diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/mem_samples.csv b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/mem_samples.csv index 30082f984e2544acf16c54c683e0bcba1737d224..80dfe841c2a238f6f3337edd3bed403a09494c3e 100644 --- a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/mem_samples.csv +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/mem_samples.csv @@ -6,3 +6,251 @@ 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, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 821 +7, 821 +0, 1955 +1, 1955 +2, 1955 +3, 1955 +4, 1955 +5, 1955 +6, 1955 +7, 1955 +0, 5821 +1, 6027 +2, 6039 +3, 5857 +4, 5881 +5, 6063 +6, 6015 +7, 6297 +0, 26775 +1, 26811 +2, 26671 +3, 26617 +4, 26579 +5, 26709 +6, 26745 +7, 26767 +0, 36451 +1, 36533 +2, 36411 +3, 36361 +4, 36371 +5, 36395 +6, 36503 +7, 36525 +0, 37785 +1, 37869 +2, 37719 +3, 37717 +4, 37717 +5, 37715 +6, 37851 +7, 37869 +0, 37873 +1, 37957 +2, 37807 +3, 37805 +4, 37805 +5, 37803 +6, 37939 +7, 37957 +0, 37875 +1, 37959 +2, 37809 +3, 37807 +4, 37807 +5, 37805 +6, 37941 +7, 37959 +0, 38821 +1, 38905 +2, 38755 +3, 38755 +4, 38755 +5, 38751 +6, 38887 +7, 38905 +0, 38821 +1, 38905 +2, 38755 +3, 38755 +4, 38755 +5, 38751 +6, 38887 +7, 38905 +0, 38827 +1, 38911 +2, 38761 +3, 38761 +4, 38761 +5, 38757 +6, 38893 +7, 38911 +0, 38829 +1, 38913 +2, 38763 +3, 38763 +4, 38763 +5, 38759 +6, 38895 +7, 38913 +0, 111963 +1, 112059 +2, 111927 +3, 111917 +4, 111879 +5, 111891 +6, 112031 +7, 112045 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 113127 +1, 113213 +2, 113061 +3, 113061 +4, 113061 +5, 113059 +6, 113193 +7, 113213 +0, 116353 +1, 116441 +2, 116287 +3, 116287 +4, 116287 +5, 116285 +6, 116419 +7, 116441 +0, 116957 +1, 117085 +2, 116911 +3, 116919 +4, 116943 +5, 116911 +6, 117073 +7, 117085 +0, 94701 +1, 88371 +2, 92559 +3, 92553 +4, 94697 +5, 92315 +6, 98971 +7, 92657 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 150347 +1, 2058 +2, 1034 +3, 150253 +4, 2314 +5, 150205 +6, 2314 +7, 150357 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/train.log b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/train.log index 1cd3359c97065bb47a27a4c1be59214bb8d3f23c..a70f9477e16d7e70ebc80d34157f6e3db64d549c 100644 --- a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/train.log +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1/train.log @@ -10,3 +10,1563 @@ W0826 18:19:37.074000 488547 torch/distributed/run.py:766] W0826 18:19:37.074000 488547 torch/distributed/run.py:766] ***************************************** W0826 18:19:37.074000 488547 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. W0826 18:19:37.074000 488547 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 ............................... False + 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 .............................. 24 + encoder_seq_length .............................. 8192 + 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 ................................. 10944 + 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 ............................... 8 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... False + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 2048 + 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 .................................... 512 + 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 .............................. 0.001 + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 1408 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ 64 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.0755208333333333 + 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 .................. seq_aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... True + moe_router_score_function ....................... softmax + moe_router_topk ................................. 6 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. 2816 + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + 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 .......................................... 0.707 + mscale_all_dim .................................. 0.707 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... True + 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-06 + normalization ................................... RMSNorm + num_attention_heads ............................. 16 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 24 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 1 + 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 .................................... True + 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 ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 40.0 + 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 ...................................... 