
+ init_method_xavier_uniform ...................... False
221
+ init_model_with_meta_device ..................... False
222
+ initial_loss_scale .............................. 4294967296
223
+ inprocess_active_world_size ..................... 8
224
+ inprocess_barrier_timeout ....................... 120
225
+ inprocess_completion_timeout .................... 120
226
+ inprocess_empty_cuda_cache ...................... False
227
+ inprocess_granularity ........................... node
228
+ inprocess_hard_timeout .......................... 90
229
+ inprocess_heartbeat_interval .................... 30
230
+ inprocess_heartbeat_timeout ..................... 60
231
+ inprocess_last_call_wait ........................ 1
232
+ inprocess_max_iterations ........................ None
233
+ inprocess_monitor_process_interval .............. 1.0
234
+ inprocess_monitor_thread_interval ............... 1.0
235
+ inprocess_progress_watchdog_interval ............ 1.0
236
+ inprocess_restart ............................... False
237
+ inprocess_soft_timeout .......................... 60
238
+ inprocess_termination_grace_time ................ 1
239
+ is_hybrid_model ................................. False
240
+ iter_per_epoch .................................. 1250
241
+ iterations_to_skip .............................. []
242
+ keep_fp8_transpose_cache_when_using_custom_fsdp . False
243
+ kitchen_config_file ............................. None
244
+ kitchen_recipe_number ........................... None
245
+ kv_channels ..................................... 128
246
+ kv_lora_rank .................................... 32
247
+ lazy_mpu_init ................................... None
248
+ load ............................................ None
249
+ load_main_params_from_ckpt ...................... None
250
+ load_model_opt_format ........................... False
251
+ local_rank ...................................... 0
252
+ log_energy ...................................... False
253
+ log_interval .................................... 1
254
+ log_loss_scale_to_tensorboard ................... True
255
+ log_memory_to_tensorboard ....................... False
256
+ log_num_zeros_in_grad ........................... False
257
+ log_params_norm ................................. False
258
+ log_progress .................................... False
259
+ log_straggler ................................... False
260
+ log_throughput .................................. True
261
+ log_timers_to_tensorboard ....................... False
262
+ log_validation_ppl_to_tensorboard ............... False
263
+ log_world_size_to_tensorboard ................... False
264
+ logging_level ................................... None
265
+ loss_scale ...................................... None
266
+ loss_scale_window ............................... 1000
267
+ lr .............................................. 0.0001
268
+ lr_decay_iters .................................. 8
269
+ lr_decay_samples ................................ None
270
+ lr_decay_style .................................. cosine
271
+ lr_warmup_fraction .............................. None
272
+ lr_warmup_init .................................. 0.0
273
+ lr_warmup_iters ................................. 5
274
+ lr_warmup_samples ............................... 0
275
+ lr_wsd_decay_iters .............................. None
276
+ lr_wsd_decay_samples ............................ None
277
+ lr_wsd_decay_style .............................. exponential
278
+ main_grads_dtype ................................ torch.float32
279
+ main_params_dtype ............................... torch.float32
280
+ make_vocab_size_divisible_by .................... 128
281
+ mamba_head_dim .................................. 64
282
+ mamba_num_groups ................................ 8
283
+ mamba_num_heads ................................. None
284
+ mamba_state_dim ................................. 128
285
+ manual_gc ....................................... False
286
+ manual_gc_eval .................................. True
287
+ manual_gc_interval .............................. 0
288
+ mask_factor ..................................... 1.0
289
+ mask_prob ....................................... 0.15
290
+ mask_type ....................................... random
291
+ masked_softmax_fusion ........................... False
292
+ max_position_embeddings ......................... 4096
293
+ max_tokens_to_oom ............................... 12000
294
+ memory_snapshot_path ............................ snapshot.pickle
295
+ merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt
296
+ micro_batch_size ................................ 1
297
+ microbatch_group_size_per_vp_stage .............. None
298
+ mid_level_dataset_surplus ....................... 0.005
299
+ min_loss_scale .................................. 1.0
300
+ min_lr .......................................... 1e-05
301
+ mlp_chunks_for_prefill .......................... 1
302
+ mmap_bin_files .................................. True
303
+ mock_data ....................................... False
304
+ moe_apply_probs_on_input ........................ False
305
+ moe_aux_loss_coeff .............................. 0.01
306
+ moe_deepep_num_sms .............................. 20
307
+ moe_echo_enable_random_offloading ............... False
308
+ moe_echo_expert_dispatch_overlap ................ False
309
+ moe_echo_expert_dispatcher_type ................. hybridep
310
+ moe_echo_recompute_expert_dispatch .............. False
311
+ moe_enable_deepep ............................... False
312
+ moe_enable_echo ................................. False
313
+ moe_enable_hybridep ............................. False
314
+ moe_enable_moonep ............................... True
315
+ moe_expert_capacity_factor ...................... None
316
+ moe_extended_tp ................................. False
317
+ moe_ffn_hidden_size ............................. 1024
318
+ moe_grouped_gemm ................................ True
319
+ moe_hybridep_num_sms ............................ 16
320
+ moe_input_jitter_eps ............................ None
321
+ moe_layer_freq .................................. 1
322
+ moe_layer_recompute ............................. False
323
+ moe_moonep_async_grad_reduce .................... True
324
+ moe_moonep_num_sms .............................. None
325
+ moe_num_echo_experts ............................ None
326
+ moe_pad_expert_input_to_capacity ................ False
327
+ moe_per_layer_logging ........................... False
328
+ moe_permute_fusion .............................. False
329
+ moe_received_token_capacity ..................... None
330
+ moe_router_bias_update_rate ..................... 0.001
331
+ moe_router_dtype ................................ fp32
332
+ moe_router_enable_expert_bias ................... False
333
+ moe_router_force_load_balancing ................. False
334
+ moe_router_group_topk ........................... None
335
