diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/mem_samples.csv b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/mem_samples.csv index 82c808260df02dc225dfb6a46e3b629624317689..125c131bb902596b420c116a7882e24d78c986ba 100644 --- a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/mem_samples.csv +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/mem_samples.csv @@ -222,3 +222,11 @@ 5, 108273 6, 108011 7, 108275 +0, 3 +1, 4 +2, 106983 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/train.log b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/train.log index d7dc7ee51a1c63f2348988ab62979bf7bde684d9..98707161942fcc6628e8bb133c7f37ee9c125c26 100644 --- a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/train.log +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1/train.log @@ -1274,3 +1274,40 @@ CUDA graph capture done!!! [rank4]:[W830 07:30:47.103262183 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()) [rank7]:[W830 07:30:48.261809486 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()) [rank1]:[W830 07:30:48.638553071 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()) +W0830 07:30:51.336000 766245 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0830 07:30:51.337000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766333 closing signal SIGTERM +W0830 07:30:51.337000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766334 closing signal SIGTERM +W0830 07:30:51.338000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766335 closing signal SIGTERM +W0830 07:30:51.339000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766336 closing signal SIGTERM +W0830 07:30:51.339000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766337 closing signal SIGTERM +W0830 07:30:51.340000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766338 closing signal SIGTERM +W0830 07:30:51.340000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766339 closing signal SIGTERM +W0830 07:30:51.341000 766245 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 766340 closing signal SIGTERM +Traceback (most recent call last): + File "/usr/local/bin/torchrun", line 8, in + sys.exit(main()) + ^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper + return f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main + run(args) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run + elastic_launch( + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ + return launch_agent(self._config, self._entrypoint, list(args)) + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent + result = agent.run() + ^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper + result = f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run + result = self._invoke_run(role) + ^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run + time.sleep(monitor_interval) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler + raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) +torch.distributed.elastic.multiprocessing.api.SignalException: Process 766245 got signal: 15 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..dc4ebb3dfd5bc19f2d7c5fb7581d40c771795ceb --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv @@ -0,0 +1,232 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 2 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 819 +3, 821 +4, 819 +5, 819 +6, 819 +7, 821 +0, 1537 +1, 1537 +2, 1537 +3, 1537 +4, 1537 +5, 1537 +6, 1537 +7, 1537 +0, 2231 +1, 2231 +2, 2231 +3, 2231 +4, 2231 +5, 2231 +6, 2231 +7, 2231 +0, 2903 +1, 2903 +2, 2903 +3, 2903 +4, 2903 +5, 2903 +6, 2903 +7, 2903 +0, 42073 +1, 42055 +2, 42065 +3, 42049 +4, 42037 +5, 42067 +6, 42077 +7, 42069 +0, 42803 +1, 42777 +2, 42795 +3, 42783 +4, 42783 +5, 42795 +6, 42835 +7, 42785 +0, 43579 +1, 43563 +2, 43563 +3, 43563 +4, 43563 +5, 43563 +6, 43563 +7, 43563 +0, 44129 +1, 44113 +2, 44113 +3, 44113 +4, 44113 +5, 44113 +6, 44113 +7, 44113 +0, 45975 +1, 45959 +2, 45959 +3, 45959 +4, 45959 +5, 45959 +6, 45959 +7, 45959 +0, 54885 +1, 55461 +2, 55461 +3, 55461 +4, 55503 +5, 55463 +6, 55465 +7, 55461 +0, 74073 +1, 74057 +2, 74057 +3, 74057 +4, 74097 +5, 74057 +6, 74057 +7, 74057 +0, 74075 +1, 74059 +2, 74059 +3, 74059 +4, 74099 +5, 74059 +6, 74059 +7, 74059 +0, 74075 +1, 74059 +2, 74059 +3, 74061 +4, 74099 +5, 74059 +6, 74061 +7, 74061 +0, 87105 +1, 87089 +2, 87089 +3, 87089 +4, 87169 +5, 87089 +6, 87089 +7, 87089 +0, 87635 +1, 87621 +2, 87613 +3, 87637 +4, 87727 +5, 87665 +6, 87657 +7, 87657 +0, 88783 +1, 88761 +2, 88763 +3, 88769 +4, 88887 +5, 88791 +6, 88809 +7, 88773 +0, 104465 +1, 104901 +2, 104103 +3, 105013 +4, 102297 +5, 104901 +6, 104733 +7, 104453 +0, 113951 +1, 113839 +2, 113837 +3, 113817 +4, 113933 +5, 113821 +6, 113837 +7, 113835 +0, 113951 +1, 113839 +2, 113837 +3, 113837 +4, 113933 +5, 113835 +6, 113837 +7, 113835 +0, 113951 +1, 113839 +2, 113837 +3, 113837 +4, 113933 +5, 113835 +6, 113837 +7, 113835 +0, 110457 +1, 110345 +2, 110343 +3, 110343 +4, 110439 +5, 110341 +6, 110343 +7, 110341 +0, 108777 +1, 108277 +2, 108663 +3, 108275 +4, 108371 +5, 108273 +6, 108275 +7, 108661 +0, 106709 +1, 4 +2, 106983 +3, 106595 +4, 107077 +5, 4 +6, 106981 +7, 106981 +0, 1 +1, 1 +2, 1 +3, 1 +4, 107074 +5, 1 +6, 1 +7, 3 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/run.env b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/run.env new file mode 100644 index 0000000000000000000000000000000000000000..4b2e571dee0cb48bce3df8aa30c5ad9c794cc0df --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/run.env @@ -0,0 +1,70 @@ +backend=ultraep +config=G3 +family=tierA +repo=/workspace/Megatron-UltraEP +branch=ultraep-bracing-regions-through-kernels +commit=03d57ef7 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=1792 +num_experts=64 +topk=16 +experts_per_rank=8 +expert_params=22020096 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=65536 +mean_tokens_per_expert=8192 +slot_policy=matched +slots_per_rank=8 +capacity_mode=static +capacity=1.072265625 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens=1024 +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/train.log b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/train.log new file mode 100644 index 0000000000000000000000000000000000000000..5b45be5b9110f52abfc28818f0591e55dc63305c --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/train.log @@ -0,0 +1,1312 @@ +============================================================== + G3 / ultraep (slots/rank=8 policy=matched) + E=64 K=16 H'=1792 m=8 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.072265625 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-UltraEP @ 03d57ef7 +============================================================== +W0830 07:30:56.845000 769356 torch/distributed/run.py:766] +W0830 07:30:56.845000 769356 torch/distributed/run.py:766] ***************************************** +W0830 07:30:56.845000 769356 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. +W0830 07:30:56.845000 769356 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +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 +WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer +Number of virtual stages per pipeline stage: None +accumulate and all-reduce gradients in fp32 for bfloat16 data type. +using torch.bfloat16 for parameters ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. True + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. False + add_position_embedding .......................... False + add_qkv_bias .................................... False + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... True + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.0 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ True + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ False + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full_iteration + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 8 + data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document'] + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + delay_wgrad_compute ............................. False + deprecated_use_mcore_models ..................... True + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_init_method_std ....................... None + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... True + enable_experimental ............................. False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 16 + encoder_seq_length .............................. 4096 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 1000000 + eval_iters ...................................... 0 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 8 + expert_tensor_parallel_size ..................... 1 + export_force_local_attention .................... False + export_kd_cfg ................................... None + export_kd_teacher_ckpt_format ................... None + export_kd_teacher_load .......................... None + export_kv_cache_quant ........................... False + export_legacy_megatron .......................... False + export_model_type ............................... GPTModel + export_moe_apply_probs_on_input ................. False + export_qk_l2_norm ............................... False + export_quant_cfg ................................ None + export_real_quant_cfg ........................... None + export_te_mcore_model ........................... False + external_cuda_graph ............................. False + ffn_hidden_size ................................. 14336 + finetune ........................................ False + finetune_data_split ............................. train + finetune_hf_dataset ............................. None + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ False + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. e4m3 + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... mxfp8 + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 16 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_dynamic_batching_num_cuda_graphs ...... 16 + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.01 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kitchen_config_file ............................. None + kitchen_recipe_number ........................... None + kv_channels ..................................... 128 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ None + load_main_params_from_ckpt ...................... None + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_energy ...................................... False + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. True + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... None + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0001 + lr_decay_iters .................................. 8 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 5 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... False + max_position_embeddings ......................... 32768 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 1e-05 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... False + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.01 + moe_capacity_monitor ............................ True + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ False + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. False + moe_enable_hybridep ............................. True + moe_enable_ultraep .............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 1792 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ None + moe_num_redundant_experts_per_rank .............. 8 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.072265625 + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ fp32 + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 16 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + moe_ultraep_max_microbatches .................... None + moe_ultraep_overlap_weight_sync ................. True + moe_ultraep_weight_sync_join .................... before_dispatch + moe_upcycling_granularity ....................... 1 + moe_use_device_initiated_grouped_gemm ........... True + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 0.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... RMSNorm + num_attention_heads ............................. 32 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 16 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 8 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. True + overlap_moe_expert_parallel_comm ................ False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ True + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.bfloat16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_layout .................. None + pipeline_model_parallel_size .................... 1 + position_embedding_type ......................... rope + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... validate_results + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + reuse_grad_buf_for_mxfp8_param_ag ............... False + rope_scaling_factor ............................. 8.0 + rope_type ....................................... None + rotary_base ..................................... 1000000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ None + save_interval ................................... None + save_retain_interval ............................ None + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 4096 + sequence_parallel ............................... False + sft ............................................. False + sft_tokenizer_prompt_format ..................... nemotron-h-aligned + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... 100,0,0 + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... True + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. True + tensor_model_parallel_size ...................... 1 + tensorboard_dir ................................. None + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 12 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. True + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... True + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2 +> building GPT2BPETokenizer tokenizer ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") + > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304) +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS +> initializing torch distributed ... +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 +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank [Gloo] Rank 1 is connected to 73 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks. +Expected number of connected peer ranks is : 7 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank +6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : [Gloo] Rank 71 + is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +> initialized tensor model parallel with size 1 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[rank5]:[W830 07:31:14.505048609 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. +[rank2]:[W830 07:31:14.505066484 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. +[rank1]:[W830 07:31:14.505100898 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. +[rank4]:[W830 07:31:14.505110555 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. +[rank3]:[W830 07:31:14.505202825 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. +[rank7]:[W830 07:31:14.505214490 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. +[rank6]:[W830 07:31:14.514790650 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.036 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +[rank0]:[W830 07:31:14.690214575 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. +>>> done with compiling and loading fused kernels. Compilation time: 3.679 seconds +time to initialize megatron (seconds): 6.890 +[after megatron is initialized] datetime: 2026-08-30 07:31:20 +building GPT model ... +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +[INFO] Use MNNVL fabric: False +[WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8. +NVSHMEM v3.7.2 +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive). + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6724653056 +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) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18 +Params for bucket 1 (206045184 elements, 206045184 padded size): + module.output_layer.weight +Params for bucket 2 (42217472 elements, 42217472 padded size): + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.final_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_proj.weight +Params for bucket 3 (42213376 elements, 42213376 padded size): + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_qkv.weight +Params for bucket 4 (42213376 elements, 42213376 padded size): + module.decoder.layers.13.self_attention.linear_qkv.weight + module.decoder.layers.13.pre_mlp_layernorm.weight + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.13.self_attention.linear_proj.weight +Params for bucket 5 (42213376 elements, 42213376 padded size): + module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.12.self_attention.linear_proj.weight +Params for bucket 6 (42213376 elements, 42213376 padded size): + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_qkv.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight +Params for bucket 7 (42213376 elements, 42213376 padded size): + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.pre_mlp_layernorm.weight +Params for bucket 8 (42213376 elements, 42213376 padded size): + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_qkv.weight +Params for bucket 9 (42213376 elements, 42213376 padded size): + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.8.pre_mlp_layernorm.weight +Params for bucket 10 (42213376 elements, 42213376 padded size): + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.7.self_attention.linear_qkv.weight + module.decoder.layers.7.pre_mlp_layernorm.weight +Params for bucket 11 (42213376 elements, 42213376 padded size): + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_qkv.weight +Params for bucket 12 (42213376 elements, 42213376 padded size): + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight +Params for bucket 13 (42213376 elements, 42213376 padded size): + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.pre_mlp_layernorm.weight +Params for bucket 14 (42213376 elements, 42213376 padded size): + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight +Params for bucket 15 (42213376 elements, 42213376 padded size): + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_qkv.weight + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight +Params for bucket 16 (42213376 elements, 42213376 padded size): + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.self_attention.linear_proj.weight +Params for bucket 17 (42213376 elements, 42213376 padded size): + module.decoder.layers.0.mlp.router.weight + module.decoder.layers.0.pre_mlp_layernorm.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_proj.weight +Params for bucket 18 (206045184 elements, 206045184 padded size): + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 64 +Params for bucket 1 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight3 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 2 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 +Params for bucket 4 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 +Params for bucket 5 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 +Params for bucket 6 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 +Params for bucket 9 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight5 + module.decoder.layers.13.mlp.experts.linear_fc2.weight3 +Params for bucket 10 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight1 + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight5 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight3 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 +Params for bucket 13 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 + module.decoder.layers.12.mlp.experts.linear_fc2.weight4 + module.decoder.layers.12.mlp.experts.linear_fc2.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight3 + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 +Params for bucket 14 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight1 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight0 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 +Params for bucket 17 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 +Params for bucket 18 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 +Params for bucket 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 +Params for bucket 21 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight4 + module.decoder.layers.10.mlp.experts.linear_fc2.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 +Params for bucket 22 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight5 + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 + module.decoder.layers.10.mlp.experts.linear_fc1.weight1 +Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 +Params for bucket 26 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 +Params for bucket 27 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 +Params for bucket 29 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight2 +Params for bucket 30 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 +Params for bucket 31 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + module.decoder.layers.7.mlp.experts.linear_fc2.weight3 + module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 + module.decoder.layers.7.mlp.experts.linear_fc2.weight2 +Params for bucket 34 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight1 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc2.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 +Params for bucket 35 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight5 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 +Params for bucket 38 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight1 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 +Params for bucket 39 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 +Params for bucket 42 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 +Params for bucket 43 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 +Params for bucket 46 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 +Params for bucket 47 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 +Params for bucket 51 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight5 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight3 +Params for bucket 52 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 + module.decoder.layers.2.mlp.experts.linear_fc2.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 +Params for bucket 54 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 +Params for bucket 55 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 +Params for bucket 56 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 +Params for bucket 59 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 +Params for bucket 60 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 +Params for bucket 63 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 +Params for bucket 64 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight2 + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight1 +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=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +[after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:31:23 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + validation: 0 + test: 0 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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) +INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers +INFO:megatron.core.datasets.indexed_dataset: Extract the document indices +INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029 +INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350 +INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices +INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-30 07:31:23 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (3754.69, 3760.61) + train/valid/test-data-iterators-setup ..........: (71.74, 91.94) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 07:31:24 +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") + [2026-08-30 07:31:50] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26859.0 | throughput per GPU (TFLOP/s/GPU): 12.4 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102352E+01 | loss scale: 1.0 | grad norm: 44.192 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67072 | min safe capacity: 1.0234 | +Number of parameters in transformer block in billions: 23.22 +Number of parameters in embedding layers in billions: 0.41 +Total number of parameters in billions: 23.63 +Number of parameters in most loaded shard in billions: 23.6320 +Theoretical memory footprints: weight and optimizer=169029.43 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.49609375 | max allocated: 76960.7119140625 | reserved: 79598.0 | max reserved: 79598.0 + [2026-08-30 07:31:51] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 1047.0 | throughput per GPU (TFLOP/s/GPU): 318.4 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101566E+01 | loss scale: 1.0 | grad norm: 45.944 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:52] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 505.3 | throughput per GPU (TFLOP/s/GPU): 659.8 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.051973E+01 | loss scale: 1.0 | grad norm: 215.084 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 07:31:55] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 3210.3 | throughput per GPU (TFLOP/s/GPU): 103.8 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.316053E+01 | loss scale: 1.0 | grad norm: 38.032 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:56] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 446.8 | throughput per GPU (TFLOP/s/GPU): 746.2 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.038130E+01 | loss scale: 1.0 | grad norm: 16.230 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:56] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 465.8 | throughput per GPU (TFLOP/s/GPU): 715.7 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.038030E+01 | loss scale: 1.0 | grad norm: 32.702 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:57] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 464.0 | throughput per GPU (TFLOP/s/GPU): 718.5 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.004398E+01 | loss scale: 1.0 | grad norm: 7.313 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:57] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 453.0 | throughput per GPU (TFLOP/s/GPU): 736.