8192 + 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 1 +> 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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 4 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 [Gloo] Rank 32 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77 + +[Gloo] Rank [Gloo] Rank 7 is connected to 76 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks. +Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 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 7 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 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 [Gloo] Rank 2 is connected to 37 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is : +7 +[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 +[rank7]:[W826 18:19:55.157225444 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. +[rank1]:[W826 18:19:55.160649081 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]:[W826 18:19:55.160653478 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]:[W826 18:19:55.160741368 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. +[rank6]:[W826 18:19:55.160757852 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. +[rank2]:[W826 18:19:55.162409651 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. +[rank5]:[W826 18:19:55.168898840 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-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.035 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]:[W826 18:19:55.312818415 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: 3.618 seconds +time to initialize megatron (seconds): 6.913 +[after megatron is initialized] datetime: 2026-08-26 18:20:00 +building GPT model ... +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + 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( +/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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 2596386816 +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: 20 +Params for bucket 1 (103022592 elements, 103022592 padded size): + module.output_layer.weight +Params for bucket 2 (48503296 elements, 48503296 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.23.pre_mlp_layernorm.weight + module.decoder.layers.23.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.23.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.23.self_attention.linear_kv_down_proj.weight + module.decoder.layers.23.self_attention.linear_q_proj.weight + module.decoder.layers.22.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.23.mlp.router.weight + module.decoder.layers.23.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.23.self_attention.linear_kv_up_proj.weight + module.decoder.layers.23.self_attention.linear_proj.weight + module.decoder.layers.23.input_layernorm.weight + module.decoder.layers.22.mlp.shared_experts.linear_fc1.weight +Params for bucket 3 (40901632 elements, 40901632 padded size): + module.decoder.layers.22.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.22.self_attention.linear_kv_down_proj.weight + module.decoder.layers.22.pre_mlp_layernorm.weight + module.decoder.layers.22.self_attention.linear_q_proj.weight + module.decoder.layers.21.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.21.mlp.router.weight + module.decoder.layers.21.self_attention.linear_kv_up_proj.weight + module.decoder.layers.21.self_attention.linear_q_proj.weight + module.decoder.layers.22.mlp.router.weight + module.decoder.layers.22.self_attention.linear_kv_up_proj.weight + module.decoder.layers.22.self_attention.linear_proj.weight + module.decoder.layers.22.input_layernorm.weight + module.decoder.layers.21.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.21.pre_mlp_layernorm.weight + module.decoder.layers.21.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.21.self_attention.linear_kv_down_proj.weight +Params for bucket 4 (41163264 elements, 41163264 padded size): + module.decoder.layers.20.mlp.router.weight + module.decoder.layers.20.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.20.self_attention.linear_kv_up_proj.weight + module.decoder.layers.20.self_attention.linear_proj.weight + module.decoder.layers.20.input_layernorm.weight + module.decoder.layers.21.self_attention.linear_proj.weight + module.decoder.layers.20.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.21.input_layernorm.weight + module.decoder.layers.20.pre_mlp_layernorm.weight + module.decoder.layers.20.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.20.self_attention.linear_kv_down_proj.weight + module.decoder.layers.20.self_attention.linear_q_proj.weight + module.decoder.layers.19.mlp.shared_experts.linear_fc2.weight +Params for bucket 5 (42734080 elements, 42734080 padded size): + module.decoder.layers.19.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.19.mlp.router.weight + module.decoder.layers.19.self_attention.linear_kv_up_proj.weight + module.decoder.layers.19.self_attention.linear_proj.weight + module.decoder.layers.19.input_layernorm.weight + module.decoder.layers.18.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.19.pre_mlp_layernorm.weight + module.decoder.layers.19.