+ moe_router_load_balancing_type .................. aux_loss
336
+ moe_router_num_groups ........................... None
337
+ moe_router_padding_for_fp8 ...................... False
338
+ moe_router_pre_softmax .......................... True
339
+ moe_router_score_function ....................... softmax
340
+ moe_router_topk ................................. 8
341
+ moe_router_topk_scaling_factor .................. None
342
+ moe_shared_expert_intermediate_size ............. None
343
+ moe_shared_expert_overlap ....................... False
344
+ moe_token_dispatcher_type ....................... flex
345
+ moe_token_drop_policy ........................... probs
346
+ moe_upcycling_granularity ....................... 1
347
+ moe_use_device_initiated_grouped_gemm ........... True
348
+ moe_use_legacy_grouped_gemm ..................... False
349
+ moe_use_upcycling ............................... False
350
+ moe_z_loss_coeff ................................ None
351
+ mrope_section ................................... None
352
+ mscale .......................................... 1.0
353
+ mscale_all_dim .................................. 0.0
354
+ mtp_loss_scaling_factor ......................... 0.1
355
+ mtp_num_layers .................................. None
356
+ multi_latent_attention .......................... False
357
+ nccl_all_reduce_for_prefill ..................... False
358
+ nccl_communicator_config_path ................... None
359
+ nccl_ub ......................................... False
360
+ no_load_optim ................................... None
361
+ no_load_rng ..................................... None
362
+ no_persist_layer_norm ........................... False
363
+ no_rope_freq .................................... None
364
+ no_save_optim ................................... None
365
+ no_save_rng ..................................... None
366
+ non_persistent_ckpt_type ........................ None
367
+ non_persistent_global_ckpt_dir .................. None
368
+ non_persistent_local_ckpt_algo .................. fully_parallel
369
+ non_persistent_local_ckpt_dir ................... None
370
+ non_persistent_save_interval .................... None
371
+ norm_epsilon .................................... 1e-05
372
+ normalization ................................... RMSNorm
373
+ num_attention_heads ............................. 16
374
+ num_channels .................................... 3
375
+ num_classes ..................................... 1000
376
+ num_dataset_builder_threads ..................... 1
377
+ num_distributed_optimizer_instances ............. 1
378
+ num_experts ..................................... 64
379
+ num_layers ...................................... 16
380
+ num_layers_at_end_in_bf16 ....................... 1
381
+ num_layers_at_start_in_bf16 ..................... 1
382
+ num_layers_per_virtual_pipeline_stage ........... None
383
+ num_query_groups ................................ 16
384
+ num_virtual_stages_per_pipeline_rank ............ None
385
+ num_workers ..................................... 2
386
+ object_storage_cache_path ....................... None
387
+ one_logger_async ................................ False
388
+ one_logger_project .............................. megatron-lm
389
+ one_logger_run_name ............................. None
390
+ onnx_safe ....................................... None
391
+ openai_gelu ..................................... False
392
+ optimizer ....................................... adam
393
+ optimizer_cpu_offload ........................... False
394
+ optimizer_offload_fraction ...................... 1.0
395
+ output_bert_embeddings .......................... False
396
+ overlap_cpu_optimizer_d2h_h2d ................... False
397
+ overlap_grad_reduce ............................. True
398
+ overlap_moe_expert_parallel_comm ................ False
399
+ overlap_p2p_comm ................................ False
400
+ overlap_p2p_comm_warmup_flush ................... False
401
+ overlap_param_gather ............................ True
402
+ overlap_param_gather_with_optimizer_step ........ False
403
+ override_opt_param_scheduler .................... False
404
+ params_dtype .................................... torch.bfloat16
405
+ patch_dim ....................................... 16
406
+ per_split_data_args_path ........................ None
407
+ perform_initialization .......................... True
408
+ pin_cpu_grads ................................... True
409
+ pin_cpu_params .................................. True
410
+ pipeline_model_parallel_comm_backend ............ None
411
+ pipeline_model_parallel_layout .................. None
412
+ pipeline_model_parallel_size .................... 1
413
+ position_embedding_type ......................... rope
414
+ pretrained_checkpoint ........................... None
415
+ profile ......................................... False
416
+ profile_ranks ................................... [0]
417
+ profile_step_end ................................ 12
418
+ profile_step_start .............................. 10
419
+ q_lora_rank ..................................... None
420
+ qk_head_dim ..................................... 128
421
+ qk_l2_norm ...................................... False
422
+ qk_layernorm .................................... False
423
+ qk_pos_emb_head_dim ............................. 64
424
+ query_in_block_prob ............................. 0.1
425
+ rampup_batch_size ............................... None
426
+ rank ............................................ 0
427
+ recompute_granularity ........................... None
428
+ recompute_method ................................ None
429
+ recompute_modules ............................... None
430
+ recompute_num_layers ............................ None
431
+ record_memory_history ........................... False
432
+ relative_attention_max_distance ................. 128
433
+ relative_attention_num_buckets .................. 32
434
+ replication ..................................... False
435
+ replication_factor .............................. 2
436
+ replication_jump ................................ None
437
+ rerun_mode ...................................... validate_results
438
+ reset_attention_mask ............................ False
439
+ reset_position_ids .............................. False
440
+ result_rejected_tracker_filename ................ None
441
+ retriever_report_topk_accuracies ................ []
442
+ retriever_score_scaling ......................... False
443
+ retriever_seq_length ............................ 256
444
+ retro_add_retriever ............................. False
445
+ retro_attention_gate ............................ 1
446
+ retro_cyclic_train_iters ........................ None
447
+ retro_encoder_attention_dropout ................. 0.1
448
+ retro_encoder_hidden_dropout .................... 0.1
449
+ retro_encoder_layers ............................ 2
450
+ retro_num_neighbors ............................. 2