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.518322E+00 | loss scale: 1.0 | grad norm: 5.761 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:57] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 462.8 | throughput per GPU (TFLOP/s/GPU): 720.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.275056E+00 | loss scale: 1.0 | grad norm: 5.596 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:58] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 451.1 | throughput per GPU (TFLOP/s/GPU): 739.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.113712E+00 | loss scale: 1.0 | grad norm: 5.426 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:58] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 455.9 | throughput per GPU (TFLOP/s/GPU): 731.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.995015E+00 | loss scale: 1.0 | grad norm: 5.228 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:31:59] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 460.3 | throughput per GPU (TFLOP/s/GPU): 724.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.873810E+00 | loss scale: 1.0 | grad norm: 5.305 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | +[after training is done] datetime: 2026-08-30 07:31:59 +[rank0]:[W830 07:32:00.773086397 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()) +[rank5]:[W830 07:32:01.861411069 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()) +[rank4]:[W830 07:32:01.868168097 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()) +[rank1]:[W830 07:32:01.891940151 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()) +[rank3]:[W830 07:32:01.892041227 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()) +[rank2]:[W830 07:32:01.892300631 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()) +[rank6]:[W830 07:32:02.334670136 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()) +[rank7]:[W830 07:32:02.549359231 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()) +W0830 07:32:05.530000 769356 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0830 07:32:05.531000 769356 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 769444 closing signal SIGTERM +W0830 07:32:05.531000 769356 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 769445 closing signal SIGTERM +W0830 07:32:05.532000 769356 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 769446 closing signal SIGTERM +W0830 07:32:05.532000 769356 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 769447 closing signal SIGTERM +W0830 07:32:05.533000 769356 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 769448 closing signal SIGTERM +W0830 07:32:05.534000 769356 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 769450 closing signal SIGTERM +W0830 07:32:05.534000 769356 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 769451 closing signal SIGTERM +Traceback (most recent call last): + File "/usr/local/bin/torchrun", line 8, in + sys.exit(main()) + ^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper + return f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main + run(args) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run + elastic_launch( + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ + return launch_agent(self._config, self._entrypoint, list(args)) + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent + result = agent.run() + ^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper + result = f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run + result = self._invoke_run(role) + ^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run + time.sleep(monitor_interval) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler + raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) +torch.distributed.elastic.multiprocessing.api.SignalException: Process 769356 got signal: 15 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..943b574d2133bb76dfa4344c25722f5fdcc48ab9 --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv @@ -0,0 +1,232 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 1 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 819 +2, 821 +3, 819 +4, 821 +5, 819 +6, 819 +7, 821 +0, 1537 +1, 1537 +2, 1537 +3, 1537 +4, 1537 +5, 1537 +6, 1537 +7, 1537 +0, 2231 +1, 2231 +2, 2231 +3, 2231 +4, 2231 +5, 2231 +6, 2231 +7, 2231 +0, 2903 +1, 2903 +2, 2903 +3, 2903 +4, 2903 +5, 2903 +6, 2903 +7, 2903 +0, 42119 +1, 42081 +2, 42093 +3, 42113 +4, 42041 +5, 42101 +6, 42135 +7, 42117 +0, 42911 +1, 42895 +2, 42895 +3, 42895 +4, 42895 +5, 42895 +6, 42895 +7, 42895 +0, 43579 +1, 43563 +2, 43563 +3, 43563 +4, 43563 +5, 43563 +6, 43563 +7, 43563 +0, 44129 +1, 44113 +2, 44113 +3, 44113 +4, 44113 +5, 44113 +6, 44113 +7, 44113 +0, 45975 +1, 45959 +2, 45959 +3, 45959 +4, 45959 +5, 45959 +6, 45959 +7, 45959 +0, 56829 +1, 56833 +2, 56793 +3, 56733 +4, 56895 +5, 56833 +6, 56793 +7, 57425 +0, 74073 +1, 74057 +2, 74057 +3, 74057 +4, 74097 +5, 74057 +6, 74057 +7, 74057 +0, 74075 +1, 74059 +2, 74059 +3, 74059 +4, 74099 +5, 74059 +6, 74059 +7, 74059 +0, 74089 +1, 74061 +2, 74061 +3, 74061 +4, 74103 +5, 74073 +6, 74061 +7, 74065 +0, 87105 +1, 87089 +2, 87089 +3, 87089 +4, 87169 +5, 87089 +6, 87089 +7, 87089 +0, 87729 +1, 87733 +2, 87723 +3, 87735 +4, 87815 +5, 87735 +6, 87735 +7, 87735 +0, 88727 +1, 88723 +2, 88743 +3, 88745 +4, 88837 +5, 88767 +6, 88773 +7, 88783 +0, 104409 +1, 102899 +2, 102465 +3, 104943 +4, 104269 +5, 105195 +6, 104551 +7, 104775 +0, 113951 +1, 113837 +2, 113835 +3, 113837 +4, 113931 +5, 113837 +6, 113835 +7, 113837 +0, 113951 +1, 113837 +2, 113835 +3, 113837 +4, 113931 +5, 113837 +6, 113835 +7, 113837 +0, 113951 +1, 113837 +2, 113835 +3, 113837 +4, 113931 +5, 113837 +6, 113835 +7, 113837 +0, 110457 +1, 110343 +2, 110341 +3, 110343 +4, 110437 +5, 110343 +6, 110341 +7, 110343 +0, 108777 +1, 108663 +2, 108661 +3, 108663 +4, 108563 +5, 108663 +6, 108661 +7, 108663 +0, 4 +1, 106983 +2, 106989 +3, 107191 +4, 107127 +5, 107047 +6, 107627 +7, 107303 +0, 1 +1, 1 +2, 9 +3, 107188 +4, 51 +5, 1 +6, 1 +7, 3 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/run.env b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/run.env new file mode 100644 index 0000000000000000000000000000000000000000..4b2e571dee0cb48bce3df8aa30c5ad9c794cc0df --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/run.env @@ -0,0 +1,70 @@ +backend=ultraep +config=G3 +family=tierA +repo=/workspace/Megatron-UltraEP +branch=ultraep-bracing-regions-through-kernels +commit=03d57ef7 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=1792 +num_experts=64 +topk=16 +experts_per_rank=8 +expert_params=22020096 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=65536 +mean_tokens_per_expert=8192 +slot_policy=matched +slots_per_rank=8 +capacity_mode=static +capacity=1.072265625 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens=1024 +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/train.log b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e47ef6af635e15883beda060ae732081dca67ecd --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/train.log @@ -0,0 +1,1313 @@ +============================================================== + G3 / ultraep (slots/rank=8 policy=matched) + E=64 K=16 H'=1792 m=8 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.072265625 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-UltraEP @ 03d57ef7 +============================================================== +W0830 07:32:11.037000 772411 torch/distributed/run.py:766] +W0830 07:32:11.037000 772411 torch/distributed/run.py:766] ***************************************** +W0830 07:32:11.037000 772411 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. +W0830 07:32:11.037000 772411 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +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 +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 +WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer +Number of virtual stages per pipeline stage: None +accumulate and all-reduce gradients in fp32 for bfloat16 data type. +using torch.bfloat16 for parameters ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. True + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. False + add_position_embedding .......................... False + add_qkv_bias .................................... False + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... True + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.0 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ True + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ False + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full_iteration + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 8 + data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document'] + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + delay_wgrad_compute ............................. False + deprecated_use_mcore_models ..................... True + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_init_method_std ....................... None + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... True + enable_experimental ............................. False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 16 + encoder_seq_length .............................. 4096 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 1000000 + eval_iters ...................................... 0 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 8 + expert_tensor_parallel_size ..................... 1 + export_force_local_attention .................... False + export_kd_cfg ................................... None + export_kd_teacher_ckpt_format ................... None + export_kd_teacher_load .......................... None + export_kv_cache_quant ........................... False + export_legacy_megatron .......................... False + export_model_type ............................... GPTModel + export_moe_apply_probs_on_input ................. False + export_qk_l2_norm ............................... False + export_quant_cfg ................................ None + export_real_quant_cfg ........................... None + export_te_mcore_model ........................... False + external_cuda_graph ............................. False + ffn_hidden_size ................................. 14336 + finetune ........................................ False + finetune_data_split ............................. train + finetune_hf_dataset ............................. None + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ False + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. e4m3 + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... mxfp8 + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 16 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_dynamic_batching_num_cuda_graphs ...... 16 + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.01 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kitchen_config_file ............................. None + kitchen_recipe_number ........................... None + kv_channels ..................................... 128 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ None + load_main_params_from_ckpt ...................... None + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_energy ...................................... False + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. True + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... None + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0001 + lr_decay_iters .................................. 8 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 5 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... False + max_position_embeddings ......................... 32768 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 1e-05 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... False + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.01 + moe_capacity_monitor ............................ True + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ False + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. False + moe_enable_hybridep ............................. True + moe_enable_ultraep .............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 1792 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ None + moe_num_redundant_experts_per_rank .............. 8 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.072265625 + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ fp32 + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 16 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + moe_ultraep_max_microbatches .................... None + moe_ultraep_overlap_weight_sync ................. True + moe_ultraep_weight_sync_join .................... before_dispatch + moe_upcycling_granularity ....................... 1 + moe_use_device_initiated_grouped_gemm ........... True + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 0.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... RMSNorm + num_attention_heads ............................. 32 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 16 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 8 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. True + overlap_moe_expert_parallel_comm ................ False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ True + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.bfloat16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_layout .................. None + pipeline_model_parallel_size .................... 1 + position_embedding_type ......................... rope + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... validate_results + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + reuse_grad_buf_for_mxfp8_param_ag ............... False + rope_scaling_factor ............................. 8.0 + rope_type ....................................... None + rotary_base ..................................... 1000000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ None + save_interval ................................... None + save_retain_interval ............................ None + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 4096 + sequence_parallel ............................... False + sft ............................................. False + sft_tokenizer_prompt_format ..................... nemotron-h-aligned + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... 100,0,0 + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... True + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. True + tensor_model_parallel_size ...................... 1 + tensorboard_dir ................................. None + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 12 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. True + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... True + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2 +> building GPT2BPETokenizer tokenizer ... + > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304) +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS +> initializing torch distributed ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank +1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +> initialized tensor model parallel with size 1 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. [Gloo] Rank Expected number of connected peer ranks is : [Gloo] Rank 00 is connected to 0 +0 is connected to peer ranks. 0Expected number of connected peer ranks is : peer ranks. 0Expected number of connected peer ranks is : +0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[rank4]:[W830 07:32:28.082772351 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. +[rank2]:[W830 07:32:28.082839766 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. +[rank1]:[W830 07:32:28.082880868 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. +[rank3]:[W830 07:32:28.082932114 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. +[rank7]:[W830 07:32:28.082950377 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. +[rank6]:[W830 07:32:28.082954305 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. +[rank5]:[W830 07:32:28.082954221 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.036 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +[rank0]:[W830 07:32:29.244335488 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. +>>> done with compiling and loading fused kernels. Compilation time: 3.686 seconds +time to initialize megatron (seconds): 7.030 +[after megatron is initialized] datetime: 2026-08-30 07:32:34 +building GPT model ... +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +[INFO] Use MNNVL fabric: False +[WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8. +NVSHMEM v3.7.2 +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive). + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6724653056 +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) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18 +Params for bucket 1 (206045184 elements, 206045184 padded size): + module.output_layer.weight +Params for bucket 2 (42217472 elements, 42217472 padded size): + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.final_layernorm.weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.15.self_attention.linear_proj.weight +Params for bucket 3 (42213376 elements, 42213376 padded size): + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_qkv.weight + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight +Params for bucket 4 (42213376 elements, 42213376 padded size): + module.decoder.layers.13.self_attention.linear_qkv.weight + module.decoder.layers.13.pre_mlp_layernorm.weight + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight +Params for bucket 5 (42213376 elements, 42213376 padded size): + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight +Params for bucket 6 (42213376 elements, 42213376 padded size): + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_qkv.weight +Params for bucket 7 (42213376 elements, 42213376 padded size): + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.pre_mlp_layernorm.weight +Params for bucket 8 (42213376 elements, 42213376 padded size): + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight +Params for bucket 9 (42213376 elements, 42213376 padded size): + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.8.pre_mlp_layernorm.weight +Params for bucket 10 (42213376 elements, 42213376 padded size): + module.decoder.layers.7.self_attention.linear_qkv.weight + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight +Params for bucket 11 (42213376 elements, 42213376 padded size): + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_qkv.weight +Params for bucket 12 (42213376 elements, 42213376 padded size): + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight +Params for bucket 13 (42213376 elements, 42213376 padded size): + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.pre_mlp_layernorm.weight +Params for bucket 14 (42213376 elements, 42213376 padded size): + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.pre_mlp_layernorm.weight +Params for bucket 15 (42213376 elements, 42213376 padded size): + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_qkv.weight + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight +Params for bucket 16 (42213376 elements, 42213376 padded size): + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight +Params for bucket 17 (42213376 elements, 42213376 padded size): + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.pre_mlp_layernorm.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.mlp.router.weight +Params for bucket 18 (206045184 elements, 206045184 padded size): + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 64 +Params for bucket 1 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 + module.decoder.layers.15.mlp.experts.linear_fc2.weight3 +Params for bucket 2 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 +Params for bucket 4 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 +Params for bucket 5 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 +Params for bucket 6 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 +Params for bucket 9 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight5 + module.decoder.layers.13.mlp.experts.linear_fc2.weight3 +Params for bucket 10 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight1 + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight5 + module.decoder.layers.13.mlp.experts.linear_fc1.weight3 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 +Params for bucket 13 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 + module.decoder.layers.12.mlp.experts.linear_fc2.weight4 + module.decoder.layers.12.mlp.experts.linear_fc2.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight3 + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 +Params for bucket 14 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight1 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc2.weight0 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 +Params for bucket 17 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 +Params for bucket 18 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 +Params for bucket 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 +Params for bucket 21 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight4 + module.decoder.layers.10.mlp.experts.linear_fc2.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 +Params for bucket 22 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight5 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 +Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 +Params for bucket 26 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 +Params for bucket 27 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 +Params for bucket 29 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 +Params for bucket 30 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 +Params for bucket 31 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.7.mlp.experts.linear_fc2.weight3 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 + module.decoder.layers.7.mlp.experts.linear_fc2.weight2 +Params for bucket 34 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 +Params for bucket 35 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight5 + module.decoder.layers.6.mlp.experts.linear_fc2.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 +Params for bucket 38 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight1 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 +Params for bucket 39 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 +Params for bucket 42 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 +Params for bucket 43 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 +Params for bucket 46 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 +Params for bucket 47 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +Params for bucket 51 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight5 + module.decoder.layers.3.mlp.experts.linear_fc1.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 +Params for bucket 52 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.3.mlp.experts.linear_fc1.weight1 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 + module.decoder.layers.2.mlp.experts.linear_fc2.