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.19.self_attention.linear_kv_down_proj.weight + module.decoder.layers.19.self_attention.linear_q_proj.weight + module.decoder.layers.18.mlp.shared_experts.linear_fc2.weight +Params for bucket 6 (40901632 elements, 40901632 padded size): + module.decoder.layers.18.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.18.self_attention.linear_kv_down_proj.weight + module.decoder.layers.18.pre_mlp_layernorm.weight + module.decoder.layers.18.self_attention.linear_proj.weight + module.decoder.layers.18.input_layernorm.weight + module.decoder.layers.17.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.17.pre_mlp_layernorm.weight + module.decoder.layers.17.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.17.self_attention.linear_kv_down_proj.weight + module.decoder.layers.18.mlp.router.weight + module.decoder.layers.18.self_attention.linear_kv_up_proj.weight + module.decoder.layers.18.self_attention.linear_q_proj.weight + module.decoder.layers.17.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.17.mlp.router.weight + module.decoder.layers.17.self_attention.linear_kv_up_proj.weight + module.decoder.layers.17.self_attention.linear_q_proj.weight +Params for bucket 7 (41163264 elements, 41163264 padded size): + module.decoder.layers.17.self_attention.linear_proj.weight + module.decoder.layers.17.input_layernorm.weight + module.decoder.layers.16.mlp.router.weight + module.decoder.layers.16.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.16.pre_mlp_layernorm.weight + module.decoder.layers.16.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.16.self_attention.linear_kv_down_proj.weight + module.decoder.layers.16.self_attention.linear_q_proj.weight + module.decoder.layers.15.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.16.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.16.self_attention.linear_kv_up_proj.weight + module.decoder.layers.16.self_attention.linear_proj.weight + module.decoder.layers.16.input_layernorm.weight +Params for bucket 8 (42734080 elements, 42734080 padded size): + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.15.self_attention.linear_kv_down_proj.weight + module.decoder.layers.15.self_attention.linear_q_proj.weight + module.decoder.layers.15.self_attention.linear_proj.weight + module.decoder.layers.15.input_layernorm.weight + module.decoder.layers.14.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.15.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.15.self_attention.linear_kv_up_proj.weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.14.mlp.shared_experts.linear_fc2.weight +Params for bucket 9 (40901632 elements, 40901632 padded size): + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_kv_up_proj.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.13.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.pre_mlp_layernorm.weight + module.decoder.layers.13.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.13.self_attention.linear_kv_down_proj.weight + module.decoder.layers.13.self_attention.linear_q_proj.weight + module.decoder.layers.14.self_attention.linear_kv_down_proj.weight + module.decoder.layers.14.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_q_proj.weight + module.decoder.layers.14.input_layernorm.weight + module.decoder.layers.13.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.13.self_attention.linear_kv_up_proj.weight +Params for bucket 10 (41163264 elements, 41163264 padded size): + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.input_layernorm.weight + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.12.self_attention.linear_kv_down_proj.weight + module.decoder.layers.12.input_layernorm.weight + module.decoder.layers.12.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.12.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_kv_up_proj.weight + module.decoder.layers.12.self_attention.linear_q_proj.weight + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.11.mlp.shared_experts.linear_fc2.weight +Params for bucket 11 (42734080 elements, 42734080 padded size): + module.decoder.layers.11.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.11.self_attention.linear_kv_up_proj.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.input_layernorm.weight + module.decoder.layers.10.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_kv_down_proj.weight + module.decoder.layers.11.self_attention.linear_q_proj.weight + module.decoder.layers.10.mlp.shared_experts.linear_fc2.weight +Params for bucket 12 (40901632 elements, 40901632 padded size): + module.decoder.layers.10.self_attention.linear_kv_down_proj.weight + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.10.self_attention.linear_q_proj.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_kv_up_proj.weight + module.decoder.layers.9.self_attention.linear_q_proj.weight + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_kv_up_proj.weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.input_layernorm.