451
+ retro_num_retrieved_chunks ...................... 2
452
+ retro_project_dir ............................... None
453
+ retro_verify_neighbor_count ..................... True
454
+ reuse_grad_buf_for_mxfp8_param_ag ............... False
455
+ rope_scaling_factor ............................. 8.0
456
+ rope_type ....................................... None
457
+ rotary_base ..................................... 10000
458
+ rotary_interleaved .............................. False
459
+ rotary_percent .................................. 1.0
460
+ rotary_scaling_factor ........................... 1.0
461
+ rotary_seq_len_interpolation_factor ............. None
462
+ run_workload_inspector_server ................... False
463
+ sample_rate ..................................... 1.0
464
+ save ............................................ None
465
+ save_interval ................................... None
466
+ save_retain_interval ............................ None
467
+ scatter_gather_tensors_in_pipeline .............. True
468
+ seed ............................................ 1234
469
+ seq_length ...................................... 4096
470
+ sequence_parallel ............................... False
471
+ sft ............................................. False
472
+ sft_tokenizer_prompt_format ..................... nemotron-h-aligned
473
+ sgd_momentum .................................... 0.9
474
+ short_seq_prob .................................. 0.1
475
+ skip_train ...................................... False
476
+ skipped_train_samples ........................... 0
477
+ spec ............................................ None
478
+ split ........................................... 100,0,0
479
+ squared_relu .................................... False
480
+ start_weight_decay .............................. 0.1
481
+ straggler_ctrlr_port ............................ 65535
482
+ straggler_minmax_count .......................... 1
483
+ suggested_communication_unit_size ............... None
484
+ swiglu .......................................... True
485
+ swin_backbone_type .............................. tiny
486
+ symmetric_ar_type ............................... None
487
+ te_rng_tracker .................................. True
488
+ tensor_model_parallel_size ...................... 1
489
+ tensorboard_dir ................................. None
490
+ tensorboard_log_interval ........................ 1
491
+ tensorboard_queue_size .......................... 1000
492
+ test_data_path .................................. None
493
+ test_mode ....................................... False
494
+ tiktoken_num_special_tokens ..................... 1000
495
+ tiktoken_pattern ................................ None
496
+ tiktoken_special_tokens ......................... None
497
+ timing_log_level ................................ 0
498
+ timing_log_option ............................... minmax
499
+ titles_data_path ................................ None
500
+ tokenizer_model ................................. None
501
+ tokenizer_type .................................. GPT2BPETokenizer
502
+ torch_fsdp2_reshard_after_forward ............... True
503
+ tp_comm_bootstrap_backend ....................... nccl
504
+ tp_comm_bulk_dgrad .............................. True
505
+ tp_comm_bulk_wgrad .............................. True
506
+ tp_comm_overlap ................................. False
507
+ tp_comm_overlap_ag .............................. True
508
+ tp_comm_overlap_cfg ............................. None
509
+ tp_comm_overlap_rs .............................. True
510
+ tp_comm_overlap_rs_dgrad ........................ False
511
+ tp_comm_split_ag ................................ True
512
+ tp_comm_split_rs ................................ True
513
+ train_data_path ................................. None
514
+ train_iters ..................................... 12
515
+ train_samples ................................... None
516
+ train_sync_interval ............................. None
517
+ transformer_impl ................................ transformer_engine
518
+ transformer_pipeline_model_parallel_size ........ 1
519
+ untie_embeddings_and_output_weights ............. True
520
+ use_checkpoint_args ............................. False
521
+ use_checkpoint_opt_param_scheduler .............. False
522
+ use_cpu_initialization .......................... None
523
+ use_custom_fsdp ................................. False
524
+ use_dist_ckpt ................................... True
525
+ use_dist_ckpt_deprecated ........................ False
526
+ use_distributed_optimizer ....................... True
527
+ use_flash_attn .................................. False
528
+ use_legacy_models ............................... False
529
+ use_mp_args_from_checkpoint_args ................ False
530
+ use_one_sent_docs ............................... False
531
+ use_persistent_ckpt_worker ...................... False
532
+ use_precision_aware_optimizer ................... False
533
+ use_pytorch_profiler ............................ False
534
+ use_ring_exchange_p2p ........................... False
535
+ use_rope_scaling ................................ False
536
+ use_rotary_position_embeddings .................. False
537
+ use_sharp ....................................... False
538
+ use_tokenizer_model_from_checkpoint_args ........ True
539
+ use_torch_fsdp2 ................................. False
540
+ use_torch_optimizer_for_cpu_offload ............. False
541
+ use_tp_pp_dp_mapping ............................ False
542
+ v_head_dim ...................................... 128
543
+ valid_data_path ................................. None
544
+ variable_seq_lengths ............................ False
545
+ virtual_pipeline_model_parallel_size ............ None
546
+ vision_backbone_type ............................ vit
547
+ vision_pretraining .............................. False
548
+ vision_pretraining_type ......................... classify
549
+ vocab_extra_ids ................................. 0
550
+ vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json
551
+ vocab_size ...................................... None
552
+ wandb_exp_name ..................................
553
+ wandb_project ...................................
554
+ wandb_save_dir ..................................
555
+ weight_decay .................................... 0.1
556
+ weight_decay_incr_style ......................... constant
557
+ wgrad_deferral_limit ............................ 0
558
+ world_size ...................................... 8
559
+ yaml_cfg ........................................ None
560
+ -------------------- end of arguments ---------------------
561
+ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 1
562
+ > building GPT2BPETokenizer tokenizer ...
563
+ > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
564
+ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS
565
+ > initializing torch distributed ...