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 +Params for bucket 54 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 +Params for bucket 55 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 +Params for bucket 56 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 +Params for bucket 59 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 +Params for bucket 60 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 +Params for bucket 63 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 +Params for bucket 64 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight2 + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight1 +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=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +[after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:32:38 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + validation: 0 + test: 0 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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) +INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers +INFO:megatron.core.datasets.indexed_dataset: Extract the document indices +INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029 +INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350 +INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices +INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-30 07:32:38 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (3667.18, 3673.84) + train/valid/test-data-iterators-setup ..........: (70.11, 88.94) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 07:32:38 +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") + [2026-08-30 07:33:05] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26929.5 | throughput per GPU (TFLOP/s/GPU): 12.4 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102352E+01 | loss scale: 1.0 | grad norm: 44.192 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67072 | min safe capacity: 1.0234 | +Number of parameters in transformer block in billions: 23.22 +Number of parameters in embedding layers in billions: 0.41 +Total number of parameters in billions: 23.63 +Number of parameters in most loaded shard in billions: 23.6320 +Theoretical memory footprints: weight and optimizer=169029.43 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.49609375 | max allocated: 76960.7119140625 | reserved: 79598.0 | max reserved: 79598.0 + [2026-08-30 07:33:06] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 1008.9 | throughput per GPU (TFLOP/s/GPU): 330.4 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101566E+01 | loss scale: 1.0 | grad norm: 45.944 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:06] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 503.9 | throughput per GPU (TFLOP/s/GPU): 661.7 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.052005E+01 | loss scale: 1.0 | grad norm: 215.129 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 07:33:09] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 3058.6 | throughput per GPU (TFLOP/s/GPU): 109.0 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.316056E+01 | loss scale: 1.0 | grad norm: 38.034 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:10] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 446.1 | throughput per GPU (TFLOP/s/GPU): 747.4 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.038146E+01 | loss scale: 1.0 | grad norm: 16.240 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:10] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 466.0 | throughput per GPU (TFLOP/s/GPU): 715.4 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.038034E+01 | loss scale: 1.0 | grad norm: 32.686 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:11] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 460.8 | throughput per GPU (TFLOP/s/GPU): 723.6 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.004451E+01 | loss scale: 1.0 | grad norm: 7.311 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:11] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 455.5 | throughput per GPU (TFLOP/s/GPU): 732.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.520081E+00 | loss scale: 1.0 | grad norm: 5.762 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:12] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 462.4 | throughput per GPU (TFLOP/s/GPU): 721.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.274012E+00 | loss scale: 1.0 | grad norm: 5.595 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:12] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 456.8 | throughput per GPU (TFLOP/s/GPU): 729.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.114054E+00 | loss scale: 1.0 | grad norm: 5.427 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:13] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 457.8 | throughput per GPU (TFLOP/s/GPU): 728.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.996742E+00 | loss scale: 1.0 | grad norm: 5.229 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | + [2026-08-30 07:33:13] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 459.3 | throughput per GPU (TFLOP/s/GPU): 725.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.877083E+00 | loss scale: 1.0 | grad norm: 5.308 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 67328 | min safe capacity: 1.0273 | +[after training is done] datetime: 2026-08-30 07:33:13 +[rank0]:[W830 07:33:14.038597114 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()) +[rank1]:[W830 07:33:16.571467371 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()) +[rank3]:[W830 07:33:16.605477777 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()) +[rank2]:[W830 07:33:16.608513211 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()) +[rank6]:[W830 07:33:16.608649446 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()) +[rank5]:[W830 07:33:16.645146558 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()) +[rank7]:[W830 07:33:16.645834804 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()) +[rank4]:[W830 07:33:16.151519809 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()) +W0830 07:33:19.737000 772411 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0830 07:33:19.737000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772499 closing signal SIGTERM +W0830 07:33:19.738000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772500 closing signal SIGTERM +W0830 07:33:19.739000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772501 closing signal SIGTERM +W0830 07:33:19.740000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772502 closing signal SIGTERM +W0830 07:33:19.740000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772503 closing signal SIGTERM +W0830 07:33:19.741000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772504 closing signal SIGTERM +W0830 07:33:19.742000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772505 closing signal SIGTERM +W0830 07:33:19.743000 772411 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 772506 closing signal SIGTERM +Traceback (most recent call last): + File "/usr/local/bin/torchrun", line 8, in + sys.exit(main()) + ^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper + return f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main + run(args) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run + elastic_launch( + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ + return launch_agent(self._config, self._entrypoint, list(args)) + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent + result = agent.run() + ^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper + result = f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run + result = self._invoke_run(role) + ^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run + time.sleep(monitor_interval) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler + raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) +torch.distributed.elastic.multiprocessing.api.SignalException: Process 772411 got signal: 15 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/mem_samples.csv b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..327c1560d367aed3e46999e20dd276b5a78b15b3 --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/mem_samples.csv @@ -0,0 +1,232 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 4 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 45 +1, 725 +2, 725 +3, 725 +4, 725 +5, 725 +6, 725 +7, 725 +0, 1003 +1, 1041 +2, 1043 +3, 1051 +4, 1069 +5, 1051 +6, 1063 +7, 1057 +0, 1895 +1, 1895 +2, 1895 +3, 1895 +4, 1895 +5, 1895 +6, 1895 +7, 1895 +0, 2419 +1, 2391 +2, 2395 +3, 2413 +4, 2389 +5, 2413 +6, 2395 +7, 2405 +0, 11957 +1, 12567 +2, 12069 +3, 12641 +4, 12761 +5, 12469 +6, 12309 +7, 12567 +0, 42917 +1, 43365 +2, 43365 +3, 43365 +4, 42917 +5, 43365 +6, 43365 +7, 43365 +0, 44145 +1, 44589 +2, 44599 +3, 44607 +4, 44141 +5, 44597 +6, 44607 +7, 44589 +0, 44463 +1, 44911 +2, 44911 +3, 44911 +4, 44463 +5, 44911 +6, 44911 +7, 44911 +0, 45585 +1, 46033 +2, 46033 +3, 46033 +4, 45585 +5, 46033 +6, 46033 +7, 46033 +0, 48337 +1, 48785 +2, 48785 +3, 48785 +4, 48337 +5, 48785 +6, 48785 +7, 48785 +0, 66585 +1, 66581 +2, 66499 +3, 66621 +4, 66645 +5, 66683 +6, 66661 +7, 67195 +0, 79825 +1, 79841 +2, 79841 +3, 79841 +4, 79845 +5, 79841 +6, 79841 +7, 79841 +0, 80949 +1, 80965 +2, 80965 +3, 80965 +4, 80969 +5, 80965 +6, 80965 +7, 80965 +0, 80961 +1, 80977 +2, 80977 +3, 80977 +4, 80981 +5, 80977 +6, 80977 +7, 80977 +0, 92565 +1, 92581 +2, 92581 +3, 92581 +4, 92585 +5, 92581 +6, 92581 +7, 92581 +0, 93157 +1, 93165 +2, 93131 +3, 93167 +4, 93221 +5, 93163 +6, 93201 +7, 93191 +0, 94301 +1, 94293 +2, 94309 +3, 94267 +4, 94333 +5, 94343 +6, 94343 +7, 94335 +0, 92789 +1, 89839 +2, 89987 +3, 89839 +4, 94335 +5, 89965 +6, 90825 +7, 87655 +0, 132221 +1, 132435 +2, 132437 +3, 132435 +4, 132221 +5, 132435 +6, 132435 +7, 132435 +0, 132221 +1, 132435 +2, 132437 +3, 132435 +4, 132221 +5, 132435 +6, 132435 +7, 132435 +0, 132221 +1, 132435 +2, 132437 +3, 132435 +4, 132221 +5, 132435 +6, 132435 +7, 132435 +0, 128061 +1, 128013 +2, 128277 +3, 128275 +4, 128061 +5, 128013 +6, 128275 +7, 128275 +0, 125347 +1, 125689 +2, 126855 +3, 126527 +4, 125731 +5, 126271 +6, 126853 +7, 126201 +0, 3 +1, 4 +2, 4 +3, 125173 +4, 4 +5, 4 +6, 4 +7, 4 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/run.env b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/run.env new file mode 100644 index 0000000000000000000000000000000000000000..37dca56d5f2930f6fa9c4c0dfb76169a3b8cbd3d --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/run.env @@ -0,0 +1,70 @@ +backend=ultraep +config=G4 +family=tierA +repo=/workspace/Megatron-UltraEP +branch=ultraep-bracing-regions-through-kernels +commit=03d57ef7 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=896 +num_experts=128 +topk=32 +experts_per_rank=16 +expert_params=11010048 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=131072 +mean_tokens_per_expert=8192 +slot_policy=matched +slots_per_rank=16 +capacity_mode=static +capacity=1.095703125 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens=1024 +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/train.log b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/train.log new file mode 100644 index 0000000000000000000000000000000000000000..b492b5be2e3183e793779dc14a1037c636a91775 --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/train.log @@ -0,0 +1,1571 @@ +============================================================== + G4 / ultraep (slots/rank=16 policy=matched) + E=128 K=32 H'=896 m=16 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.095703125 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-UltraEP @ 03d57ef7 +============================================================== +W0830 07:33:25.255000 775465 torch/distributed/run.py:766] +W0830 07:33:25.255000 775465 torch/distributed/run.py:766] ***************************************** +W0830 07:33:25.255000 775465 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. +W0830 07:33:25.255000 775465 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +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 +WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer +Number of virtual stages per pipeline stage: None +accumulate and all-reduce gradients in fp32 for bfloat16 data type. +using torch.bfloat16 for parameters ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. True + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. False + add_position_embedding .......................... False + add_qkv_bias .................................... False + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... True + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.0 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ True + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ False + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full_iteration + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 8 + data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document'] + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + delay_wgrad_compute ............................. False + deprecated_use_mcore_models ..................... True + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_init_method_std ....................... None + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... True + enable_experimental ............................. False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 16 + encoder_seq_length .............................. 4096 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 1000000 + eval_iters ...................................... 0 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 8 + expert_tensor_parallel_size ..................... 1 + export_force_local_attention .................... False + export_kd_cfg ................................... None + export_kd_teacher_ckpt_format ................... None + export_kd_teacher_load .......................... None + export_kv_cache_quant ........................... False + export_legacy_megatron .......................... False + export_model_type ............................... GPTModel + export_moe_apply_probs_on_input ................. False + export_qk_l2_norm ............................... False + export_quant_cfg ................................ None + export_real_quant_cfg ........................... None + export_te_mcore_model ........................... False + external_cuda_graph ............................. False + ffn_hidden_size ................................. 14336 + finetune ........................................ False + finetune_data_split ............................. train + finetune_hf_dataset ............................. None + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ False + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. e4m3 + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... mxfp8 + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 16 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_dynamic_batching_num_cuda_graphs ...... 16 + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.01 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kitchen_config_file ............................. None + kitchen_recipe_number ........................... None + kv_channels ..................................... 128 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ None + load_main_params_from_ckpt ...................... None + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_energy ...................................... False + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. True + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... None + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0001 + lr_decay_iters .................................. 8 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 5 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... False + max_position_embeddings ......................... 32768 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 1e-05 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... False + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.01 + moe_capacity_monitor ............................ True + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ False + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. False + moe_enable_hybridep ............................. True + moe_enable_ultraep .............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 896 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ None + moe_num_redundant_experts_per_rank .............. 16 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.095703125 + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ fp32 + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 32 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + moe_ultraep_max_microbatches .................... None + moe_ultraep_overlap_weight_sync ................. True + moe_ultraep_weight_sync_join .................... before_dispatch + moe_upcycling_granularity ....................... 1 + moe_use_device_initiated_grouped_gemm ........... True + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 0.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... RMSNorm + num_attention_heads ............................. 32 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 128 + num_layers ...................................... 16 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 8 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. True + overlap_moe_expert_parallel_comm ................ False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ True + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.bfloat16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_layout .................. None + pipeline_model_parallel_size .................... 1 + position_embedding_type ......................... rope + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... validate_results + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + reuse_grad_buf_for_mxfp8_param_ag ............... False + rope_scaling_factor ............................. 8.0 + rope_type ....................................... None + rotary_base ..................................... 1000000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ None + save_interval ................................... None + save_retain_interval ............................ None + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 4096 + sequence_parallel ............................... False + sft ............................................. False + sft_tokenizer_prompt_format ..................... nemotron-h-aligned + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... 100,0,0 + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... True + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. True + tensor_model_parallel_size ...................... 1 + tensorboard_dir ................................. None + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 12 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. True + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... True + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2 +> building GPT2BPETokenizer tokenizer ... +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 + > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304) +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS +> initializing torch distributed ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank +4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank [Gloo] Rank 1 is connected to 0[Gloo] Rank 7 is connected to peer ranks. 75Expected number of connected peer ranks is : peer ranks. is connected to 7Expected number of connected peer ranks is : 7 +7 peer ranks. +Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. [Gloo] Rank Expected number of connected peer ranks is : 74 + is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +> initialized tensor model parallel with size 1 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[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 + +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[rank6]:[W830 07:33:42.035273235 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. +[rank5]:[W830 07:33:42.038243278 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. +[rank7]:[W830 07:33:42.038907429 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. +[rank3]:[W830 07:33:42.041743142 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. +[rank4]:[W830 07:33:42.041855450 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. +[rank2]:[W830 07:33:42.041896239 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. +[rank1]:[W830 07:33:42.042049203 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.036 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +[rank0]:[W830 07:33:42.213761334 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. +>>> done with compiling and loading fused kernels. Compilation time: 4.030 seconds +time to initialize megatron (seconds): 7.184 +[after megatron is initialized] datetime: 2026-08-30 07:33:48 +building GPT model ... +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +[INFO] Use MNNVL fabric: False +[WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8. +NVSHMEM v3.7.2 +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive). + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6728847360 +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) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18 +Params for bucket 1 (206045184 elements, 206045184 padded size): + module.output_layer.weight +Params for bucket 2 (42479616 elements, 42479616 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_proj.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.layers.15.pre_mlp_layernorm.weight +Params for bucket 3 (42475520 elements, 42475520 padded size): + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.14.self_attention.linear_qkv.weight + module.decoder.layers.14.pre_mlp_layernorm.weight +Params for bucket 4 (42475520 elements, 42475520 padded size): + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.13.self_attention.linear_qkv.weight + module.decoder.layers.13.pre_mlp_layernorm.weight +Params for bucket 5 (42475520 elements, 42475520 padded size): + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.pre_mlp_layernorm.weight +Params for bucket 6 (42475520 elements, 42475520 padded size): + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_qkv.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight +Params for bucket 7 (42475520 elements, 42475520 padded size): + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.10.self_attention.linear_proj.weight +Params for bucket 8 (42475520 elements, 42475520 padded size): + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight +Params for bucket 9 (42475520 elements, 42475520 padded size): + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.8.self_attention.linear_proj.weight +Params for bucket 10 (42475520 elements, 42475520 padded size): + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.7.self_attention.linear_qkv.weight + module.decoder.layers.7.pre_mlp_layernorm.weight +Params for bucket 11 (42475520 elements, 42475520 padded size): + module.decoder.layers.6.self_attention.linear_qkv.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.6.self_attention.linear_proj.weight +Params for bucket 12 (42475520 elements, 42475520 padded size): + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.pre_mlp_layernorm.weight +Params for bucket 13 (42475520 elements, 42475520 padded size): + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.pre_mlp_layernorm.weight +Params for bucket 14 (42475520 elements, 42475520 padded size): + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.pre_mlp_layernorm.weight +Params for bucket 15 (42475520 elements, 42475520 padded size): + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.2.self_attention.linear_qkv.weight + module.decoder.layers.2.pre_mlp_layernorm.weight +Params for bucket 16 (42475520 elements, 42475520 padded size): + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.self_attention.linear_qkv.weight +Params for bucket 17 (42475520 elements, 42475520 padded size): + module.decoder.layers.0.mlp.router.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.pre_mlp_layernorm.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight +Params for bucket 18 (206045184 elements, 206045184 padded size): + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 67 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight15 + module.decoder.layers.15.mlp.experts.linear_fc2.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight11 + module.decoder.layers.15.mlp.experts.linear_fc2.weight9 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 + module.decoder.layers.15.mlp.experts.linear_fc2.weight14 + module.decoder.layers.15.mlp.experts.linear_fc2.weight12 + module.decoder.layers.15.mlp.experts.linear_fc2.weight10 + module.decoder.layers.15.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 +Params for bucket 2 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight3 + module.decoder.layers.15.mlp.experts.linear_fc2.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight14 + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight15 + module.decoder.layers.15.mlp.experts.linear_fc1.weight13 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight12 + module.decoder.layers.15.mlp.experts.linear_fc1.weight10 + module.decoder.layers.15.mlp.experts.linear_fc1.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight9 + module.decoder.layers.15.mlp.experts.linear_fc1.weight11 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 +Params for bucket 4 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + module.decoder.layers.14.mlp.experts.linear_fc2.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc2.weight11 + module.decoder.layers.14.mlp.experts.linear_fc2.weight9 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 + module.decoder.layers.14.mlp.experts.linear_fc2.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight12 + module.decoder.layers.14.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc2.weight8 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight15 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight12 + module.decoder.layers.14.mlp.experts.linear_fc1.weight8 + module.decoder.layers.14.mlp.experts.linear_fc1.