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.9.self_attention.linear_kv_down_proj.weight +Params for bucket 13 (41163264 elements, 41163264 padded size): + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.8.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.8.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.9.input_layernorm.weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.8.self_attention.linear_kv_down_proj.weight + module.decoder.layers.8.input_layernorm.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_kv_up_proj.weight + module.decoder.layers.8.self_attention.linear_q_proj.weight + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc2.weight +Params for bucket 14 (42734080 elements, 42734080 padded size): + module.decoder.layers.7.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_kv_up_proj.weight + module.decoder.layers.7.self_attention.linear_q_proj.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.7.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_kv_down_proj.weight + module.decoder.layers.7.input_layernorm.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc1.weight +Params for bucket 15 (40901632 elements, 40901632 padded size): + module.decoder.layers.6.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_kv_down_proj.weight + module.decoder.layers.6.self_attention.linear_q_proj.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.input_layernorm.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.5.self_attention.linear_kv_down_proj.weight + module.decoder.layers.5.self_attention.linear_q_proj.weight + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_kv_up_proj.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_kv_up_proj.weight +Params for bucket 16 (41163264 elements, 41163264 padded size): + module.decoder.layers.4.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.4.self_attention.linear_kv_down_proj.weight + module.decoder.layers.4.input_layernorm.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.input_layernorm.weight + module.decoder.layers.4.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.4.self_attention.linear_kv_up_proj.weight + module.decoder.layers.4.self_attention.linear_q_proj.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.3.mlp.shared_experts.linear_fc2.weight +Params for bucket 17 (42734080 elements, 42734080 padded size): + module.decoder.layers.3.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.3.self_attention.linear_kv_down_proj.weight + module.decoder.layers.3.self_attention.linear_q_proj.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.input_layernorm.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_kv_up_proj.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc2.weight +Params for bucket 18 (40901632 elements, 40901632 padded size): + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_kv_up_proj.weight + module.decoder.layers.2.input_layernorm.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.1.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.1.self_attention.linear_kv_down_proj.weight + module.decoder.layers.1.self_attention.linear_q_proj.weight + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_kv_down_proj.weight + module.decoder.layers.2.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.2.self_attention.linear_q_proj.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.self_attention.linear_kv_up_proj.weight +Params for bucket 19 (71436288 elements, 71436288 padded size): + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.1.input_layernorm.weight +Params for bucket 20 (116789760 elements, 116789760 padded size): + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_q_proj.weight + module.decoder.layers.0.self_attention.linear_kv_up_proj.weight + module.decoder.layers.0.self_attention.linear_kv_down_proj.weight + module.decoder.layers.0.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.input_layernorm.weight + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 40 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.23.mlp.experts.linear_fc2.weight6 + module.decoder.layers.23.mlp.experts.linear_fc2.weight4 + module.decoder.layers.23.mlp.experts.linear_fc2.weight2 + module.decoder.layers.23.mlp.experts.linear_fc2.weight0 + module.decoder.layers.23.mlp.experts.linear_fc1.weight7 + module.decoder.layers.23.mlp.experts.linear_fc1.weight5 + module.decoder.layers.23.mlp.experts.linear_fc2.weight7 + module.decoder.layers.23.mlp.experts.linear_fc2.weight5 + module.decoder.layers.23.mlp.experts.linear_fc2.weight3 + module.decoder.layers.23.mlp.experts.linear_fc2.weight1 + module.decoder.layers.23.mlp.experts.linear_fc1.weight6 +Params for bucket 2 (40370176 elements, 40370176 padded size): + module.decoder.layers.23.mlp.experts.linear_fc1.weight1 + module.decoder.layers.23.mlp.experts.linear_fc1.weight3 + module.decoder.layers.22.mlp.experts.linear_fc2.weight6 + module.decoder.layers.22.mlp.experts.linear_fc2.weight4 + module.decoder.layers.23.mlp.experts.linear_fc1.weight4 + module.decoder.layers.23.mlp.experts.linear_fc1.weight0 + module.decoder.layers.23.mlp.experts.linear_fc1.weight2 + module.decoder.layers.22.mlp.experts.linear_fc2.weight7 + module.decoder.layers.22.mlp.experts.linear_fc2.weight5 +Params for bucket 3 (40370176 elements, 40370176 padded size): + module.decoder.layers.22.mlp.experts.linear_fc2.weight2 + module.decoder.layers.22.mlp.experts.linear_fc1.weight7 + module.decoder.layers.22.mlp.experts.linear_fc2.weight0 + module.decoder.layers.22.mlp.experts.linear_fc1.weight5 + module.decoder.layers.22.mlp.experts.linear_fc1.weight3 + module.decoder.layers.22.mlp.experts.linear_fc2.weight3 + module.decoder.layers.22.mlp.experts.linear_fc2.weight1 + module.decoder.layers.22.mlp.experts.linear_fc1.weight6 + module.decoder.layers.22.mlp.experts.linear_fc1.weight4 +Params for bucket 4 (40370176 elements, 40370176 padded size): + module.decoder.layers.21.mlp.experts.linear_fc2.weight6 + module.decoder.layers.22.mlp.experts.linear_fc1.weight1 + module.decoder.layers.21.mlp.experts.linear_fc2.weight4 + module.decoder.layers.21.mlp.experts.linear_fc2.weight2 + module.decoder.layers.21.mlp.experts.linear_fc2.weight0 + module.decoder.layers.22.mlp.experts.linear_fc1.weight2 + module.decoder.layers.22.mlp.experts.linear_fc1.weight0 + module.decoder.layers.21.mlp.experts.linear_fc2.weight7 + module.decoder.layers.21.mlp.experts.linear_fc2.weight5 + module.decoder.layers.21.mlp.experts.linear_fc2.weight3 + module.decoder.layers.21.mlp.experts.linear_fc2.weight1 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.21.mlp.experts.linear_fc1.weight7 + module.decoder.layers.21.mlp.experts.linear_fc1.weight5 + module.decoder.layers.21.mlp.experts.linear_fc1.weight3 + module.decoder.layers.21.mlp.experts.linear_fc1.weight1 + module.decoder.layers.21.mlp.experts.linear_fc1.weight4 + module.decoder.layers.21.mlp.experts.linear_fc1.weight6 + module.decoder.layers.21.mlp.experts.linear_fc1.weight2 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.20.mlp.experts.linear_fc2.weight4 + module.decoder.layers.20.mlp.experts.linear_fc2.weight6 + module.decoder.layers.20.mlp.experts.linear_fc2.weight2 + module.decoder.layers.20.mlp.experts.linear_fc2.weight0 + module.decoder.layers.20.mlp.experts.linear_fc1.weight7 + module.decoder.layers.21.mlp.experts.linear_fc1.weight0 + module.decoder.layers.20.mlp.experts.linear_fc2.weight7 + module.decoder.layers.20.mlp.experts.linear_fc2.weight5 + module.decoder.layers.20.mlp.experts.linear_fc2.weight3 + module.decoder.layers.20.mlp.experts.linear_fc2.weight1 + module.decoder.layers.20.mlp.experts.linear_fc1.weight6 +Params for bucket 7 (40370176 elements, 40370176 padded size): + module.decoder.layers.20.mlp.experts.linear_fc1.weight5 + module.decoder.layers.20.mlp.experts.linear_fc1.weight3 + module.decoder.layers.20.mlp.experts.linear_fc1.weight1 + module.decoder.layers.19.mlp.experts.linear_fc2.weight7 + module.decoder.layers.20.mlp.experts.linear_fc1.weight4 + module.decoder.layers.20.mlp.experts.linear_fc1.weight2 + module.decoder.layers.20.mlp.experts.linear_fc1.weight0 + module.decoder.layers.19.mlp.experts.linear_fc2.weight6 +Params for bucket 8 (40370176 elements, 40370176 padded size): + module.decoder.layers.19.mlp.experts.linear_fc2.weight5 + module.decoder.layers.19.mlp.experts.linear_fc2.weight1 + module.decoder.layers.19.mlp.experts.linear_fc2.weight3 + module.decoder.layers.19.mlp.experts.linear_fc1.weight6 + module.decoder.layers.19.mlp.experts.linear_fc1.weight4 + module.decoder.layers.19.mlp.experts.linear_fc2.weight2 + module.decoder.layers.19.mlp.experts.linear_fc2.weight4 + module.decoder.layers.19.mlp.experts.linear_fc2.weight0 + module.decoder.layers.19.mlp.experts.linear_fc1.weight7 + module.decoder.layers.19.mlp.experts.linear_fc1.weight5 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.18.mlp.experts.linear_fc2.weight7 + module.decoder.layers.19.mlp.experts.linear_fc1.weight2 + module.decoder.layers.19.mlp.experts.linear_fc1.weight0 + module.decoder.layers.18.mlp.experts.linear_fc2.weight5 + module.decoder.layers.18.mlp.experts.linear_fc2.weight3 + module.decoder.layers.19.mlp.experts.linear_fc1.weight3 + module.decoder.layers.19.mlp.experts.linear_fc1.weight1 + module.decoder.layers.18.mlp.experts.linear_fc2.weight6 + module.decoder.layers.18.mlp.experts.linear_fc2.weight4 + module.decoder.layers.18.mlp.experts.linear_fc2.weight2 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.18.mlp.experts.linear_fc2.weight1 + module.decoder.layers.18.mlp.experts.linear_fc1.weight6 + module.decoder.layers.18.mlp.experts.linear_fc1.weight4 + module.decoder.layers.18.mlp.experts.linear_fc1.weight2 + module.decoder.layers.18.mlp.experts.linear_fc1.weight7 + module.decoder.layers.18.mlp.experts.linear_fc1.weight5 + module.decoder.layers.18.mlp.experts.linear_fc2.weight0 + module.decoder.layers.18.mlp.experts.linear_fc1.weight3 +Params for bucket 11 (40370176 elements, 40370176 padded size): + module.decoder.layers.18.mlp.experts.linear_fc1.weight0 + module.decoder.layers.17.mlp.experts.linear_fc2.weight7 + module.decoder.layers.17.mlp.experts.linear_fc2.weight5 + module.decoder.layers.17.mlp.experts.linear_fc2.weight3 + module.decoder.layers.17.mlp.experts.linear_fc2.weight1 + module.decoder.layers.18.mlp.experts.linear_fc1.weight1 + module.decoder.layers.17.mlp.experts.linear_fc2.weight6 + module.decoder.layers.17.mlp.experts.linear_fc2.weight4 + module.decoder.layers.17.mlp.experts.linear_fc2.weight2 + module.decoder.layers.17.mlp.experts.linear_fc2.weight0 + module.decoder.layers.17.mlp.experts.linear_fc1.weight7 +Params for bucket 12 (40370176 elements, 40370176 padded size): + module.decoder.layers.17.mlp.experts.linear_fc1.weight6 + module.decoder.layers.17.mlp.experts.linear_fc1.weight4 + module.decoder.layers.17.mlp.experts.linear_fc1.weight2 + module.decoder.layers.17.mlp.experts.linear_fc1.weight0 + module.decoder.layers.17.mlp.experts.linear_fc1.weight3 + module.decoder.layers.17.mlp.experts.linear_fc1.weight5 + module.decoder.layers.17.mlp.experts.linear_fc1.weight1 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.16.mlp.experts.linear_fc2.weight7 + module.decoder.layers.16.mlp.experts.linear_fc2.weight3 + module.decoder.layers.16.mlp.experts.linear_fc2.weight5 + module.decoder.layers.16.mlp.experts.linear_fc2.weight1 + module.decoder.layers.16.mlp.experts.linear_fc1.weight6 + module.decoder.layers.16.mlp.experts.linear_fc2.weight6 + module.decoder.layers.16.mlp.experts.linear_fc2.weight4 + module.decoder.layers.16.mlp.experts.linear_fc2.weight2 + module.decoder.layers.16.mlp.experts.linear_fc2.weight0 + module.decoder.layers.16.mlp.experts.linear_fc1.weight7 + module.decoder.layers.16.mlp.experts.linear_fc1.weight5 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.16.mlp.experts.linear_fc1.weight4 + module.decoder.layers.16.mlp.experts.linear_fc1.weight0 + module.decoder.layers.16.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.16.mlp.experts.linear_fc1.weight3 + module.decoder.layers.16.mlp.experts.linear_fc1.weight1 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 15 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + 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_fc2.weight3 + 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_fc1.weight4 +Params for bucket 16 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 + 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.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + 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 +Params for bucket 17 (40370176 elements, 40370176 padded size): + 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 + 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 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + 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.weight7 + 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.weight6 +Params for bucket 19 (40370176 elements, 40370176 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.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 +Params for bucket 20 (40370176 elements, 40370176 padded size): + 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_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 + 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_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + 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.weight5 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 + 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.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 +Params for bucket 22 (40370176 elements, 40370176 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_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 +Params for bucket 23 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + 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.weight2 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 + 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.weight1 +Params for bucket 24 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 + 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.weight6 + 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 25 (40370176 elements, 40370176 padded size): + 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 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + 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.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + 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.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 +Params for bucket 27 (40370176 elements, 40370176 padded size): + 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.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight2 + 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.