566
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
567
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
568
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
569
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
570
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
571
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
572
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
573
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
574
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
575
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
576
+ 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
577
+ /workspace/Megatron-MoE-Echo/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled
578
+ warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled")
579
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
580
+ [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
581
+ [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
582
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
583
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
584
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
585
+ [Gloo] Rank [Gloo] Rank 5 is connected to 76 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks.
586
+ Expected number of connected peer ranks is : 7
587
+ [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
588
+ [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
589
+ [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank
590
+ 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : [Gloo] Rank 71
591
+ is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
592
+ [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7
593
+ [Gloo] Rank [Gloo] Rank 7 is connected to 67 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is :
594
+ 7
595
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
596
+ > initialized tensor model parallel with size 1
597
+ > initialized pipeline model parallel with size 1
598
+ > setting random seeds to 1234 ...
599
+ [Gloo] Rank [Gloo] Rank 00 is connected to is connected to 00 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 00
600
+
601
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
602
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
603
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
604
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
605
+ [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
606
+ [rank6]:[W921 08:43:10.468942155 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.
607
+ [rank4]:[W921 08:43:10.468967190 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
608
+ [rank3]:[W921 08:43:10.469064243 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
609
+ [rank7]:[W921 08:43:10.469538235 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
610
+ [rank5]:[W921 08:43:10.470979099 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
611
+ [rank1]:[W921 08:43:10.471069283 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
612
+ [rank2]:[W921 08:43:10.472257569 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.
613
+ > compiling dataset index builder ...
614
+ make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
615
+ make: Nothing to be done for 'default'.
616
+ make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets'
617
+ >>> done with dataset index builder. Compilation time: 0.035 seconds
618
+ WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations.
619
+ > compiling and loading fused kernels ...
620
+ [rank0]:[W921 08:43:10.630240739 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device.
621
+ >>> done with compiling and loading fused kernels. Compilation time: 3.860 seconds
622
+ time to initialize megatron (seconds): 7.558
623
+ [after megatron is initialized] datetime: 2026-09-21 08:43:16
624
+ building GPT model ...
625
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
626
+ warnings.warn(
627
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
628
+ warnings.warn(
629
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
630
+ warnings.warn(
631
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
632
+ warnings.warn(
633
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
634
+ warnings.warn(
635
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
636
+ warnings.warn(
637
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
638
+ warnings.warn(
639
+ /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect.
640
+ warnings.warn(
641
+ > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 1281951744
642
+ INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False)
643
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 7
644
+ Params for bucket 1 (103022592 elements, 103022592 padded size):
645
+ module.output_layer.weight
646
+ Params for bucket 2 (46542848 elements, 46542848 padded size):
647
+ module.decoder.final_layernorm.weight
648
+ module.decoder.layers.15.self_attention.linear_qkv.weight
649
+ module.decoder.layers.15.pre_mlp_layernorm.weight
650
+ module.decoder.layers.14.mlp.router.weight
651
+ module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight
652
+ module.decoder.layers.14.self_attention.linear_proj.weight
653
+ module.decoder.layers.13.pre_mlp_layernorm.weight
654
+ module.decoder.layers.13.self_attention.linear_qkv.weight
655
+ module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight
656
+ module.decoder.layers.15.mlp.router.weight
657
+ module.decoder.layers.15.self_attention.linear_proj.weight
658
+ module.decoder.layers.14.pre_mlp_layernorm.weight
659
+ module.decoder.layers.14.self_attention.linear_qkv.weight
660
+ module.decoder.layers.13.mlp.router.weight
661
+ Params for bucket 3 (50737152 elements, 50737152 padded size):
662
+ module.decoder.layers.12.mlp.router.weight
663
+ module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight
664
+ module.decoder.layers.12.self_attention.linear_proj.weight
665
+ module.decoder.layers.11.mlp.router.weight
666
+ module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight
667
+ module.decoder.layers.11.self_attention.linear_proj.weight
668
+ module.decoder.layers.10.pre_mlp_layernorm.weight
669
+ module.decoder.layers.10.self_attention.linear_qkv.weight
670
+ module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight
671
+ module.decoder.layers.12.pre_mlp_layernorm.weight
672
+ module.decoder.layers.12.self_attention.linear_qkv.weight
673
+ module.decoder.layers.13.self_attention.linear_proj.weight
674
+ module.decoder.layers.11.pre_mlp_layernorm.weight
675
+ module.decoder.layers.11.self_attention.linear_qkv.weight
676
+ module.decoder.layers.10.mlp.router.weight
677
+ Params for bucket 4 (50737152 elements, 50737152 padded size):
678
+ module.decoder.layers.9.pre_mlp_layernorm.weight
679
+ module.decoder.layers.9.self_attention.linear_qkv.weight
680
+ module.decoder.layers.8.pre_mlp_layernorm.weight
681
+ module.decoder.layers.8.self_attention.linear_qkv.weight
682
+ module.decoder.layers.7.pre_mlp_layernorm.weight
683
+ module.decoder.layers.7.self_attention.linear_qkv.weight
684
+ module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight
685
+ module.decoder.layers.10.self_attention.linear_proj.weight
686
+ module.decoder.layers.9.mlp.router.weight
687
+ module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight
688
+ module.decoder.layers.9.self_attention.linear_proj.weight
689
+ module.decoder.layers.8.mlp.router.weight
690
+ module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight
691
+ module.decoder.layers.8.self_attention.linear_proj.weight
692
+ module.decoder.layers.7.mlp.router.weight
693
+ Params for bucket 5 (50737152 elements, 50737152 padded size):
694
+ module.decoder.layers.6.mlp.router.weight
695
+ module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight
696
+ module.decoder.layers.6.self_attention.linear_proj.weight
697
+ module.decoder.layers.5.mlp.router.weight
698
+ module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight
699
+ module.decoder.layers.4.pre_mlp_layernorm.weight
700
+ module.decoder.layers.4.self_attention.linear_qkv.weight
701
+ module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight
702
+ module.decoder.layers.6.self_attention.linear_qkv.weight
703
+ module.decoder.layers.7.self_attention.linear_proj.weight
704
+ module.decoder.layers.6.pre_mlp_layernorm.weight
705
+ module.decoder.layers.5.pre_mlp_layernorm.weight
706
+ module.decoder.layers.5.self_attention.linear_qkv.weight
707
+ module.decoder.layers.5.self_attention.linear_proj.weight
708
+ module.decoder.layers.4.mlp.router.weight
709
+ Params for bucket 6 (50737152 elements, 50737152 padded size):
710
+ module.decoder.layers.3.mlp.router.weight
711
+ module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight
712
+ module.decoder.layers.3.self_attention.linear_proj.weight
713
+ module.decoder.layers.2.mlp.router.weight
714
+ module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight
715
+ module.decoder.layers.2.self_attention.linear_proj.weight
716
+ module.decoder.layers.1.mlp.router.weight
717
+ module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight
718
+ module.decoder.layers.4.self_attention.linear_proj.weight
719
+ module.decoder.layers.3.self_attention.linear_qkv.weight
720
+ module.decoder.layers.3.pre_mlp_layernorm.weight
721
+ module.decoder.layers.2.pre_mlp_layernorm.weight
722
+ module.decoder.layers.2.self_attention.linear_qkv.weight
723
+ module.decoder.layers.1.pre_mlp_layernorm.weight
724
+ module.decoder.layers.1.self_attention.linear_qkv.weight
725
+ Params for bucket 7 (124131328 elements, 124131328 padded size):
726
+ module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight
727
+ module.decoder.layers.1.self_attention.linear_proj.weight
728