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight11 + module.decoder.layers.14.mlp.experts.linear_fc1.weight9 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + module.decoder.layers.13.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc2.weight9 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 + module.decoder.layers.13.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc2.weight10 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight5 + module.decoder.layers.13.mlp.experts.linear_fc2.weight3 + module.decoder.layers.13.mlp.experts.linear_fc2.weight1 + module.decoder.layers.13.mlp.experts.linear_fc2.weight8 + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 + module.decoder.layers.13.mlp.experts.linear_fc1.weight15 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight12 + module.decoder.layers.13.mlp.experts.linear_fc1.weight10 + module.decoder.layers.13.mlp.experts.linear_fc1.weight13 + module.decoder.layers.13.mlp.experts.linear_fc1.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight9 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight8 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + module.decoder.layers.13.mlp.experts.linear_fc1.weight5 + module.decoder.layers.13.mlp.experts.linear_fc1.weight3 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc2.weight11 + module.decoder.layers.12.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight12 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight9 + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 + module.decoder.layers.12.mlp.experts.linear_fc2.weight3 + module.decoder.layers.12.mlp.experts.linear_fc2.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight10 + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 + module.decoder.layers.12.mlp.experts.linear_fc2.weight8 + module.decoder.layers.12.mlp.experts.linear_fc2.weight4 + module.decoder.layers.12.mlp.experts.linear_fc2.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight0 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight12 + module.decoder.layers.12.mlp.experts.linear_fc1.weight10 + module.decoder.layers.12.mlp.experts.linear_fc1.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight11 + module.decoder.layers.12.mlp.experts.linear_fc1.weight13 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight8 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 + module.decoder.layers.12.mlp.experts.linear_fc1.weight9 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 +Params for bucket 17 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc2.weight14 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight9 + module.decoder.layers.11.mlp.experts.linear_fc2.weight11 + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 + module.decoder.layers.11.mlp.experts.linear_fc2.weight12 + module.decoder.layers.11.mlp.experts.linear_fc2.weight10 + module.decoder.layers.11.mlp.experts.linear_fc2.weight8 + module.decoder.layers.11.mlp.experts.linear_fc2.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 +Params for bucket 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight15 + module.decoder.layers.11.mlp.experts.linear_fc1.weight13 + module.decoder.layers.11.mlp.experts.linear_fc1.weight11 + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight14 + module.decoder.layers.11.mlp.experts.linear_fc1.weight12 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight9 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 + module.decoder.layers.11.mlp.experts.linear_fc1.weight10 + module.decoder.layers.11.mlp.experts.linear_fc1.weight8 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight15 +Params for bucket 22 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight14 + module.decoder.layers.10.mlp.experts.linear_fc2.weight10 + module.decoder.layers.10.mlp.experts.linear_fc2.weight12 + module.decoder.layers.10.mlp.experts.linear_fc2.weight8 + module.decoder.layers.10.mlp.experts.linear_fc2.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight4 + module.decoder.layers.10.mlp.experts.linear_fc2.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight11 + module.decoder.layers.10.mlp.experts.linear_fc2.weight9 + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight15 + module.decoder.layers.10.mlp.experts.linear_fc1.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight14 + module.decoder.layers.10.mlp.experts.linear_fc1.weight12 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight11 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.10.mlp.experts.linear_fc1.weight9 + module.decoder.layers.10.mlp.experts.linear_fc1.weight10 + module.decoder.layers.10.mlp.experts.linear_fc1.weight8 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 +Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight5 + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 + module.decoder.layers.10.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight12 + module.decoder.layers.9.mlp.experts.linear_fc2.weight14 + module.decoder.layers.9.mlp.experts.linear_fc2.weight10 + module.decoder.layers.9.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight15 + module.decoder.layers.9.mlp.experts.linear_fc2.weight11 + module.decoder.layers.9.mlp.experts.linear_fc2.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight9 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 +Params for bucket 27 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight15 + module.decoder.layers.9.mlp.experts.linear_fc1.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight14 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight11 + module.decoder.layers.9.mlp.experts.linear_fc1.weight9 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight12 + module.decoder.layers.9.mlp.experts.linear_fc1.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight10 +Params for bucket 29 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 +Params for bucket 30 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight14 + module.decoder.layers.8.mlp.experts.linear_fc2.weight12 + module.decoder.layers.8.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight13 + module.decoder.layers.8.mlp.experts.linear_fc2.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight11 + module.decoder.layers.8.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 +Params for bucket 31 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight14 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight13 + module.decoder.layers.8.mlp.experts.linear_fc1.weight11 + module.decoder.layers.8.mlp.experts.linear_fc1.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight12 + module.decoder.layers.8.mlp.experts.linear_fc1.weight8 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight15 + module.decoder.layers.7.mlp.experts.linear_fc2.weight13 + module.decoder.layers.7.mlp.experts.linear_fc2.weight11 + module.decoder.layers.7.mlp.experts.linear_fc2.weight9 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight8 + module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 + module.decoder.layers.7.mlp.experts.linear_fc2.weight2 + module.decoder.layers.7.mlp.experts.linear_fc2.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.7.mlp.experts.linear_fc2.weight3 + module.decoder.layers.7.mlp.experts.linear_fc2.weight1 + module.decoder.layers.7.mlp.experts.linear_fc1.weight15 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight10 + module.decoder.layers.7.mlp.experts.linear_fc1.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight11 + module.decoder.layers.7.mlp.experts.linear_fc1.weight13 + module.decoder.layers.7.mlp.experts.linear_fc1.weight9 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight8 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight15 + module.decoder.layers.6.mlp.experts.linear_fc2.weight13 + module.decoder.layers.6.mlp.experts.linear_fc2.weight11 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight10 + module.decoder.layers.6.mlp.experts.linear_fc2.weight8 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight9 + module.decoder.layers.6.mlp.experts.linear_fc2.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight5 + module.decoder.layers.6.mlp.experts.linear_fc2.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight1 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight11 + module.decoder.layers.6.mlp.experts.linear_fc1.weight13 + module.decoder.layers.6.mlp.experts.linear_fc1.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight12 + module.decoder.layers.6.mlp.experts.linear_fc1.weight10 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight9 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc1.weight8 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 42 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc2.weight13 +Params for bucket 43 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight12 + module.decoder.layers.5.mlp.experts.linear_fc2.weight10 + module.decoder.layers.5.mlp.experts.linear_fc2.weight8 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight9 + module.decoder.layers.5.mlp.experts.linear_fc2.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight13 + module.decoder.layers.5.mlp.experts.linear_fc1.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight14 + module.decoder.layers.5.mlp.experts.linear_fc1.weight12 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight9 + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight10 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc1.weight8 +Params for bucket 46 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc2.weight15 +Params for bucket 47 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight14 + module.decoder.layers.4.mlp.experts.linear_fc2.weight10 + module.decoder.layers.4.mlp.experts.linear_fc2.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight8 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 + module.decoder.layers.4.mlp.experts.linear_fc2.weight13 + module.decoder.layers.4.mlp.experts.linear_fc2.weight11 + module.decoder.layers.4.mlp.experts.linear_fc2.weight9 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight15 + module.decoder.layers.4.mlp.experts.linear_fc1.weight13 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight14 + module.decoder.layers.4.mlp.experts.linear_fc1.weight12 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight11 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight9 + module.decoder.layers.4.mlp.experts.linear_fc1.weight10 + module.decoder.layers.4.mlp.experts.linear_fc1.weight8 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 +Params for bucket 51 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight15 + module.decoder.layers.3.mlp.experts.linear_fc2.weight11 + module.decoder.layers.3.mlp.experts.linear_fc2.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight9 + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + module.decoder.layers.3.mlp.experts.linear_fc2.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight12 + module.decoder.layers.3.mlp.experts.linear_fc2.weight10 + module.decoder.layers.3.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 +Params for bucket 52 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight15 + module.decoder.layers.3.mlp.experts.linear_fc1.weight13 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight12 + module.decoder.layers.3.mlp.experts.linear_fc1.weight10 + module.decoder.layers.3.mlp.experts.linear_fc1.weight8 + module.decoder.layers.3.mlp.experts.linear_fc1.weight11 + module.decoder.layers.3.mlp.experts.linear_fc1.weight9 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +Params for bucket 54 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight5 + module.decoder.layers.3.mlp.experts.linear_fc1.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight1 +Params for bucket 55 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight12 + module.decoder.layers.2.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc2.weight8 + module.decoder.layers.2.mlp.experts.linear_fc2.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight13 + module.decoder.layers.2.mlp.experts.linear_fc2.weight11 + module.decoder.layers.2.mlp.experts.linear_fc2.weight9 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 +Params for bucket 56 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 + module.decoder.layers.2.mlp.experts.linear_fc2.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight15 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight12 + module.decoder.layers.2.mlp.experts.linear_fc1.weight8 + module.decoder.layers.2.mlp.experts.linear_fc1.weight13 + module.decoder.layers.2.mlp.experts.linear_fc1.weight11 + module.decoder.layers.2.mlp.experts.linear_fc1.weight9 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 +Params for bucket 59 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight13 + module.decoder.layers.1.mlp.experts.linear_fc2.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc2.weight9 + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight14 + module.decoder.layers.1.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc2.weight10 +Params for bucket 60 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight8 + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight15 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight10 + module.decoder.layers.1.mlp.experts.linear_fc1.weight12 + module.decoder.layers.1.mlp.experts.linear_fc1.weight13 + module.decoder.layers.1.mlp.experts.linear_fc1.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight9 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight8 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + module.decoder.layers.1.mlp.experts.linear_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 +Params for bucket 63 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight14 + module.decoder.layers.0.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight15 + module.decoder.layers.0.mlp.experts.linear_fc2.weight13 + module.decoder.layers.0.mlp.experts.linear_fc2.weight11 +Params for bucket 64 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight10 + module.decoder.layers.0.mlp.experts.linear_fc2.weight8 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight9 + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 +Params for bucket 65 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight15 + module.decoder.layers.0.mlp.experts.linear_fc1.weight13 + module.decoder.layers.0.mlp.experts.linear_fc1.weight11 + module.decoder.layers.0.mlp.experts.linear_fc1.weight12 + module.decoder.layers.0.mlp.experts.linear_fc1.weight14 + module.decoder.layers.0.mlp.experts.linear_fc1.weight10 +Params for bucket 66 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight9 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight8 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 +Params for bucket 67 (29360128 elements, 29360128 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + module.decoder.layers.0.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight2 +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=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +[after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:33:52 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + validation: 0 + test: 0 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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) +INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers +INFO:megatron.core.datasets.indexed_dataset: Extract the document indices +INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029 +INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350 +INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices +INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-30 07:33:52 +done with setup ... +training ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (3857.92, 3871.89) + train/valid/test-data-iterators-setup ..........: (72.62, 93.48) +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 07:33:53 +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +Number of parameters in transformer block in billions: 23.22 [2026-08-30 07:34:19] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26749.6 | throughput per GPU (TFLOP/s/GPU): 12.5 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102500E+01 | loss scale: 1.0 | grad norm: 44.121 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 133888 | min safe capacity: 1.0215 | + +Number of parameters in embedding layers in billions: 0.41 +Total number of parameters in billions: 23.63 +Number of parameters in most loaded shard in billions: 23.6320 +Theoretical memory footprints: weight and optimizer=169029.43 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 58109.861328125 | max allocated: 83199.2578125 | reserved: 85038.0 | max reserved: 85038.0 + [2026-08-30 07:34:20] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 682.2 | throughput per GPU (TFLOP/s/GPU): 488.7 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101850E+01 | loss scale: 1.0 | grad norm: 45.886 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:20] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 514.2 | throughput per GPU (TFLOP/s/GPU): 648.4 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.060152E+01 | loss scale: 1.0 | grad norm: 216.512 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 07:34:23] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2487.8 | throughput per GPU (TFLOP/s/GPU): 134.0 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.301575E+01 | loss scale: 1.0 | grad norm: 46.073 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:23] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 476.7 | throughput per GPU (TFLOP/s/GPU): 699.3 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.033463E+01 | loss scale: 1.0 | grad norm: 17.368 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:24] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 483.8 | throughput per GPU (TFLOP/s/GPU): 689.1 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.016443E+01 | loss scale: 1.0 | grad norm: 17.568 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:24] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 478.6 | throughput per GPU (TFLOP/s/GPU): 696.6 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.898163E+00 | loss scale: 1.0 | grad norm: 6.180 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:25] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 486.3 | throughput per GPU (TFLOP/s/GPU): 685.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.318243E+00 | loss scale: 1.0 | grad norm: 5.515 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:25] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 480.5 | throughput per GPU (TFLOP/s/GPU): 693.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.066381E+00 | loss scale: 1.0 | grad norm: 5.390 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:26] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 475.2 | throughput per GPU (TFLOP/s/GPU): 701.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.913552E+00 | loss scale: 1.0 | grad norm: 5.195 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:26] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 485.9 | throughput per GPU (TFLOP/s/GPU): 686.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.801952E+00 | loss scale: 1.0 | grad norm: 4.958 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:34:27] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 492.0 | throughput per GPU (TFLOP/s/GPU): 677.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.685153E+00 | loss scale: 1.0 | grad norm: 4.900 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | +[after training is done] datetime: 2026-08-30 07:34:27 +[rank0]:[W830 07:34:28.737938015 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()) +[rank4]:[W830 07:34:29.879350971 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()) +[rank5]:[W830 07:34:29.879822390 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()) +[rank3]:[W830 07:34:29.881715536 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()) +[rank7]:[W830 07:34:29.883267640 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()) +[rank2]:[W830 07:34:29.883599961 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()) +[rank6]:[W830 07:34:30.271114218 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()) +[rank1]:[W830 07:34:30.423137865 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()) +W0830 07:34:33.945000 775465 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0830 07:34:33.945000 775465 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 775553 closing signal SIGTERM +W0830 07:34:33.946000 775465 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 775554 closing signal SIGTERM +W0830 07:34:33.946000 775465 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 775555 closing signal SIGTERM +W0830 07:34:33.946000 775465 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 775556 closing signal SIGTERM +W0830 07:34:33.947000 775465 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 775557 closing signal SIGTERM +W0830 07:34:33.947000 775465 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 775558 closing signal SIGTERM +W0830 07:34:33.948000 775465 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 775559 closing signal SIGTERM +Traceback (most recent call last): + File "/usr/local/bin/torchrun", line 8, in + sys.exit(main()) + ^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper + return f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main + run(args) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run + elastic_launch( + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ + return launch_agent(self._config, self._entrypoint, list(args)) + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent + result = agent.run() + ^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper + result = f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run + result = self._invoke_run(role) + ^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run + time.sleep(monitor_interval) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler + raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) +torch.distributed.elastic.multiprocessing.api.SignalException: Process 775465 got signal: 15 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/mem_samples.csv b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..e1aa4cdd4525cfcf96a97f507c605e6f44e67293 --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/mem_samples.csv @@ -0,0 +1,232 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 1 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 4 +1, 4 +2, 629 +3, 4 +4, 629 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 821 +7, 821 +0, 1895 +1, 1895 +2, 1895 +3, 1895 +4, 1895 +5, 1895 +6, 1895 +7, 1895 +0, 2307 +1, 2315 +2, 2313 +3, 2317 +4, 2313 +5, 2313 +6, 2317 +7, 2325 +0, 12055 +1, 11543 +2, 12069 +3, 12069 +4, 12149 +5, 12229 +6, 12129 +7, 12069 +0, 43095 +1, 43543 +2, 43543 +3, 43543 +4, 43095 +5, 43543 +6, 43543 +7, 43543 +0, 44163 +1, 44611 +2, 44611 +3, 44611 +4, 44163 +5, 44611 +6, 44611 +7, 44611 +0, 44463 +1, 44911 +2, 44911 +3, 44911 +4, 44463 +5, 44911 +6, 44911 +7, 44911 +0, 45585 +1, 46033 +2, 46033 +3, 46033 +4, 45585 +5, 46033 +6, 46033 +7, 46033 +0, 48337 +1, 48785 +2, 48785 +3, 48785 +4, 48337 +5, 48785 +6, 48785 +7, 48785 +0, 75509 +1, 75465 +2, 75463 +3, 75443 +4, 75489 +5, 75505 +6, 75443 +7, 75649 +0, 79825 +1, 79841 +2, 79841 +3, 79841 +4, 79845 +5, 79841 +6, 79841 +7, 79841 +0, 80949 +1, 80965 +2, 80965 +3, 80965 +4, 80969 +5, 80965 +6, 80965 +7, 80965 +0, 80963 +1, 80979 +2, 80979 +3, 80979 +4, 80983 +5, 80979 +6, 80979 +7, 80979 +0, 92107 +1, 92167 +2, 92167 +3, 92179 +4, 92195 +5, 92191 +6, 92159 +7, 92185 +0, 93101 +1, 93079 +2, 93103 +3, 93079 +4, 93077 +5, 93111 +6, 93119 +7, 93095 +0, 94241 +1, 94255 +2, 94209 +3, 94225 +4, 94285 +5, 94243 +6, 94293 +7, 94281 +0, 90589 +1, 87745 +2, 87725 +3, 87787 +4, 89771 +5, 90029 +6, 89837 +7, 89967 +0, 132221 +1, 132435 +2, 132437 +3, 132437 +4, 132221 +5, 132437 +6, 132435 +7, 132437 +0, 132221 +1, 132435 +2, 132437 +3, 132437 +4, 132221 +5, 132437 +6, 132435 +7, 132437 +0, 132221 +1, 132435 +2, 132437 +3, 132437 +4, 132221 +5, 132437 +6, 132435 +7, 132437 +0, 128061 +1, 128275 +2, 128277 +3, 128277 +4, 128061 +5, 128277 +6, 128275 +7, 128277 +0, 126639 +1, 126853 +2, 126593 +3, 126855 +4, 126639 +5, 126855 +6, 126853 +7, 126855 +0, 1 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 125563 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/run.env b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/run.env new file mode 100644 index 0000000000000000000000000000000000000000..37dca56d5f2930f6fa9c4c0dfb76169a3b8cbd3d --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/run.env @@ -0,0 +1,70 @@ +backend=ultraep +config=G4 +family=tierA +repo=/workspace/Megatron-UltraEP +branch=ultraep-bracing-regions-through-kernels +commit=03d57ef7 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=896 +num_experts=128 +topk=32 +experts_per_rank=16 +expert_params=11010048 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=131072 +mean_tokens_per_expert=8192 +slot_policy=matched +slots_per_rank=16 +capacity_mode=static +capacity=1.095703125 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens=1024 +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/train.log b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/train.log new file mode 100644 index 0000000000000000000000000000000000000000..db6209115981484a9da1fdca65cb70d6f8557467 --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/train.log @@ -0,0 +1,1572 @@ +============================================================== + G4 / ultraep (slots/rank=16 policy=matched) + E=128 K=32 H'=896 m=16 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.095703125 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-UltraEP @ 03d57ef7 +============================================================== +W0830 07:34:39.450000 778522 torch/distributed/run.py:766] +W0830 07:34:39.450000 778522 torch/distributed/run.py:766] ***************************************** +W0830 07:34:39.450000 778522 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. +W0830 07:34:39.450000 778522 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +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 +WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer +Number of virtual stages per pipeline stage: None +accumulate and all-reduce gradients in fp32 for bfloat16 data type. +using torch.bfloat16 for parameters ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. True + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. False + add_position_embedding .......................... False + add_qkv_bias .................................... False + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... True + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.0 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ True + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ False + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full_iteration + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 8 + data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document'] + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + delay_wgrad_compute ............................. False + deprecated_use_mcore_models ..................... True + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_init_method_std ....................... None + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... True + enable_experimental ............................. False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 16 + encoder_seq_length .............................. 