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 +Params for bucket 28 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + 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 + module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + 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 +Params for bucket 29 (40370176 elements, 40370176 padded size): + 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.weight2 + 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.weight1 +Params for bucket 30 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + 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 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + 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_fc1.weight6 +Params for bucket 31 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 + 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.weight7 +Params for bucket 32 (40370176 elements, 40370176 padded size): + 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.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + 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.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 +Params for bucket 33 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + 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 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + 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_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + 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.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 + 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 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +Params for bucket 36 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + 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 37 (40370176 elements, 40370176 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.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + 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 + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 +Params for bucket 39 (40370176 elements, 40370176 padded size): + 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.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_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 +Params for bucket 40 (17301504 elements, 17301504 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 +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: 2026-08-26 18:20:01 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 96 + 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=(96, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=8192, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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 eadc3756447e39d609e8aa728a91f451-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from eadc3756447e39d609e8aa728a91f451-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from eadc3756447e39d609e8aa728a91f451-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 14536 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-26 18:20:01 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (674.83, 689.99) + train/valid/test-data-iterators-setup ..........: (62.91, 71.76) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-26 18:20:01 +/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: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: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: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: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: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: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: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.") + [2026-08-26 18:20:41] iteration 1/ 12 | consumed samples: 8 | elapsed time per iteration (ms): 39434.7 | throughput per GPU (TFLOP/s/GPU): 3.2 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.092139E+01 | loss scale: 1.0 | grad norm: 30.620 | number of skipped iterations: 0 | number of nan iterations: 0 |Number of parameters in transformer block in billions: 13.53 + +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 13.74 +Number of parameters in most loaded shard in billions: 13.7356 +Theoretical memory footprints: weight and optimizer=98244.96 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 40509.6416015625 | max allocated: 106239.89453125 | reserved: 109990.0 | max reserved: 109990.0 +/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.") + [2026-08-26 18:20:41] iteration 2/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 633.9 | throughput per GPU (TFLOP/s/GPU): 201.2 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.092003E+01 | loss scale: 1.0 | grad norm: 30.566 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:42] iteration 3/ 12 | consumed samples: 24 | elapsed time per iteration (ms): 636.6 | throughput per GPU (TFLOP/s/GPU): 200.4 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 9.952471E+00 | loss scale: 1.0 | grad norm: 42.237 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-26 18:20:44] iteration 4/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 2037.8 | throughput per