+ module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight
729
+ module.decoder.layers.0.self_attention.linear_proj.weight
730
+ module.embedding.word_embeddings.weight
731
+ module.decoder.layers.0.mlp.router.weight
732
+ module.decoder.layers.0.self_attention.linear_qkv.weight
733
+ module.decoder.layers.0.pre_mlp_layernorm.weight
734
+ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 20
735
+ Params for bucket 1 (41943040 elements, 41943040 padded size):
736
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight0
737
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight2
738
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight4
739
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight6
740
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight0
741
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight2
742
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight4
743
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight3
744
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight1
745
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight5
746
+ module.decoder.layers.15.mlp.experts.linear_fc2.weight7
747
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight1
748
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight3
749
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight5
750
+ Params for bucket 2 (41943040 elements, 41943040 padded size):
751
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight6
752
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight2
753
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight0
754
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight4
755
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight6
756
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight1
757
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight3
758
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight1
759
+ module.decoder.layers.15.mlp.experts.linear_fc1.weight7
760
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight3
761
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight5
762
+ module.decoder.layers.14.mlp.experts.linear_fc2.weight7
763
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight0
764
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight2
765
+ Params for bucket 3 (41943040 elements, 41943040 padded size):
766
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight5
767
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight7
768
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight1
769
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight3
770
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight5
771
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight7
772
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight0
773
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight4
774
+ module.decoder.layers.14.mlp.experts.linear_fc1.weight6
775
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight0
776
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight2
777
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight4
778
+ module.decoder.layers.13.mlp.experts.linear_fc2.weight6
779
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight1
780
+ Params for bucket 4 (41943040 elements, 41943040 padded size):
781
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight2
782
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight6
783
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight4
784
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight0
785
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight2
786
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight4
787
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight6
788
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight5
789
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight3
790
+ module.decoder.layers.13.mlp.experts.linear_fc1.weight7
791
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight1
792
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight3
793
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight5
794
+ module.decoder.layers.12.mlp.experts.linear_fc2.weight7
795
+ Params for bucket 5 (41943040 elements, 41943040 padded size):
796
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight3
797
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight1
798
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight5
799
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight7
800
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight0
801
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight2
802
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight0
803
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight2
804
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight4
805
+ module.decoder.layers.12.mlp.experts.linear_fc1.weight6
806
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight1
807
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight3
808
+ Params for bucket 6 (41943040 elements, 41943040 padded size):
809
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight4
810
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight6
811
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight1
812
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight3
813
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight5
814
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight7
815
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight7
816
+ module.decoder.layers.11.mlp.experts.linear_fc2.weight5
817
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight0
818
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight2
819
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight4
820
+ module.decoder.layers.11.mlp.experts.linear_fc1.weight6
821
+ Params for bucket 7 (41943040 elements, 41943040 padded size):
822
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight1
823
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight3
824
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight5
825
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight7
826
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight0
827
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight2
828
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight4
829
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight0
830
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight2
831
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight4
832
+ module.decoder.layers.10.mlp.experts.linear_fc2.weight6
833
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight1
834
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight3
835
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight5
836
+ Params for bucket 8 (41943040 elements, 41943040 padded size):
837
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight6
838
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight1
839
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight3
840
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight5
841
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight7
842
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight0
843
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight2
844
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight0
845
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight2
846
+ module.decoder.layers.10.mlp.experts.linear_fc1.weight7
847
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight4
848
+ module.decoder.layers.9.mlp.experts.linear_fc2.weight6
849
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight1
850
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight3
851
+ Params for bucket 9 (41943040 elements, 41943040 padded size):
852
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight4
853
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight0
854
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight6
855
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight2
856
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight4
857
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight6
858
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight0
859
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight7
860
+ module.decoder.layers.9.mlp.experts.linear_fc1.weight5
861
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight1
862
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight3
863
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight5
864
+ module.decoder.layers.8.mlp.experts.linear_fc2.weight7
865
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight1
866
+ Params for bucket 10 (41943040 elements, 41943040 padded size):
867
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight2
868
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight6
869
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight4
870
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight1
871
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight3
872
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight5
873
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight7
874
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight3
875
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight5
876
+ module.decoder.layers.8.mlp.experts.linear_fc1.weight7
877
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight0
878
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight2
879
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight4
880
+ module.decoder.layers.7.mlp.experts.linear_fc2.weight6
881
+ Params for bucket 11 (41943040 elements, 41943040 padded size):
882
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight0
883