4096 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 1000000 + eval_iters ...................................... 0 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 8 + expert_tensor_parallel_size ..................... 1 + export_force_local_attention .................... False + export_kd_cfg ................................... None + export_kd_teacher_ckpt_format ................... None + export_kd_teacher_load .......................... None + export_kv_cache_quant ........................... False + export_legacy_megatron .......................... False + export_model_type ............................... GPTModel + export_moe_apply_probs_on_input ................. False + export_qk_l2_norm ............................... False + export_quant_cfg ................................ None + export_real_quant_cfg ........................... None + export_te_mcore_model ........................... False + external_cuda_graph ............................. False + ffn_hidden_size ................................. 14336 + finetune ........................................ False + finetune_data_split ............................. train + finetune_hf_dataset ............................. None + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ False + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. e4m3 + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... mxfp8 + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 16 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_dynamic_batching_num_cuda_graphs ...... 16 + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.01 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kitchen_config_file ............................. None + kitchen_recipe_number ........................... None + kv_channels ..................................... 128 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ None + load_main_params_from_ckpt ...................... None + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_energy ...................................... False + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. True + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... None + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0001 + lr_decay_iters .................................. 8 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 5 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... False + max_position_embeddings ......................... 32768 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 1e-05 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... False + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.01 + moe_capacity_monitor ............................ True + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ False + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. False + moe_enable_hybridep ............................. True + moe_enable_ultraep .............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 896 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ None + moe_num_redundant_experts_per_rank .............. 16 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.095703125 + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ fp32 + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 32 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + moe_ultraep_max_microbatches .................... None + moe_ultraep_overlap_weight_sync ................. True + moe_ultraep_weight_sync_join .................... before_dispatch + moe_upcycling_granularity ....................... 1 + moe_use_device_initiated_grouped_gemm ........... True + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 0.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... RMSNorm + num_attention_heads ............................. 32 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 128 + num_layers ...................................... 16 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 8 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. True + overlap_moe_expert_parallel_comm ................ False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ True + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.bfloat16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_layout .................. None + pipeline_model_parallel_size .................... 1 + position_embedding_type ......................... rope + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... validate_results + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + reuse_grad_buf_for_mxfp8_param_ag ............... False + rope_scaling_factor ............................. 8.0 + rope_type ....................................... None + rotary_base ..................................... 1000000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ None + save_interval ................................... None + save_retain_interval ............................ None + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 4096 + sequence_parallel ............................... False + sft ............................................. False + sft_tokenizer_prompt_format ..................... nemotron-h-aligned + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... 100,0,0 + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... True + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. True + tensor_model_parallel_size ...................... 1 + tensorboard_dir ................................. None + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 12 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. True + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... True + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2 +> building GPT2BPETokenizer tokenizer ... + > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304) +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS +> initializing torch distributed ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +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 +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +[Gloo] Rank [Gloo] Rank 1 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 : 7 + +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank [Gloo] Rank [Gloo] Rank 257 is connected to is connected to is connected to 777 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 777 + + +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +> initialized tensor model parallel with size 1 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +[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 + +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[rank6]:[W830 07:34:57.857321504 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. +[rank1]:[W830 07:34:57.857323995 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. +[rank5]:[W830 07:34:57.857329590 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. +[rank4]:[W830 07:34:57.857374155 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. +[rank2]:[W830 07:34:57.857414867 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. +[rank3]:[W830 07:34:57.857429948 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. +[rank7]:[W830 07:34:57.857441563 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.033 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +[rank0]:[W830 07:34:57.018291220 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. +>>> done with compiling and loading fused kernels. Compilation time: 4.002 seconds +time to initialize megatron (seconds): 7.633 +[after megatron is initialized] datetime: 2026-08-30 07:35:03 +building GPT model ... +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +[INFO] Use MNNVL fabric: False +[WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8. +NVSHMEM v3.7.2 +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive). + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6728847360 +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) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18 +Params for bucket 1 (206045184 elements, 206045184 padded size): + module.output_layer.weight +Params for bucket 2 (42479616 elements, 42479616 padded size): + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.final_layernorm.weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.15.self_attention.linear_proj.weight +Params for bucket 3 (42475520 elements, 42475520 padded size): + module.decoder.layers.14.self_attention.linear_qkv.weight + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.14.self_attention.linear_proj.weight +Params for bucket 4 (42475520 elements, 42475520 padded size): + module.decoder.layers.13.pre_mlp_layernorm.weight + module.decoder.layers.13.self_attention.linear_qkv.weight + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight +Params for bucket 5 (42475520 elements, 42475520 padded size): + module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.12.self_attention.linear_proj.weight +Params for bucket 6 (42475520 elements, 42475520 padded size): + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_qkv.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight +Params for bucket 7 (42475520 elements, 42475520 padded size): + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.10.self_attention.linear_proj.weight +Params for bucket 8 (42475520 elements, 42475520 padded size): + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.pre_mlp_layernorm.weight +Params for bucket 9 (42475520 elements, 42475520 padded size): + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.8.pre_mlp_layernorm.weight +Params for bucket 10 (42475520 elements, 42475520 padded size): + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_qkv.weight +Params for bucket 11 (42475520 elements, 42475520 padded size): + module.decoder.layers.6.self_attention.linear_qkv.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight +Params for bucket 12 (42475520 elements, 42475520 padded size): + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.pre_mlp_layernorm.weight +Params for bucket 13 (42475520 elements, 42475520 padded size): + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.pre_mlp_layernorm.weight +Params for bucket 14 (42475520 elements, 42475520 padded size): + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.3.self_attention.linear_proj.weight +Params for bucket 15 (42475520 elements, 42475520 padded size): + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_qkv.weight + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight +Params for bucket 16 (42475520 elements, 42475520 padded size): + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight +Params for bucket 17 (42475520 elements, 42475520 padded size): + module.decoder.layers.0.pre_mlp_layernorm.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.mlp.router.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight +Params for bucket 18 (206045184 elements, 206045184 padded size): + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 67 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight12 + module.decoder.layers.15.mlp.experts.linear_fc2.weight14 + module.decoder.layers.15.mlp.experts.linear_fc2.weight10 + module.decoder.layers.15.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight15 + module.decoder.layers.15.mlp.experts.linear_fc2.weight11 + module.decoder.layers.15.mlp.experts.linear_fc2.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight9 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 2 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight15 + module.decoder.layers.15.mlp.experts.linear_fc1.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight3 + module.decoder.layers.15.mlp.experts.linear_fc2.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight14 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight11 + module.decoder.layers.15.mlp.experts.linear_fc1.weight9 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight12 + module.decoder.layers.15.mlp.experts.linear_fc1.weight10 + module.decoder.layers.15.mlp.experts.linear_fc1.weight8 +Params for bucket 4 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight12 + module.decoder.layers.14.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + module.decoder.layers.14.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc2.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight11 + module.decoder.layers.14.mlp.experts.linear_fc2.weight9 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight14 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight11 + module.decoder.layers.14.mlp.experts.linear_fc1.weight9 + module.decoder.layers.14.mlp.experts.linear_fc1.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight12 + module.decoder.layers.14.mlp.experts.linear_fc1.weight8 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 + module.decoder.layers.13.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + module.decoder.layers.13.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight13 + module.decoder.layers.13.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc2.weight9 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight8 + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 + module.decoder.layers.13.mlp.experts.linear_fc1.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight5 + module.decoder.layers.13.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight3 + module.decoder.layers.13.mlp.experts.linear_fc2.weight1 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight13 + module.decoder.layers.13.mlp.experts.linear_fc1.weight9 + module.decoder.layers.13.mlp.experts.linear_fc1.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight12 + module.decoder.layers.13.mlp.experts.linear_fc1.weight10 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight5 + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + module.decoder.layers.13.mlp.experts.linear_fc1.weight3 + module.decoder.layers.13.mlp.experts.linear_fc1.weight8 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight14 + module.decoder.layers.12.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc2.weight11 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight10 + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 + module.decoder.layers.12.mlp.experts.linear_fc2.weight8 + module.decoder.layers.12.mlp.experts.linear_fc2.weight4 + module.decoder.layers.12.mlp.experts.linear_fc2.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight9 + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 + module.decoder.layers.12.mlp.experts.linear_fc2.weight3 + module.decoder.layers.12.mlp.experts.linear_fc2.weight1 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight13 + module.decoder.layers.12.mlp.experts.linear_fc1.weight11 + module.decoder.layers.12.mlp.experts.linear_fc1.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight12 + module.decoder.layers.12.mlp.experts.linear_fc1.weight10 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight9 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight8 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 +Params for bucket 17 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc2.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight13 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight12 + module.decoder.layers.11.mlp.experts.linear_fc2.weight10 + module.decoder.layers.11.mlp.experts.linear_fc2.weight8 + module.decoder.layers.11.mlp.experts.linear_fc2.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight11 + module.decoder.layers.11.mlp.experts.linear_fc2.weight9 + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 +Params for bucket 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight15 + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight13 + module.decoder.layers.11.mlp.experts.linear_fc1.weight11 + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight14 + module.decoder.layers.11.mlp.experts.linear_fc1.weight12 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight9 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 + module.decoder.layers.11.mlp.experts.linear_fc1.weight10 + module.decoder.layers.11.mlp.experts.linear_fc1.weight8 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight15 +Params for bucket 22 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight14 + module.decoder.layers.10.mlp.experts.linear_fc2.weight10 + module.decoder.layers.10.mlp.experts.linear_fc2.weight12 + module.decoder.layers.10.mlp.experts.linear_fc2.weight8 + module.decoder.layers.10.mlp.experts.linear_fc2.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight4 + module.decoder.layers.10.mlp.experts.linear_fc2.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight11 + module.decoder.layers.10.mlp.experts.linear_fc2.weight9 + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight15 + module.decoder.layers.10.mlp.experts.linear_fc1.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight14 + module.decoder.layers.10.mlp.experts.linear_fc1.weight12 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight11 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.10.mlp.experts.linear_fc1.weight9 + module.decoder.layers.10.mlp.experts.linear_fc1.weight10 + module.decoder.layers.10.mlp.experts.linear_fc1.weight8 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 +Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight5 + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 + module.decoder.layers.10.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight12 + module.decoder.layers.9.mlp.experts.linear_fc2.weight14 + module.decoder.layers.9.mlp.experts.linear_fc2.weight10 + module.decoder.layers.9.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight15 + module.decoder.layers.9.mlp.experts.linear_fc2.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight11 + module.decoder.layers.9.mlp.experts.linear_fc2.weight9 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 +Params for bucket 27 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight14 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight15 + module.decoder.layers.9.mlp.experts.linear_fc1.weight13 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight12 + module.decoder.layers.9.mlp.experts.linear_fc1.weight10 + module.decoder.layers.9.mlp.experts.linear_fc1.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight11 + module.decoder.layers.9.mlp.experts.linear_fc1.weight9 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 +Params for bucket 29 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 +Params for bucket 30 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight14 + module.decoder.layers.8.mlp.experts.linear_fc2.weight12 + module.decoder.layers.8.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc2.weight8 + module.decoder.layers.8.mlp.experts.linear_fc2.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight13 + module.decoder.layers.8.mlp.experts.linear_fc2.weight11 + module.decoder.layers.8.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 +Params for bucket 31 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight14 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight15 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight12 + module.decoder.layers.8.mlp.experts.linear_fc1.weight8 + module.decoder.layers.8.mlp.experts.linear_fc1.weight13 + module.decoder.layers.8.mlp.experts.linear_fc1.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight11 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight15 + module.decoder.layers.7.mlp.experts.linear_fc2.weight13 + module.decoder.layers.7.mlp.experts.linear_fc2.weight11 + module.decoder.layers.7.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc2.weight10 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.7.mlp.experts.linear_fc2.weight3 + module.decoder.layers.7.mlp.experts.linear_fc2.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight8 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 + module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.7.mlp.experts.linear_fc2.weight2 + module.decoder.layers.7.mlp.experts.linear_fc2.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight15 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight10 + module.decoder.layers.7.mlp.experts.linear_fc1.weight13 + module.decoder.layers.7.mlp.experts.linear_fc1.weight11 + module.decoder.layers.7.mlp.experts.linear_fc1.weight9 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight8 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight15 + module.decoder.layers.6.mlp.experts.linear_fc2.weight13 + module.decoder.layers.6.mlp.experts.linear_fc2.weight11 + module.decoder.layers.6.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight12 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight9 + module.decoder.layers.6.mlp.experts.linear_fc2.weight5 + module.decoder.layers.6.mlp.experts.linear_fc2.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight10 + module.decoder.layers.6.mlp.experts.linear_fc2.weight8 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight13 + module.decoder.layers.6.mlp.experts.linear_fc1.weight11 + module.decoder.layers.6.mlp.experts.linear_fc1.weight12 + module.decoder.layers.6.mlp.experts.linear_fc1.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight10 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight9 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight8 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 42 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight13 +Params for bucket 43 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight12 + module.decoder.layers.5.mlp.experts.linear_fc2.weight8 + module.decoder.layers.5.mlp.experts.linear_fc2.weight10 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight9 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight13 + module.decoder.layers.5.mlp.experts.linear_fc1.weight15 + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight14 + module.decoder.layers.5.mlp.experts.linear_fc1.weight12 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight9 + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight10 + module.decoder.layers.5.mlp.experts.linear_fc1.weight8 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 +Params for bucket 46 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc2.weight15 +Params for bucket 47 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight14 + module.decoder.layers.4.mlp.experts.linear_fc2.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight10 + module.decoder.layers.4.mlp.experts.linear_fc2.weight8 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 + module.decoder.layers.4.mlp.experts.linear_fc2.weight11 + module.decoder.layers.4.mlp.experts.linear_fc2.weight13 + module.decoder.layers.4.mlp.experts.linear_fc2.weight9 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight14 + module.decoder.layers.4.mlp.experts.linear_fc1.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight15 + module.decoder.layers.4.mlp.experts.linear_fc1.weight13 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight10 + module.decoder.layers.4.mlp.experts.linear_fc1.weight8 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight11 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight9 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 +Params for bucket 51 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight15 + module.decoder.layers.3.mlp.experts.linear_fc2.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight11 + module.decoder.layers.3.mlp.experts.linear_fc2.weight9 + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + module.decoder.layers.3.mlp.experts.linear_fc2.weight12 + module.decoder.layers.3.mlp.experts.linear_fc2.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight10 + module.decoder.layers.3.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 +Params for bucket 52 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight15 + module.decoder.layers.3.mlp.experts.linear_fc1.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc1.weight14 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight11 + module.decoder.layers.3.mlp.experts.linear_fc1.weight9 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 + module.decoder.layers.3.mlp.experts.linear_fc1.weight12 + module.decoder.layers.3.mlp.experts.linear_fc1.weight10 + module.decoder.layers.3.mlp.experts.linear_fc1.weight8 +Params for bucket 54 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight5 + module.decoder.layers.3.mlp.experts.linear_fc1.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 +Params for bucket 55 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight12 + module.decoder.layers.2.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight13 + module.decoder.layers.2.mlp.experts.linear_fc2.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight11 + module.decoder.layers.2.mlp.experts.linear_fc2.weight9 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 +Params for bucket 56 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight14 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight13 + module.decoder.layers.2.mlp.experts.linear_fc1.weight9 + module.decoder.layers.2.mlp.experts.linear_fc1.weight11 + module.decoder.layers.2.mlp.experts.linear_fc1.weight12 + module.decoder.layers.2.mlp.experts.linear_fc1.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight8 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 +Params for bucket 59 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight14 + module.decoder.layers.1.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight13 + module.decoder.layers.1.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc2.weight9 +Params for bucket 60 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight8 + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight15 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight12 + module.decoder.layers.1.mlp.experts.linear_fc1.weight10 + module.decoder.layers.1.mlp.experts.linear_fc1.weight13 + module.decoder.layers.1.mlp.experts.linear_fc1.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight9 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight8 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + module.decoder.layers.1.mlp.experts.linear_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 +Params for bucket 63 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight14 + module.decoder.layers.0.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight15 + module.decoder.layers.0.mlp.experts.linear_fc2.weight13 + module.decoder.layers.0.mlp.experts.linear_fc2.weight11 +Params for bucket 64 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight10 + module.decoder.layers.0.mlp.experts.linear_fc2.weight8 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight9 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 +Params for bucket 65 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight12 + module.decoder.layers.0.mlp.experts.linear_fc1.weight14 + module.decoder.layers.0.mlp.experts.linear_fc1.weight10 + module.decoder.layers.0.mlp.experts.linear_fc1.weight15 + module.decoder.layers.0.mlp.experts.linear_fc1.weight11 + module.decoder.layers.0.mlp.experts.linear_fc1.weight13 +Params for bucket 66 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight8 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 + module.decoder.layers.0.mlp.experts.linear_fc1.weight9 + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 +Params for bucket 67 (29360128 elements, 29360128 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + module.decoder.layers.0.mlp.experts.linear_fc1.weight2 + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight1 +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=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +[after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:35:07 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + validation: 0 + test: 0 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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) +INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers +INFO:megatron.core.datasets.indexed_dataset: Extract the document indices +INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029 +INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350 +INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices +INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-30 07:35:07 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (3910.47, 3922.97) + train/valid/test-data-iterators-setup ..........: (73.58, 95.27) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 07:35:07 +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") + [2026-08-30 07:35:34] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 27093.9 | throughput per GPU (TFLOP/s/GPU): 12.3 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102500E+01 | loss scale: 1.0 | grad norm: 44.121 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 133888 | min safe capacity: 1.0215 |Number of parameters in transformer block in billions: 23.22 + +Number of parameters in embedding layers in billions: 0.41 +Total number of parameters in billions: 23.63 +Number of parameters in most loaded shard in billions: 23.6320 +Theoretical memory footprints: weight and optimizer=169029.43 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 58109.861328125 | max allocated: 83199.2578125 | reserved: 85038.0 | max reserved: 85038.0 + [2026-08-30 07:35:35] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 674.8 | throughput per GPU (TFLOP/s/GPU): 494.1 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101850E+01 | loss scale: 1.0 | grad norm: 45.886 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:36] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 514.4 | throughput per GPU (TFLOP/s/GPU): 648.1 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.060148E+01 | loss scale: 1.0 | grad norm: 216.512 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 07:35:38] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2509.8 | throughput per GPU (TFLOP/s/GPU): 132.8 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.301596E+01 | loss scale: 1.0 | grad norm: 46.053 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:39] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 476.6 | throughput per GPU (TFLOP/s/GPU): 699.5 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.033482E+01 | loss scale: 1.0 | grad norm: 17.345 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:39] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 484.9 | throughput per GPU (TFLOP/s/GPU): 687.6 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.016502E+01 | loss scale: 1.0 | grad norm: 17.634 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:40] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 479.5 | throughput per GPU (TFLOP/s/GPU): 695.2 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.899235E+00 | loss scale: 1.0 | grad norm: 6.181 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:40] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 484.1 | throughput per GPU (TFLOP/s/GPU): 688.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.319888E+00 | loss scale: 1.0 | grad norm: 5.520 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:41] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 479.3 | throughput per GPU (TFLOP/s/GPU): 695.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.068393E+00 | loss scale: 1.0 | grad norm: 5.392 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:41] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 474.5 | throughput per GPU (TFLOP/s/GPU): 702.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.914932E+00 | loss scale: 1.0 | grad norm: 5.197 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:42] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 485.1 | throughput per GPU (TFLOP/s/GPU): 687.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.803726E+00 | loss scale: 1.0 | grad norm: 4.961 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:35:42] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 490.6 | throughput per GPU (TFLOP/s/GPU): 679.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.686855E+00 | loss scale: 1.0 | grad norm: 4.905 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | +[after training is done] datetime: 2026-08-30 07:35:42 +[rank0]:[W830 07:35:43.794089223 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()) +[rank4]:[W830 07:35:45.312410786 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()) +[rank5]:[W830 07:35:45.374710140 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()) +[rank2]:[W830 07:35:45.374777612 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()) +[rank1]:[W830 07:35:45.376260309 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()) +[rank6]:[W830 07:35:45.376626907 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()) +[rank3]:[W830 07:35:45.376929325 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()) +[rank7]:[W830 07:35:45.891230243 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()) +W0830 07:35:48.142000 778522 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0830 07:35:48.143000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778610 closing signal SIGTERM +W0830 07:35:48.143000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778611 closing signal SIGTERM +W0830 07:35:48.144000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778612 closing signal SIGTERM +W0830 07:35:48.145000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778613 closing signal SIGTERM +W0830 07:35:48.146000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778614 closing signal SIGTERM +W0830 07:35:48.146000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778615 closing signal SIGTERM +W0830 07:35:48.147000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778616 closing signal SIGTERM +W0830 07:35:48.148000 778522 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 778617 closing signal SIGTERM +Traceback (most recent call last): + File "/usr/local/bin/torchrun", line 8, in + sys.exit(main()) + ^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper + return f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main + run(args) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run + elastic_launch( + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ + return launch_agent(self._config, self._entrypoint, list(args)) + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent + result = agent.run() + ^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper + result = f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run + result = self._invoke_run(role) + ^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run + time.sleep(monitor_interval) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler + raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) +torch.distributed.elastic.multiprocessing.api.SignalException: Process 778522 got signal: 15 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/mem_samples.csv b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..4527f6d583894ee21c5aac0aa62cbc2f22239174 --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/mem_samples.csv @@ -0,0 +1,232 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 2 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 819 +2, 819 +3, 819 +4, 819 +5, 819 +6, 819 +7, 819 +0, 1537 +1, 1537 +2, 1537 +3, 1537 +4, 1537 +5, 1537 +6, 1537 +7, 1537 +0, 2231 +1, 2231 +2, 2231 +3, 2231 +4, 2231 +5, 2231 +6, 2231 +7, 2231 +0, 2903 +1, 2903 +2, 2903 +3, 2903 +4, 2903 +5, 2903 +6, 2903 +7, 2903 +0, 42429 +1, 42877 +2, 42877 +3, 42877 +4, 42429 +5, 42877 +6, 42877 +7, 42877 +0, 43129 +1, 43577 +2, 43577 +3, 43577 +4, 43129 +5, 43577 +6, 43577 +7, 43577 +0, 44463 +1, 44911 +2, 44911 +3, 44911 +4, 44463 +5, 44911 +6, 44911 +7, 44911 +0, 44463 +1, 44911 +2, 44911 +3, 44911 +4, 44463 +5, 44911 +6, 44911 +7, 44911 +0, 45585 +1, 46033 +2, 46033 +3, 46033 +4, 45585 +5, 46033 +6, 46033 +7, 46033 +0, 48337 +1, 48785 +2, 48785 +3, 48785 +4, 48337 +5, 48785 +6, 48785 +7, 48785 +0, 79697 +1, 79713 +2, 79777 +3, 79713 +4, 79717 +5, 79777 +6, 79777 +7, 79777 +0, 79825 +1, 79841 +2, 79841 +3, 79841 +4, 79845 +5, 79841 +6, 79841 +7, 79841 +0, 80949 +1, 80965 +2, 80965 +3, 80965 +4, 80969 +5, 80965 +6, 80965 +7, 80965 +0, 82223 +1, 80979 +2, 80979 +3, 82101 +4, 82243 +5, 80979 +6, 80979 +7, 80979 +0, 92791 +1, 92795 +2, 92839 +3, 92841 +4, 92837 +5, 92845 +6, 92901 +7, 92831 +0, 93621 +1, 93901 +2, 93641 +3, 93961 +4, 93937 +5, 93889 +6, 93887 +7, 93885 +0, 117479 +1, 117613 +2, 117613 +3, 117613 +4, 117577 +5, 117613 +6, 117613 +7, 117613 +0, 125871 +1, 121721 +2, 121745 +3, 126861 +4, 125655 +5, 122485 +6, 122461 +7, 123823 +0, 132221 +1, 132437 +2, 132435 +3, 132435 +4, 132221 +5, 132435 +6, 132435 +7, 132435 +0, 132221 +1, 132437 +2, 132435 +3, 132435 +4, 132221 +5, 132435 +6, 132435 +7, 132435 +0, 131659 +1, 132077 +2, 131891 +3, 131553 +4, 131921 +5, 131809 +6, 131583 +7, 131803 +0, 127027 +1, 126855 +2, 127305 +3, 127887 +4, 127347 +5, 127241 +6, 127629 +7, 127241 +0, 125029 +1, 125175 +2, 125561 +3, 125705 +4, 125347 +5, 125173 +6, 125561 +7, 125173 +0, 4 +1, 4 +2, 4 +3, 125515 +4, 4 +5, 4 +6, 4 +7, 4 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/run.env b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/run.env new file mode 100644 index 0000000000000000000000000000000000000000..37dca56d5f2930f6fa9c4c0dfb76169a3b8cbd3d --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/run.env @@ -0,0 +1,70 @@ +backend=ultraep +config=G4 +family=tierA +repo=/workspace/Megatron-UltraEP +branch=ultraep-bracing-regions-through-kernels +commit=03d57ef7 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=896 +num_experts=128 +topk=32 +experts_per_rank=16 +expert_params=11010048 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=131072 +mean_tokens_per_expert=8192 +slot_policy=matched +slots_per_rank=16 +capacity_mode=static +capacity=1.095703125 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens=1024 +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/train.log b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/train.log new file mode 100644 index 0000000000000000000000000000000000000000..ab4dd5d56ac6acba7b934ef74fd6d9939acd0e01 --- /dev/null +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/train.log @@ -0,0 +1,1570 @@ +============================================================== + G4 / ultraep (slots/rank=16 policy=matched) + E=128 K=32 H'=896 m=16 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.095703125 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-UltraEP @ 03d57ef7 +============================================================== +W0830 07:35:54.659000 781569 torch/distributed/run.py:766] +W0830 07:35:54.659000 781569 torch/distributed/run.py:766] ***************************************** +W0830 07:35:54.659000 781569 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. +W0830 07:35:54.659000 781569 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +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 +WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer +Number of virtual stages per pipeline stage: None +accumulate and all-reduce gradients in fp32 for bfloat16 data type. +using torch.bfloat16 for parameters ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +------------------------ arguments ------------------------ + account_for_embedding_in_pipeline_split ......... False + account_for_loss_in_pipeline_split .............. False + accumulate_allreduce_grads_in_fp32 .............. True + adam_beta1 ...................................... 0.9 + adam_beta2 ...................................... 0.999 + adam_eps ........................................ 1e-08 + add_bias_linear ................................. False + add_position_embedding .......................... False + add_qkv_bias .................................... False + adlr_autoresume ................................. False + adlr_autoresume_interval ........................ 1000 + align_grad_reduce ............................... True + align_param_gather .............................. False + app_tag_run_name ................................ None + app_tag_run_version ............................. 0.0.0 + apply_layernorm_1p .............................. False + apply_query_key_layer_scaling ................... False + apply_residual_connection_post_layernorm ........ False + apply_rope_fusion ............................... True + async_save ...................................... None + async_tensor_model_parallel_allreduce ........... True + attention_backend ............................... AttnBackend.auto + attention_dropout ............................... 0.0 + attention_softmax_in_fp32 ....................... False + auto_detect_ckpt_format ......................... False + barrier_with_L1_time ............................ True + bert_binary_head ................................ True + bert_embedder_type .............................. megatron + bert_load ....................................... None + bf16 ............................................ True + bias_dropout_fusion ............................. True + bias_gelu_fusion ................................ False + bias_swiglu_fusion .............................. True + biencoder_projection_dim ........................ 0 + biencoder_shared_query_context_model ............ False + block_data_path ................................. None + calc_ft_timeouts ................................ False + calculate_per_token_loss ........................ False + check_for_large_grads ........................... False + check_for_nan_in_loss_and_grad .................. False + check_for_spiky_loss ............................ False + check_weight_hash_across_dp_replicas_interval ... None + ckpt_assume_constant_structure .................. False + ckpt_convert_format ............................. None + ckpt_convert_save ............................... None + ckpt_convert_update_legacy_dist_opt_format ...... False + ckpt_format ..................................... torch_dist + ckpt_fully_parallel_load ........................ False + ckpt_fully_parallel_save ........................ True + ckpt_fully_parallel_save_deprecated ............. False + ckpt_step ....................................... None + classes_fraction ................................ 1.0 + clip_grad ....................................... 1.0 + clone_scatter_output_in_embedding ............... True + config_logger_dir ............................... + consumed_train_samples .......................... 0 + consumed_valid_samples .......................... 0 + context_parallel_size ........................... 1 + cp_comm_type .................................... ['p2p'] + create_attention_mask_in_dataloader ............. True + cross_entropy_fusion_impl ....................... native + cross_entropy_loss_fusion ....................... False + cuda_graph_scope ................................ full_iteration + cuda_graph_warmup_steps ......................... 3 + data_args_path .................................. None + data_cache_path ................................. None + data_parallel_random_init ....................... False + data_parallel_sharding_strategy ................. no_shard + data_parallel_size .............................. 8 + data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document'] + data_per_class_fraction ......................... 1.0 + data_sharding ................................... True + dataloader_type ................................. single + ddp_average_in_collective ....................... False + ddp_bucket_size ................................. None + ddp_num_buckets ................................. None + ddp_pad_buckets_for_high_nccl_busbw ............. False + decoder_first_pipeline_num_layers ............... None + decoder_last_pipeline_num_layers ................ None + decoder_num_layers .............................. None + decoder_seq_length .............................. None + decoupled_lr .................................... None + decoupled_min_lr ................................ None + decrease_batch_size_if_needed ................... False + defer_embedding_wgrad_compute ................... False + delay_wgrad_compute ............................. False + deprecated_use_mcore_models ..................... True + deterministic_mode .............................. False + dino_bottleneck_size ............................ 256 + dino_freeze_last_layer .......................... 1 + dino_head_hidden_size ........................... 2048 + dino_local_crops_number ......................... 10 + dino_local_img_size ............................. 96 + dino_norm_last_layer ............................ False + dino_teacher_temp ............................... 0.07 + dino_warmup_teacher_temp ........................ 0.04 + dino_warmup_teacher_temp_epochs ................. 30 + disable_bf16_reduced_precision_matmul ........... False + disable_mamba_mem_eff_path ...................... False + disable_straggler_on_startup .................... False + dist_ckpt_format_deprecated ..................... None + dist_ckpt_strictness ............................ assume_ok_unexpected + distribute_saved_activations .................... False + distributed_backend ............................. nccl + distributed_timeout_minutes ..................... 10 + embedding_init_method_std ....................... None + embedding_path .................................. None + empty_unused_memory_level ....................... 0 + enable_cuda_graph ............................... True + enable_experimental ............................. False + enable_ft_package ............................... False + enable_gloo_process_groups ...................... True + enable_msc ...................................... True + enable_one_logger ............................... True + encoder_num_layers .............................. 16 + encoder_seq_length .............................. 4096 + end_weight_decay ................................ 0.1 + eod_mask_loss ................................... False + error_injection_rate ............................ 0 + error_injection_type ............................ transient_error + eval_interval ................................... 1000000 + eval_iters ...................................... 0 + evidence_data_path .............................. None + exit_duration_in_mins ........................... None + exit_interval ................................... None + exit_on_missing_checkpoint ...................... False + exit_signal_handler ............................. False + exp_avg_dtype ................................... torch.float32 + exp_avg_sq_dtype ................................ torch.float32 + expert_model_parallel_size ...................... 8 + expert_tensor_parallel_size ..................... 1 + export_force_local_attention .................... False + export_kd_cfg ................................... None + export_kd_teacher_ckpt_format ................... None + export_kd_teacher_load .......................... None + export_kv_cache_quant ........................... False + export_legacy_megatron .......................... False + export_model_type ............................... GPTModel + export_moe_apply_probs_on_input ................. False + export_qk_l2_norm ............................... False + export_quant_cfg ................................ None + export_real_quant_cfg ........................... None + export_te_mcore_model ........................... False + external_cuda_graph ............................. False + ffn_hidden_size ................................. 14336 + finetune ........................................ False + finetune_data_split ............................. train + finetune_hf_dataset ............................. None + first_last_layers_bf16 .......................... False + flash_decode .................................... False + fp16 ............................................ False + fp16_lm_cross_entropy ........................... False + fp32_residual_connection ........................ False + fp8 ............................................. e4m3 + fp8_amax_compute_algo ........................... most_recent + fp8_amax_history_len ............................ 1 + fp8_interval .................................... 1 + fp8_margin ...................................... 0 + fp8_param_gather ................................ False + fp8_recipe ...................................... mxfp8 + fp8_wgrad ....................................... True + fsdp_double_buffer .............................. False + global_batch_size ............................... 