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight2
884
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight4
885
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight6
886
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight0
887
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight2
888
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight1
889
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight3
890
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight5
891
+ module.decoder.layers.7.mlp.experts.linear_fc1.weight7
892
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight1
893
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight3
894
+ Params for bucket 12 (41943040 elements, 41943040 padded size):
895
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight4
896
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight6
897
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight1
898
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight3
899
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight5
900
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight7
901
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight5
902
+ module.decoder.layers.6.mlp.experts.linear_fc2.weight7
903
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight0
904
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight2
905
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight4
906
+ module.decoder.layers.6.mlp.experts.linear_fc1.weight6
907
+ Params for bucket 13 (41943040 elements, 41943040 padded size):
908
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight3
909
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight1
910
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight5
911
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight7
912
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight0
913
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight2
914
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight4
915
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight0
916
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight2
917
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight4
918
+ module.decoder.layers.5.mlp.experts.linear_fc2.weight6
919
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight1
920
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight3
921
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight5
922
+ Params for bucket 14 (41943040 elements, 41943040 padded size):
923
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight6
924
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight0
925
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight2
926
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight4
927
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight6
928
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight0
929
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight2
930
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight1
931
+ module.decoder.layers.5.mlp.experts.linear_fc1.weight7
932
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight3
933
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight5
934
+ module.decoder.layers.4.mlp.experts.linear_fc2.weight7
935
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight1
936
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight3
937
+ Params for bucket 15 (41943040 elements, 41943040 padded size):
938
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight4
939
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight0
940
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight6
941
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight2
942
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight4
943
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight6
944
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight1
945
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight7
946
+ module.decoder.layers.4.mlp.experts.linear_fc1.weight5
947
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight1
948
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight3
949
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight5
950
+ module.decoder.layers.3.mlp.experts.linear_fc2.weight7
951
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight0
952
+ Params for bucket 16 (41943040 elements, 41943040 padded size):
953
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight3
954
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight5
955
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight7
956
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight1
957
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight3
958
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight5
959
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight7
960
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight2
961
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight6
962
+ module.decoder.layers.3.mlp.experts.linear_fc1.weight4
963
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight0
964
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight2
965
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight4
966
+ module.decoder.layers.2.mlp.experts.linear_fc2.weight6
967
+ Params for bucket 17 (41943040 elements, 41943040 padded size):
968
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight3
969
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight1
970
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight5
971
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight7
972
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight0
973
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight2
974
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight0
975
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight4
976
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight2
977
+ module.decoder.layers.2.mlp.experts.linear_fc1.weight6
978
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight1
979
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight3
980
+ Params for bucket 18 (41943040 elements, 41943040 padded size):
981
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight4
982
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight6
983
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight1
984
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight3
985
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight5
986
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight7
987
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight5
988
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight0
989
+ module.decoder.layers.1.mlp.experts.linear_fc2.weight7
990
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight2
991
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight4
992
+ module.decoder.layers.1.mlp.experts.linear_fc1.weight6
993
+ Params for bucket 19 (41943040 elements, 41943040 padded size):
994
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight1
995
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight5
996
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight7
997
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight2
998
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight4
999
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight5
1000
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight0
1001
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight2
1002
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight4
1003
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight3
1004
+ module.decoder.layers.0.mlp.experts.linear_fc2.weight6
1005
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight0
1006
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight1
1007
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight3
1008
+ Params for bucket 20 (8388608 elements, 8388608 padded size):
1009
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight6
1010
+ module.decoder.layers.0.mlp.experts.linear_fc1.weight7
1011
+ 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 0x7f918e8a9160>, config_logger_dir='')
1012
+ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine
1013
+ [after model, optimizer, and learning rate scheduler are built] datetime: 2026-09-21 08:43:19
1014
+ > building train, validation, and test datasets ...
1015
+ > datasets target sizes (minimum size):
1016
+ train: 96
1017
+ validation: 0
1018
+ test: 0
1019
+ INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None]
1020
+ > building train, validation, and test datasets for GPT ...
1021
+ INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(96, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=<megatron.training.tokenizer.tokenizer._GPT2BPETokenizer object at 0x7f919706b6b0>, 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)
1022
+ INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx
1023
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths
1024
+ INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers
1025
+ INFO:megatron.core.datasets.indexed_dataset: Extract the document indices
1026
+ INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029
1027
+ INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350
1028
+ INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices
1029
+ INFO:megatron.core.datasets.gpt_dataset: Load the document index from 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-document_index.npy
1030
+ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-sample_index.npy
1031
+ INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-shuffle_index.npy
1032
+ INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073
1033
+ > finished creating GPT datasets ...