16 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... True + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 4096 + hierarchical_context_parallel_sizes ............. None + high_priority_stream_groups ..................... [] + hybrid_attention_ratio .......................... 0.0 + hybrid_mlp_ratio ................................ 0.0 + hybrid_override_pattern ......................... None + hysteresis ...................................... 2 + ict_head_size ................................... None + ict_load ........................................ None + img_h ........................................... 224 + img_w ........................................... 224 + indexer_batch_size .............................. 128 + indexer_log_interval ............................ 1000 + inference_batch_times_seqlen_threshold .......... -1 + inference_dynamic_batching ...................... False + inference_dynamic_batching_buffer_guaranteed_fraction 0.2 + inference_dynamic_batching_buffer_overflow_factor None + inference_dynamic_batching_buffer_size_gb ....... 40.0 + inference_dynamic_batching_chunk_size ........... 256 + inference_dynamic_batching_max_requests_override None + inference_dynamic_batching_max_tokens_override .. None + inference_dynamic_batching_num_cuda_graphs ...... 16 + inference_max_batch_size ........................ 8 + inference_max_seq_length ........................ 2560 + inference_rng_tracker ........................... False + init_method_std ................................. 0.01 + init_method_xavier_uniform ...................... False + init_model_with_meta_device ..................... False + initial_loss_scale .............................. 4294967296 + inprocess_active_world_size ..................... 8 + inprocess_barrier_timeout ....................... 120 + inprocess_completion_timeout .................... 120 + inprocess_empty_cuda_cache ...................... False + inprocess_granularity ........................... node + inprocess_hard_timeout .......................... 90 + inprocess_heartbeat_interval .................... 30 + inprocess_heartbeat_timeout ..................... 60 + inprocess_last_call_wait ........................ 1 + inprocess_max_iterations ........................ None + inprocess_monitor_process_interval .............. 1.0 + inprocess_monitor_thread_interval ............... 1.0 + inprocess_progress_watchdog_interval ............ 1.0 + inprocess_restart ............................... False + inprocess_soft_timeout .......................... 60 + inprocess_termination_grace_time ................ 1 + is_hybrid_model ................................. False + iter_per_epoch .................................. 1250 + iterations_to_skip .............................. [] + keep_fp8_transpose_cache_when_using_custom_fsdp . False + kitchen_config_file ............................. None + kitchen_recipe_number ........................... None + kv_channels ..................................... 128 + kv_lora_rank .................................... 32 + lazy_mpu_init ................................... None + load ............................................ None + load_main_params_from_ckpt ...................... None + load_model_opt_format ........................... False + local_rank ...................................... 0 + log_energy ...................................... False + log_interval .................................... 1 + log_loss_scale_to_tensorboard ................... True + log_memory_to_tensorboard ....................... False + log_num_zeros_in_grad ........................... False + log_params_norm ................................. False + log_progress .................................... False + log_straggler ................................... False + log_throughput .................................. True + log_timers_to_tensorboard ....................... False + log_validation_ppl_to_tensorboard ............... False + log_world_size_to_tensorboard ................... False + logging_level ................................... None + loss_scale ...................................... None + loss_scale_window ............................... 1000 + lr .............................................. 0.0001 + lr_decay_iters .................................. 8 + lr_decay_samples ................................ None + lr_decay_style .................................. cosine + lr_warmup_fraction .............................. None + lr_warmup_init .................................. 0.0 + lr_warmup_iters ................................. 5 + lr_warmup_samples ............................... 0 + lr_wsd_decay_iters .............................. None + lr_wsd_decay_samples ............................ None + lr_wsd_decay_style .............................. exponential + main_grads_dtype ................................ torch.float32 + main_params_dtype ............................... torch.float32 + make_vocab_size_divisible_by .................... 128 + mamba_head_dim .................................. 64 + mamba_num_groups ................................ 8 + mamba_num_heads ................................. None + mamba_state_dim ................................. 128 + manual_gc ....................................... False + manual_gc_eval .................................. True + manual_gc_interval .............................. 0 + mask_factor ..................................... 1.0 + mask_prob ....................................... 0.15 + mask_type ....................................... random + masked_softmax_fusion ........................... False + max_position_embeddings ......................... 32768 + max_tokens_to_oom ............................... 12000 + memory_snapshot_path ............................ snapshot.pickle + merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt + micro_batch_size ................................ 1 + microbatch_group_size_per_vp_stage .............. None + mid_level_dataset_surplus ....................... 0.005 + min_loss_scale .................................. 1.0 + min_lr .......................................... 1e-05 + mlp_chunks_for_prefill .......................... 1 + mmap_bin_files .................................. True + mock_data ....................................... False + moe_apply_probs_on_input ........................ False + moe_aux_loss_coeff .............................. 0.01 + moe_capacity_monitor ............................ True + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ False + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. False + moe_enable_hybridep ............................. True + moe_enable_ultraep .............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 896 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. 1 + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ None + moe_num_redundant_experts_per_rank .............. 16 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.095703125 + moe_router_bias_update_rate ..................... 0.001 + moe_router_dtype ................................ fp32 + moe_router_enable_expert_bias ................... False + moe_router_force_load_balancing ................. False + moe_router_group_topk ........................... None + moe_router_load_balancing_type .................. aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... False + moe_router_score_function ....................... softmax + moe_router_topk ................................. 32 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. None + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + moe_ultraep_max_microbatches .................... None + moe_ultraep_overlap_weight_sync ................. True + moe_ultraep_weight_sync_join .................... before_dispatch + moe_upcycling_granularity ....................... 1 + moe_use_device_initiated_grouped_gemm ........... True + moe_use_legacy_grouped_gemm ..................... False + moe_use_upcycling ............................... False + moe_z_loss_coeff ................................ None + mrope_section ................................... None + mscale .......................................... 1.0 + mscale_all_dim .................................. 0.0 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... False + nccl_all_reduce_for_prefill ..................... False + nccl_communicator_config_path ................... None + nccl_ub ......................................... False + no_load_optim ................................... None + no_load_rng ..................................... None + no_persist_layer_norm ........................... False + no_rope_freq .................................... None + no_save_optim ................................... None + no_save_rng ..................................... None + non_persistent_ckpt_type ........................ None + non_persistent_global_ckpt_dir .................. None + non_persistent_local_ckpt_algo .................. fully_parallel + non_persistent_local_ckpt_dir ................... None + non_persistent_save_interval .................... None + norm_epsilon .................................... 1e-05 + normalization ................................... RMSNorm + num_attention_heads ............................. 32 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 128 + num_layers ...................................... 16 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 8 + num_virtual_stages_per_pipeline_rank ............ None + num_workers ..................................... 2 + object_storage_cache_path ....................... None + one_logger_async ................................ False + one_logger_project .............................. megatron-lm + one_logger_run_name ............................. None + onnx_safe ....................................... None + openai_gelu ..................................... False + optimizer ....................................... adam + optimizer_cpu_offload ........................... False + optimizer_offload_fraction ...................... 1.0 + output_bert_embeddings .......................... False + overlap_cpu_optimizer_d2h_h2d ................... False + overlap_grad_reduce ............................. True + overlap_moe_expert_parallel_comm ................ False + overlap_p2p_comm ................................ False + overlap_p2p_comm_warmup_flush ................... False + overlap_param_gather ............................ True + overlap_param_gather_with_optimizer_step ........ False + override_opt_param_scheduler .................... False + params_dtype .................................... torch.bfloat16 + patch_dim ....................................... 16 + per_split_data_args_path ........................ None + perform_initialization .......................... True + pin_cpu_grads ................................... True + pin_cpu_params .................................. True + pipeline_model_parallel_comm_backend ............ None + pipeline_model_parallel_layout .................. None + pipeline_model_parallel_size .................... 1 + position_embedding_type ......................... rope + pretrained_checkpoint ........................... None + profile ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + q_lora_rank ..................................... None + qk_head_dim ..................................... 128 + qk_l2_norm ...................................... False + qk_layernorm .................................... False + qk_pos_emb_head_dim ............................. 64 + query_in_block_prob ............................. 0.1 + rampup_batch_size ............................... None + rank ............................................ 0 + recompute_granularity ........................... None + recompute_method ................................ None + recompute_modules ............................... None + recompute_num_layers ............................ None + record_memory_history ........................... False + relative_attention_max_distance ................. 128 + relative_attention_num_buckets .................. 32 + replication ..................................... False + replication_factor .............................. 2 + replication_jump ................................ None + rerun_mode ...................................... validate_results + reset_attention_mask ............................ False + reset_position_ids .............................. False + result_rejected_tracker_filename ................ None + retriever_report_topk_accuracies ................ [] + retriever_score_scaling ......................... False + retriever_seq_length ............................ 256 + retro_add_retriever ............................. False + retro_attention_gate ............................ 1 + retro_cyclic_train_iters ........................ None + retro_encoder_attention_dropout ................. 0.1 + retro_encoder_hidden_dropout .................... 0.1 + retro_encoder_layers ............................ 2 + retro_num_neighbors ............................. 2 + retro_num_retrieved_chunks ...................... 2 + retro_project_dir ............................... None + retro_verify_neighbor_count ..................... True + reuse_grad_buf_for_mxfp8_param_ag ............... False + rope_scaling_factor ............................. 8.0 + rope_type ....................................... None + rotary_base ..................................... 1000000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 1.0 + rotary_seq_len_interpolation_factor ............. None + run_workload_inspector_server ................... False + sample_rate ..................................... 1.0 + save ............................................ None + save_interval ................................... None + save_retain_interval ............................ None + scatter_gather_tensors_in_pipeline .............. True + seed ............................................ 1234 + seq_length ...................................... 4096 + sequence_parallel ............................... False + sft ............................................. False + sft_tokenizer_prompt_format ..................... nemotron-h-aligned + sgd_momentum .................................... 0.9 + short_seq_prob .................................. 0.1 + skip_train ...................................... False + skipped_train_samples ........................... 0 + spec ............................................ None + split ........................................... 100,0,0 + squared_relu .................................... False + start_weight_decay .............................. 0.1 + straggler_ctrlr_port ............................ 65535 + straggler_minmax_count .......................... 1 + suggested_communication_unit_size ............... None + swiglu .......................................... True + swin_backbone_type .............................. tiny + symmetric_ar_type ............................... None + te_rng_tracker .................................. True + tensor_model_parallel_size ...................... 1 + tensorboard_dir ................................. None + tensorboard_log_interval ........................ 1 + tensorboard_queue_size .......................... 1000 + test_data_path .................................. None + test_mode ....................................... False + tiktoken_num_special_tokens ..................... 1000 + tiktoken_pattern ................................ None + tiktoken_special_tokens ......................... None + timing_log_level ................................ 0 + timing_log_option ............................... minmax + titles_data_path ................................ None + tokenizer_model ................................. None + tokenizer_type .................................. GPT2BPETokenizer + torch_fsdp2_reshard_after_forward ............... True + tp_comm_bootstrap_backend ....................... nccl + tp_comm_bulk_dgrad .............................. True + tp_comm_bulk_wgrad .............................. True + tp_comm_overlap ................................. False + tp_comm_overlap_ag .............................. True + tp_comm_overlap_cfg ............................. None + tp_comm_overlap_rs .............................. True + tp_comm_overlap_rs_dgrad ........................ False + tp_comm_split_ag ................................ True + tp_comm_split_rs ................................ True + train_data_path ................................. None + train_iters ..................................... 12 + train_samples ................................... None + train_sync_interval ............................. None + transformer_impl ................................ transformer_engine + transformer_pipeline_model_parallel_size ........ 1 + untie_embeddings_and_output_weights ............. True + use_checkpoint_args ............................. False + use_checkpoint_opt_param_scheduler .............. False + use_cpu_initialization .......................... None + use_custom_fsdp ................................. False + use_dist_ckpt ................................... True + use_dist_ckpt_deprecated ........................ False + use_distributed_optimizer ....................... True + use_flash_attn .................................. False + use_legacy_models ............................... False + use_mp_args_from_checkpoint_args ................ False + use_one_sent_docs ............................... False + use_persistent_ckpt_worker ...................... False + use_precision_aware_optimizer ................... False + use_pytorch_profiler ............................ False + use_ring_exchange_p2p ........................... False + use_rope_scaling ................................ False + use_rotary_position_embeddings .................. False + use_sharp ....................................... False + use_tokenizer_model_from_checkpoint_args ........ True + use_torch_fsdp2 ................................. False + use_torch_optimizer_for_cpu_offload ............. False + use_tp_pp_dp_mapping ............................ False + v_head_dim ...................................... 128 + valid_data_path ................................. None + variable_seq_lengths ............................ False + virtual_pipeline_model_parallel_size ............ None + vision_backbone_type ............................ vit + vision_pretraining .............................. False + vision_pretraining_type ......................... classify + vocab_extra_ids ................................. 0 + vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json + vocab_size ...................................... None + wandb_exp_name .................................. + wandb_project ................................... + wandb_save_dir .................................. + weight_decay .................................... 0.1 + weight_decay_incr_style ......................... constant + wgrad_deferral_limit ............................ 0 + world_size ...................................... 8 + yaml_cfg ........................................ None +-------------------- end of arguments --------------------- +INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2 +> building GPT2BPETokenizer tokenizer ... + > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304) +WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS +> initializing torch distributed ... +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +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 +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank 3457 is connected to 6 is connected to is connected to is connected to 7 is connected to 777 peer ranks. 7 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 7777 + + + + +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +> initialized tensor model parallel with size 1 +> initialized pipeline model parallel with size 1 +> setting random seeds to 1234 ... +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : [Gloo] Rank 00 + is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[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 + +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[rank5]:[W830 07:36:12.426728407 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. +[rank7]:[W830 07:36:12.428804969 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. +[rank4]:[W830 07:36:12.428938810 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. +[rank3]:[W830 07:36:12.428956742 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. +[rank1]:[W830 07:36:12.428975498 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. +[rank2]:[W830 07:36:12.429298040 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. +[rank6]:[W830 07:36:12.429387302 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.033 seconds +WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. +> compiling and loading fused kernels ... +[rank0]:[W830 07:36:12.587981169 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. +>>> done with compiling and loading fused kernels. Compilation time: 3.705 seconds +time to initialize megatron (seconds): 7.262 +[after megatron is initialized] datetime: 2026-08-30 07:36:18 +building GPT model ... +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +/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. + warnings.warn( +[INFO] Use MNNVL fabric: False +[WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8. +NVSHMEM v3.7.2 +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. + +UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive). + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6728847360 +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) +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18 +Params for bucket 1 (206045184 elements, 206045184 padded size): + module.output_layer.weight +Params for bucket 2 (42479616 elements, 42479616 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.15.self_attention.linear_proj.weight +Params for bucket 3 (42475520 elements, 42475520 padded size): + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_qkv.weight + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight +Params for bucket 4 (42475520 elements, 42475520 padded size): + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.self_attention.linear_qkv.weight + module.decoder.layers.13.pre_mlp_layernorm.weight +Params for bucket 5 (42475520 elements, 42475520 padded size): + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.self_attention.linear_qkv.weight +Params for bucket 6 (42475520 elements, 42475520 padded size): + module.decoder.layers.11.self_attention.linear_qkv.weight + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.11.self_attention.linear_proj.weight +Params for bucket 7 (42475520 elements, 42475520 padded size): + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight +Params for bucket 8 (42475520 elements, 42475520 padded size): + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.9.self_attention.linear_proj.weight +Params for bucket 9 (42475520 elements, 42475520 padded size): + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight +Params for bucket 10 (42475520 elements, 42475520 padded size): + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_qkv.weight + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.7.self_attention.linear_proj.weight +Params for bucket 11 (42475520 elements, 42475520 padded size): + module.decoder.layers.6.self_attention.linear_qkv.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight +Params for bucket 12 (42475520 elements, 42475520 padded size): + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight +Params for bucket 13 (42475520 elements, 42475520 padded size): + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_qkv.weight +Params for bucket 14 (42475520 elements, 42475520 padded size): + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight +Params for bucket 15 (42475520 elements, 42475520 padded size): + module.decoder.layers.2.self_attention.linear_qkv.weight + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight +Params for bucket 16 (42475520 elements, 42475520 padded size): + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight +Params for bucket 17 (42475520 elements, 42475520 padded size): + module.decoder.layers.0.mlp.router.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.pre_mlp_layernorm.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight +Params for bucket 18 (206045184 elements, 206045184 padded size): + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 67 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight14 + module.decoder.layers.15.mlp.experts.linear_fc2.weight12 + module.decoder.layers.15.mlp.experts.linear_fc2.weight10 + module.decoder.layers.15.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight15 + module.decoder.layers.15.mlp.experts.linear_fc2.weight11 + module.decoder.layers.15.mlp.experts.linear_fc2.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight9 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 2 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight15 + module.decoder.layers.15.mlp.experts.linear_fc1.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight1 + module.decoder.layers.15.mlp.experts.linear_fc2.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight14 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight11 + module.decoder.layers.15.mlp.experts.linear_fc1.weight9 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight12 + module.decoder.layers.15.mlp.experts.linear_fc1.