1034
+ [after dataloaders are built] datetime: 2026-09-21 08:43:19
1035
+ done with setup ...
1036
+ training ...
1037
+ Overwriting rerun_state_machine.current_iteration from -1 to 0...
1038
+ (min, max) time across ranks (ms):
1039
+ model-and-optimizer-setup ......................: (2893.45, 2893.90)
1040
+ train/valid/test-data-iterators-setup ..........: (67.82, 70.60)
1041
+ [before the start of training step] datetime: 2026-09-21 08:43:19
1042
+ INFO:moonep.api:MoonEP comm buffer sizes:
1043
+ config : R=8, S=4096, H=2048, E=64, K=8, NvS=34800, NvS_padded=34816
1044
+ sm_count : num_sms=32, num_sms_dedup=148
1045
+ hidden_buffer_size : 136.00 MiB
1046
+ meta_buffer_size : 2.00 MiB
1047
+ local_temp_buffer_size : 1.26 MiB
1048
+ total_size : 139.26 MiB
1049
+ [MoonEP selfcheck 1/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [16512, 5632, 1152, 2944, 1920, 3840, 1024, 384, 128, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[44, -1, -1, -1, -1, -1, -1, -1], [18, -1, -1, -1, -1, -1, -1, -1], [52, -1, -1, -1, -1, -1, -1, -1], [44, -1, -1, -1, -1, -1, -1, -1], [27, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [27, -1, -1, -1, -1, -1, -1, -1], [19, -1, -1, -1, -1, -1, -1, -1]]
1050
+ [MoonEP selfcheck 2/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [256, 4736, 8960, 256, 256, 768, 8704, 384, 9088, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[27, -1, -1, -1, -1, -1, -1, -1], [29, -1, -1, -1, -1, -1, -1, -1], [31, -1, -1, -1, -1, -1, -1, -1], [39, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [28, -1, -1, -1, -1, -1, -1, -1], [31, -1, -1, -1, -1, -1, -1, -1], [27, -1, -1, -1, -1, -1, -1, -1]]
1051
+ [MoonEP selfcheck 3/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [128, 384, 128, 1280, 128, 128, 1408, 384, 27904, 1664, 0, 0, 0, 0, 0, 0] | duplicated experts [[9, 8, -1, -1, -1, -1, -1, -1], [61, -1, -1, -1, -1, -1, -1, -1], [60, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [27, -1, -1, -1, -1, -1, -1, -1], [27, -1, -1, -1, -1, -1, -1, -1], [60, -1, -1, -1, -1, -1, -1, -1], [27, -1, -1, -1, -1, -1, -1, -1]]
1052
+ [MoonEP selfcheck 4/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [128, 7296, 4352, 1408, 8320, 4864, 128, 6528, 256, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[9, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [42, -1, -1, -1, -1, -1, -1, -1], [52, -1, -1, -1, -1, -1, -1, -1], [1, -1, -1, -1, -1, -1, -1, -1], [34, -1, -1, -1, -1, -1, -1, -1], [44, -1, -1, -1, -1, -1, -1, -1], [46, -1, -1, -1, -1, -1, -1, -1]]
1053
+ [2026-09-21 08:43:39] iteration 1/ 12 | consumed samples: 8 | elapsed time per iteration (ms): 19754.4 | throughput per GPU (TFLOP/s/GPU): 1.6 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.118505E+01 | loss scale: 1.0 | grad norm: 35.488 | number of skipped iterations: 0 | number of nan iterations: 0 |
1054
+ Number of parameters in transformer block in billions: 6.71
1055
+ Number of parameters in embedding layers in billions: 0.21
1056
+ Total number of parameters in billions: 6.92
1057
+ Number of parameters in most loaded shard in billions: 6.9171
1058
+ Theoretical memory footprints: weight and optimizer=49474.72 MB
1059
+ [Rank 0] (after 1 iterations) memory (MB) | allocated: 16643.67431640625 | max allocated: 16999.85595703125 | reserved: 17484.0 | max reserved: 17484.0
1060
+ [2026-09-21 08:43:39] iteration 2/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 212.6 | throughput per GPU (TFLOP/s/GPU): 151.6 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.120947E+01 | loss scale: 1.0 | grad norm: 36.301 | number of skipped iterations: 0 | number of nan iterations: 0 |
1061
+ [2026-09-21 08:43:40] iteration 3/ 12 | consumed samples: 24 | elapsed time per iteration (ms): 202.3 | throughput per GPU (TFLOP/s/GPU): 159.3 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 1.006625E+01 | loss scale: 1.0 | grad norm: 101.414 | number of skipped iterations: 0 | number of nan iterations: 0 |
1062
+ Capture CUDA graph for training!!!
1063
+ CUDA graph capture done!!!