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight10 +Params for bucket 4 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight12 + module.decoder.layers.14.mlp.experts.linear_fc2.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + module.decoder.layers.14.mlp.experts.linear_fc2.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc2.weight11 + module.decoder.layers.14.mlp.experts.linear_fc2.weight9 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight14 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight9 + module.decoder.layers.14.mlp.experts.linear_fc1.weight11 + module.decoder.layers.14.mlp.experts.linear_fc1.weight12 + module.decoder.layers.14.mlp.experts.linear_fc1.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight8 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 + module.decoder.layers.13.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + module.decoder.layers.13.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight13 + module.decoder.layers.13.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc2.weight9 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight8 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 + module.decoder.layers.13.mlp.experts.linear_fc2.weight5 + module.decoder.layers.13.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight3 + module.decoder.layers.13.mlp.experts.linear_fc2.weight1 + module.decoder.layers.13.mlp.experts.linear_fc1.weight15 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight12 + module.decoder.layers.13.mlp.experts.linear_fc1.weight10 + module.decoder.layers.13.mlp.experts.linear_fc1.weight13 + module.decoder.layers.13.mlp.experts.linear_fc1.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight9 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight8 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + module.decoder.layers.13.mlp.experts.linear_fc1.weight5 + module.decoder.layers.13.mlp.experts.linear_fc1.weight3 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc2.weight11 + module.decoder.layers.12.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight12 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight9 + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 + module.decoder.layers.12.mlp.experts.linear_fc2.weight3 + module.decoder.layers.12.mlp.experts.linear_fc2.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight10 + module.decoder.layers.12.mlp.experts.linear_fc2.weight8 + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 + module.decoder.layers.12.mlp.experts.linear_fc2.weight4 + module.decoder.layers.12.mlp.experts.linear_fc2.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight0 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight12 + module.decoder.layers.12.mlp.experts.linear_fc1.weight10 + module.decoder.layers.12.mlp.experts.linear_fc1.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight13 + module.decoder.layers.12.mlp.experts.linear_fc1.weight11 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight8 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 + module.decoder.layers.12.mlp.experts.linear_fc1.weight9 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 +Params for bucket 17 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight15 + module.decoder.layers.11.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc2.weight14 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight11 + module.decoder.layers.11.mlp.experts.linear_fc2.weight9 + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 + module.decoder.layers.11.mlp.experts.linear_fc2.weight12 + module.decoder.layers.11.mlp.experts.linear_fc2.weight8 + module.decoder.layers.11.mlp.experts.linear_fc2.weight10 + module.decoder.layers.11.mlp.experts.linear_fc2.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 +Params for bucket 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight13 + module.decoder.layers.11.mlp.experts.linear_fc1.weight15 + module.decoder.layers.11.mlp.experts.linear_fc1.weight11 + module.decoder.layers.11.mlp.experts.linear_fc1.weight14 + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight12 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight9 + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 + module.decoder.layers.11.mlp.experts.linear_fc1.weight10 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 + module.decoder.layers.11.mlp.experts.linear_fc1.weight8 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc2.weight15 +Params for bucket 22 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight14 + module.decoder.layers.10.mlp.experts.linear_fc2.weight12 + module.decoder.layers.10.mlp.experts.linear_fc2.weight10 + module.decoder.layers.10.mlp.experts.linear_fc2.weight8 + module.decoder.layers.10.mlp.experts.linear_fc2.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight4 + module.decoder.layers.10.mlp.experts.linear_fc2.weight11 + module.decoder.layers.10.mlp.experts.linear_fc2.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight9 + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight15 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc1.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight14 + module.decoder.layers.10.mlp.experts.linear_fc1.weight12 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight11 + module.decoder.layers.10.mlp.experts.linear_fc1.weight9 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.10.mlp.experts.linear_fc1.weight8 + module.decoder.layers.10.mlp.experts.linear_fc1.weight10 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 +Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight5 + module.decoder.layers.10.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight14 + module.decoder.layers.9.mlp.experts.linear_fc2.weight12 + module.decoder.layers.9.mlp.experts.linear_fc2.weight10 + module.decoder.layers.9.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight15 + module.decoder.layers.9.mlp.experts.linear_fc2.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight11 + module.decoder.layers.9.mlp.experts.linear_fc2.weight9 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 +Params for bucket 27 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight15 + module.decoder.layers.9.mlp.experts.linear_fc1.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight14 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight11 + module.decoder.layers.9.mlp.experts.linear_fc1.weight9 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight12 + module.decoder.layers.9.mlp.experts.linear_fc1.weight10 + module.decoder.layers.9.mlp.experts.linear_fc1.weight8 +Params for bucket 29 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 +Params for bucket 30 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight12 + module.decoder.layers.8.mlp.experts.linear_fc2.weight14 + module.decoder.layers.8.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight13 + module.decoder.layers.8.mlp.experts.linear_fc2.weight11 + module.decoder.layers.8.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 +Params for bucket 31 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight14 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight13 + module.decoder.layers.8.mlp.experts.linear_fc1.weight11 + module.decoder.layers.8.mlp.experts.linear_fc1.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight12 + module.decoder.layers.8.mlp.experts.linear_fc1.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight8 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc2.weight10 + module.decoder.layers.7.mlp.experts.linear_fc2.weight13 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight15 + module.decoder.layers.7.mlp.experts.linear_fc2.weight11 + module.decoder.layers.7.mlp.experts.linear_fc2.weight9 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight8 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 + module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.7.mlp.experts.linear_fc2.weight2 + module.decoder.layers.7.mlp.experts.linear_fc2.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight15 + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + module.decoder.layers.7.mlp.experts.linear_fc2.weight3 + module.decoder.layers.7.mlp.experts.linear_fc2.weight1 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight13 + module.decoder.layers.7.mlp.experts.linear_fc1.weight11 + module.decoder.layers.7.mlp.experts.linear_fc1.weight9 + module.decoder.layers.7.mlp.experts.linear_fc1.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight10 + module.decoder.layers.7.mlp.experts.linear_fc1.weight12 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 + module.decoder.layers.7.mlp.experts.linear_fc1.weight8 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight14 + module.decoder.layers.6.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight15 + module.decoder.layers.6.mlp.experts.linear_fc2.weight13 + module.decoder.layers.6.mlp.experts.linear_fc2.weight11 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight10 + module.decoder.layers.6.mlp.experts.linear_fc2.weight8 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight9 + module.decoder.layers.6.mlp.experts.linear_fc2.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight5 + module.decoder.layers.6.mlp.experts.linear_fc2.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight1 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight11 + module.decoder.layers.6.mlp.experts.linear_fc1.weight13 + module.decoder.layers.6.mlp.experts.linear_fc1.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight12 + module.decoder.layers.6.mlp.experts.linear_fc1.weight10 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight9 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc1.weight8 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 42 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc2.weight13 +Params for bucket 43 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight12 + module.decoder.layers.5.mlp.experts.linear_fc2.weight10 + module.decoder.layers.5.mlp.experts.linear_fc2.weight8 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight9 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight13 + module.decoder.layers.5.mlp.experts.linear_fc1.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight14 + module.decoder.layers.5.mlp.experts.linear_fc1.weight12 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight9 + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight10 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc1.weight8 +Params for bucket 46 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 +Params for bucket 47 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight13 + module.decoder.layers.4.mlp.experts.linear_fc2.weight11 + module.decoder.layers.4.mlp.experts.linear_fc2.weight9 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight14 + module.decoder.layers.4.mlp.experts.linear_fc2.weight10 + module.decoder.layers.4.mlp.experts.linear_fc2.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight8 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight14 + module.decoder.layers.4.mlp.experts.linear_fc1.weight12 + module.decoder.layers.4.mlp.experts.linear_fc1.weight15 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight13 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight10 + module.decoder.layers.4.mlp.experts.linear_fc1.weight8 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight11 + module.decoder.layers.4.mlp.experts.linear_fc1.weight9 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +Params for bucket 51 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight12 + module.decoder.layers.3.mlp.experts.linear_fc2.weight10 + module.decoder.layers.3.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 + module.decoder.layers.3.mlp.experts.linear_fc2.weight15 + module.decoder.layers.3.mlp.experts.linear_fc2.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight11 + module.decoder.layers.3.mlp.experts.linear_fc2.weight9 + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 +Params for bucket 52 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight15 + module.decoder.layers.3.mlp.experts.linear_fc1.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight14 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight11 + module.decoder.layers.3.mlp.experts.linear_fc1.weight9 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 + module.decoder.layers.3.mlp.experts.linear_fc1.weight12 + module.decoder.layers.3.mlp.experts.linear_fc1.weight10 + module.decoder.layers.3.mlp.experts.linear_fc1.weight8 +Params for bucket 54 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight5 + module.decoder.layers.3.mlp.experts.linear_fc1.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 +Params for bucket 55 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight12 + module.decoder.layers.2.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight13 + module.decoder.layers.2.mlp.experts.linear_fc2.weight11 + module.decoder.layers.2.mlp.experts.linear_fc2.weight9 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 +Params for bucket 56 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 + module.decoder.layers.2.mlp.experts.linear_fc2.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight14 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight13 + module.decoder.layers.2.mlp.experts.linear_fc1.weight11 + module.decoder.layers.2.mlp.experts.linear_fc1.weight9 + module.decoder.layers.2.mlp.experts.linear_fc1.weight12 + module.decoder.layers.2.mlp.experts.linear_fc1.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight8 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 +Params for bucket 59 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight14 + module.decoder.layers.1.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc2.weight10 + module.decoder.layers.1.mlp.experts.linear_fc2.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight13 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc2.weight9 +Params for bucket 60 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight8 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight13 + module.decoder.layers.1.mlp.experts.linear_fc1.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight9 + module.decoder.layers.1.mlp.experts.linear_fc1.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight10 + module.decoder.layers.1.mlp.experts.linear_fc1.weight12 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + module.decoder.layers.1.mlp.experts.linear_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 + module.decoder.layers.1.mlp.experts.linear_fc1.weight8 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 +Params for bucket 63 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight15 + module.decoder.layers.0.mlp.experts.linear_fc2.weight13 + module.decoder.layers.0.mlp.experts.linear_fc2.weight11 +Params for bucket 64 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight10 + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight8 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight9 + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 +Params for bucket 65 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight12 + module.decoder.layers.0.mlp.experts.linear_fc1.weight14 + module.decoder.layers.0.mlp.experts.linear_fc1.weight10 + module.decoder.layers.0.mlp.experts.linear_fc1.weight15 + module.decoder.layers.0.mlp.experts.linear_fc1.weight13 + module.decoder.layers.0.mlp.experts.linear_fc1.weight11 +Params for bucket 66 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight8 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 + module.decoder.layers.0.mlp.experts.linear_fc1.weight9 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 +Params for bucket 67 (29360128 elements, 29360128 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + module.decoder.layers.0.mlp.experts.linear_fc1.weight2 + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight1 +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=, config_logger_dir='') +INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine +[after model, optimizer, and learning rate scheduler are built] datetime: 2026-08-30 07:36:21 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + validation: 0 + test: 0 +INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] +> building train, validation, and test datasets for GPT ... +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(192, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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) +INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths +INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers +INFO:megatron.core.datasets.indexed_dataset: Extract the document indices +INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029 +INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350 +INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices +INFO:megatron.core.datasets.gpt_dataset: Load the document index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-30 07:36:22 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (3902.27, 3910.78) + train/valid/test-data-iterators-setup ..........: (75.82, 90.03) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 07:36:22 +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") + [2026-08-30 07:36:48] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 26799.7 | throughput per GPU (TFLOP/s/GPU): 12.4 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102500E+01 | loss scale: 1.0 | grad norm: 44.121 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 133888 | min safe capacity: 1.0215 | +Number of parameters in transformer block in billions: 23.22 +Number of parameters in embedding layers in billions: 0.41 +Total number of parameters in billions: 23.63 +Number of parameters in most loaded shard in billions: 23.6320 +Theoretical memory footprints: weight and optimizer=169029.43 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 58109.861328125 | max allocated: 83199.2578125 | reserved: 85038.0 | max reserved: 85038.0 + [2026-08-30 07:36:49] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 669.0 | throughput per GPU (TFLOP/s/GPU): 498.3 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101850E+01 | loss scale: 1.0 | grad norm: 45.885 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:50] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 513.7 | throughput per GPU (TFLOP/s/GPU): 649.0 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.060117E+01 | loss scale: 1.0 | grad norm: 216.505 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 07:36:52] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2492.3 | throughput per GPU (TFLOP/s/GPU): 133.8 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.301611E+01 | loss scale: 1.0 | grad norm: 46.057 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:53] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 477.7 | throughput per GPU (TFLOP/s/GPU): 697.8 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.033481E+01 | loss scale: 1.0 | grad norm: 17.343 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:53] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 484.5 | throughput per GPU (TFLOP/s/GPU): 688.2 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.016366E+01 | loss scale: 1.0 | grad norm: 17.504 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:54] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 479.5 | throughput per GPU (TFLOP/s/GPU): 695.3 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.896972E+00 | loss scale: 1.0 | grad norm: 6.175 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:54] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 483.2 | throughput per GPU (TFLOP/s/GPU): 690.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.315825E+00 | loss scale: 1.0 | grad norm: 5.513 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:54] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 480.0 | throughput per GPU (TFLOP/s/GPU): 694.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.064344E+00 | loss scale: 1.0 | grad norm: 5.391 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:55] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 475.6 | throughput per GPU (TFLOP/s/GPU): 701.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.911621E+00 | loss scale: 1.0 | grad norm: 5.192 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:55] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 483.8 | throughput per GPU (TFLOP/s/GPU): 689.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.800047E+00 | loss scale: 1.0 | grad norm: 4.956 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | + [2026-08-30 07:36:56] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 488.6 | throughput per GPU (TFLOP/s/GPU): 682.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.683390E+00 | loss scale: 1.0 | grad norm: 4.922 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 137984 | min safe capacity: 1.0527 | +[after training is done] datetime: 2026-08-30 07:36:56 +[rank0]:[W830 07:36:57.787196433 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()) +[rank5]:[W830 07:36:58.997897598 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()) +[rank2]:[W830 07:36:58.998346933 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()) +[rank1]:[W830 07:36:58.998532348 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()) +[rank3]:[W830 07:36:58.000839015 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()) +[rank7]:[W830 07:36:59.235395638 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()) +[rank4]:[W830 07:36:59.365629886 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()) +[rank6]:[W830 07:36:59.386431558 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()) +W0830 07:37:03.351000 781569 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0830 07:37:03.352000 781569 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 781657 closing signal SIGTERM +W0830 07:37:03.352000 781569 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 781659 closing signal SIGTERM +W0830 07:37:03.352000 781569 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 781660 closing signal SIGTERM +W0830 07:37:03.352000 781569 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 781661 closing signal SIGTERM +W0830 07:37:03.353000 781569 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 781663 closing signal SIGTERM +W0830 07:37:03.354000 781569 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 781664 closing signal SIGTERM +Traceback (most recent call last): + File "/usr/local/bin/torchrun", line 8, in + sys.exit(main()) + ^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper + return f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main + run(args) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run + elastic_launch( + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ + return launch_agent(self._config, self._entrypoint, list(args)) + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent + result = agent.run() + ^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper + result = f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run + result = self._invoke_run(role) + ^^^^^^^^^^^^^^^^^^^^^^ + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run + time.sleep(monitor_interval) + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler + raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) +torch.distributed.elastic.multiprocessing.api.SignalException: Process 781569 got signal: 15 diff --git a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/results.tsv b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/results.tsv index 962062c0e4c7137680217979f8c004fead783c97..1c2346f896b83082bdbce8e5a01bfbe74ee07db2 100644 --- a/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/results.tsv +++ b/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/results.tsv @@ -5,3 +5,9 @@ G1 ultraep matched 1 4096 16 2.40625 none 3 OK 891.92 874.90 129475 /workspace/t G2 ultraep matched 4 4096 16 1.08203125 none 1 OK 492.27 465.20 112315 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G2__ultraep__matched__seq4096__gbs16__cap1p08203125/rep1 G2 ultraep matched 4 4096 16 1.08203125 none 2 OK 492.21 462.50 112315 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G2__ultraep__matched__seq4096__gbs16__cap1p08203125/rep2 G2 ultraep matched 4 4096 16 1.08203125 none 3 OK 492.54 462.50 112315 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G2__ultraep__matched__seq4096__gbs16__cap1p08203125/rep3 +G3 ultraep matched 8 4096 16 1.072265625 none 1 OK 460.01 449.60 113951 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep1 +G3 ultraep matched 8 4096 16 1.072265625 none 2 OK 459.50 449.60 113951 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep2 +G3 ultraep matched 8 4096 16 1.072265625 none 3 OK 459.60 449.60 113951 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G3__ultraep__matched__seq4096__gbs16__cap1p072265625/rep3 +G4 ultraep matched 16 4096 16 1.095703125 none 1 OK 483.19 475.20 132437 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep1 +G4 ultraep matched 16 4096 16 1.095703125 none 2 OK 482.88 474.50 132437 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep2 +G4 ultraep matched 16 4096 16 1.095703125 none 3 OK 482.64 474.50 132437 /workspace/tiera-rerun-fixed-moonep/ultraep/a2/a2_matched_gbs16/G4__ultraep__matched__seq4096__gbs16__cap1p095703125/rep3