1064
+ [2026-09-21 08:43:40] iteration 4/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 680.7 | throughput per GPU (TFLOP/s/GPU): 47.3 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 1.023816E+01 | loss scale: 1.0 | grad norm: 28.877 | number of skipped iterations: 0 | number of nan iterations: 0 |
1065
+ [2026-09-21 08:43:40] iteration 5/ 12 | consumed samples: 40 | elapsed time per iteration (ms): 106.6 | throughput per GPU (TFLOP/s/GPU): 302.3 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 1.073935E+01 | loss scale: 1.0 | grad norm: 20.840 | number of skipped iterations: 0 | number of nan iterations: 0 |
1066
+ [2026-09-21 08:43:40] iteration 6/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 105.7 | throughput per GPU (TFLOP/s/GPU): 304.7 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 9.628673E+00 | loss scale: 1.0 | grad norm: 8.183 | number of skipped iterations: 0 | number of nan iterations: 0 |
1067
+ [2026-09-21 08:43:41] iteration 7/ 12 | consumed samples: 56 | elapsed time per iteration (ms): 105.7 | throughput per GPU (TFLOP/s/GPU): 304.8 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 9.080661E+00 | loss scale: 1.0 | grad norm: 16.450 | number of skipped iterations: 0 | number of nan iterations: 0 |
1068
+ [2026-09-21 08:43:41] iteration 8/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 105.7 | throughput per GPU (TFLOP/s/GPU): 304.8 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.818626E+00 | loss scale: 1.0 | grad norm: 8.619 | number of skipped iterations: 0 | number of nan iterations: 0 |
1069
+ [2026-09-21 08:43:41] iteration 9/ 12 | consumed samples: 72 | elapsed time per iteration (ms): 105.8 | throughput per GPU (TFLOP/s/GPU): 304.5 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.710756E+00 | loss scale: 1.0 | grad norm: 3.827 | number of skipped iterations: 0 | number of nan iterations: 0 |
1070
+ [2026-09-21 08:43:41] iteration 10/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 105.6 | throughput per GPU (TFLOP/s/GPU): 305.2 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.653936E+00 | loss scale: 1.0 | grad norm: 3.502 | number of skipped iterations: 0 | number of nan iterations: 0 |
1071
+ [2026-09-21 08:43:41] iteration 11/ 12 | consumed samples: 88 | elapsed time per iteration (ms): 105.7 | throughput per GPU (TFLOP/s/GPU): 304.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.642041E+00 | loss scale: 1.0 | grad norm: 3.475 | number of skipped iterations: 0 | number of nan iterations: 0 |
1072
+ [2026-09-21 08:43:41] iteration 12/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 105.5 | throughput per GPU (TFLOP/s/GPU): 305.5 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.591676E+00 | loss scale: 1.0 | grad norm: 3.174 | number of skipped iterations: 0 | number of nan iterations: 0 |
1073
+ [after training is done] datetime: 2026-09-21 08:43:41
1074
+ [rank0]:[W921 08:43:42.856866385 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1075
+ [rank4]:[W921 08:43:44.966062431 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1076
+ [rank2]:[W921 08:43:44.968857372 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1077
+ [rank6]:[W921 08:43:44.971597179 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1078
+ [rank3]:[W921 08:43:44.124049719 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1079
+ [rank5]:[W921 08:43:44.263522995 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
1080
+ W0921 08:43:44.354000 714978 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers
1081
+ W0921 08:43:44.355000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715335 closing signal SIGTERM
1082
+ W0921 08:43:44.355000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715337 closing signal SIGTERM
1083
+ W0921 08:43:44.357000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715339 closing signal SIGTERM
1084
+ W0921 08:43:44.358000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715340 closing signal SIGTERM
1085
+ W0921 08:43:44.359000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715341 closing signal SIGTERM
1086
+ W0921 08:43:44.360000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715342 closing signal SIGTERM
1087
+ W0921 08:43:44.361000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715343 closing signal SIGTERM
1088
+ W0921 08:43:44.362000 714978 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 715344 closing signal SIGTERM
1089
+ Traceback (most recent call last):
1090
+ File "/usr/local/bin/torchrun", line 8, in <module>
1091
+ sys.exit(main())
1092
+ ^^^^^^
1093
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
1094
+ return f(*args, **kwargs)
1095
+ ^^^^^^^^^^^^^^^^^^
1096
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
1097
+ run(args)
1098
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
1099
+ elastic_launch(
1100
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
1101
+ return launch_agent(self._config, self._entrypoint, list(args))
1102
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1103
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent
1104
+ result = agent.run()
1105
+ ^^^^^^^^^^^
1106
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper
1107
+ result = f(*args, **kwargs)
1108
+ ^^^^^^^^^^^^^^^^^^
1109
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run
1110
+ result = self._invoke_run(role)
1111
+ ^^^^^^^^^^^^^^^^^^^^^^
1112
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run
1113
+ time.sleep(monitor_interval)
1114
+ File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler
1115
+ raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval)
1116
+ torch.distributed.elastic.multiprocessing.api.SignalException: Process 714978 got signal: 15
granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/results.tsv CHANGED
@@ -1,3 +1,4 @@
1
  config backend policy slots seq gbs cap recompute rep status iter_ms_mean iter_ms_min peak_mib run_dir
2
  OLMOE_1B_7B_L16 moonep matched 8 4096 8 none none 1 OK 105.50 105.30 36429 /workspace/granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/OLMOE_1B_7B_L16__moonep__matched__seq4096__gbs8__capnone__async/rep1
3
  OLMOE_1B_7B_L16 moonep matched 8 4096 8 none none 2 OK 105.68 105.20 36429 /workspace/granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/OLMOE_1B_7B_L16__moonep__matched__seq4096__gbs8__capnone__async/rep2
 
 
1
  config backend policy slots seq gbs cap recompute rep status iter_ms_mean iter_ms_min peak_mib run_dir
2
  OLMOE_1B_7B_L16 moonep matched 8 4096 8 none none 1 OK 105.50 105.30 36429 /workspace/granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/OLMOE_1B_7B_L16__moonep__matched__seq4096__gbs8__capnone__async/rep1
3
  OLMOE_1B_7B_L16 moonep matched 8 4096 8 none none 2 OK 105.68 105.20 36429 /workspace/granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/OLMOE_1B_7B_L16__moonep__matched__seq4096__gbs8__capnone__async/rep2
4
+ OLMOE_1B_7B_L16 moonep matched 8 4096 8 none none 3 OK 105.68 105.20 36429 /workspace/granularity_next_b200/tierb_models/attempts/OLMOE_1B_7B_L16/moonep/tierb/tierb_gbs8/OLMOE_1B_7B_L16__moonep__matched__seq4096__gbs8__capnone__async/rep3