GPU (TFLOP/s/GPU): 62.6 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 9.476898E+00 | loss scale: 1.0 | grad norm: 7.154 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:45] iteration 5/ 12 | consumed samples: 40 | elapsed time per iteration (ms): 583.7 | throughput per GPU (TFLOP/s/GPU): 218.6 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 1.114793E+01 | loss scale: 1.0 | grad norm: 400.648 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:45] iteration 6/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 589.9 | throughput per GPU (TFLOP/s/GPU): 216.3 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 9.368878E+00 | loss scale: 1.0 | grad norm: 5.139 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:46] iteration 7/ 12 | consumed samples: 56 | elapsed time per iteration (ms): 588.2 | throughput per GPU (TFLOP/s/GPU): 216.9 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 8.975649E+00 | loss scale: 1.0 | grad norm: 3.681 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:46] iteration 8/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 586.3 | throughput per GPU (TFLOP/s/GPU): 217.6 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.785368E+00 | loss scale: 1.0 | grad norm: 5.697 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:47] iteration 9/ 12 | consumed samples: 72 | elapsed time per iteration (ms): 584.7 | throughput per GPU (TFLOP/s/GPU): 218.2 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.613759E+00 | loss scale: 1.0 | grad norm: 3.002 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:47] iteration 10/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 589.2 | throughput per GPU (TFLOP/s/GPU): 216.5 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.899922E+00 | loss scale: 1.0 | grad norm: 1155.511 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:48] iteration 11/ 12 | consumed samples: 88 | elapsed time per iteration (ms): 586.0 | throughput per GPU (TFLOP/s/GPU): 217.7 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.563369E+00 | loss scale: 1.0 | grad norm: 14.007 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:20:49] iteration 12/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 588.1 | throughput per GPU (TFLOP/s/GPU): 217.0 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.509861E+00 | loss scale: 1.0 | grad norm: 83.626 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-26 18:20:49 +[rank0]:[W826 18:20:50.482862524 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()) diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/mem_samples.csv b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..a7958aa9266fd3dfad2a8d7ef0b548f4722ff677 --- /dev/null +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/mem_samples.csv @@ -0,0 +1,256 @@ +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, 2 +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, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 819 +7, 821 +0, 1537 +1, 1537 +2, 1537 +3, 1537 +4, 1537 +5, 1537 +6, 1537 +7, 1537 +0, 2471 +1, 2471 +2, 2471 +3, 2471 +4, 2471 +5, 2471 +6, 2471 +7, 2471 +0, 26621 +1, 26701 +2, 26499 +3, 26491 +4, 26523 +5, 26525 +6, 26677 +7, 26719 +0, 36435 +1, 36509 +2, 36387 +3, 36395 +4, 36381 +5, 36397 +6, 36493 +7, 36519 +0, 37785 +1, 37869 +2, 37719 +3, 37717 +4, 37717 +5, 37715 +6, 37851 +7, 37869 +0, 37873 +1, 37957 +2, 37807 +3, 37805 +4, 37805 +5, 37803 +6, 37939 +7, 37957 +0, 38117 +1, 38191 +2, 38017 +3, 38063 +4, 38021 +5, 38033 +6, 38175 +7, 38161 +0, 38821 +1, 38905 +2, 38755 +3, 38755 +4, 38755 +5, 38751 +6, 38887 +7, 38905 +0, 38821 +1, 38905 +2, 38755 +3, 38755 +4, 38755 +5, 38751 +6, 38887 +7, 38905 +0, 38829 +1, 38913 +2, 38763 +3, 38763 +4, 38763 +5, 38759 +6, 38895 +7, 38913 +0, 38829 +1, 38913 +2, 38763 +3, 38763 +4, 38763 +5, 38759 +6, 38895 +7, 38913 +0, 111991 +1, 112065 +2, 111957 +3, 111973 +4, 111917 +5, 111911 +6, 112059 +7, 112087 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 115705 +1, 115791 +2, 115639 +3, 115639 +4, 115639 +5, 115637 +6, 115771 +7, 115791 +0, 116353 +1, 116441 +2, 116287 +3, 116287 +4, 116287 +5, 116285 +6, 116419 +7, 116439 +0, 117041 +1, 117121 +2, 116937 +3, 116959 +4, 116925 +5, 116945 +6, 117093 +7, 117121 +0, 99081 +1, 88359 +2, 92559 +3, 94705 +4, 94607 +5, 92499 +6, 103447 +7, 92645 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 2314 +1, 150357 +2, 2314 +3, 150253 +4, 150253 +5, 2058 +6, 1034 +7, 150357 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/run.env b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/run.env new file mode 100644 index 0000000000000000000000000000000000000000..186fc15cda04c26f2455c01b15c7ee8b67526e58 --- /dev/null +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/run.env @@ -0,0 +1,61 @@ +backend=echo +config=DSV2_LITE_L24 +family=tierB +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=24 +hidden=2048 +ffn_hidden=10944 +moe_ffn_hidden=1408 +num_experts=64 +topk=6 +experts_per_rank=8 +expert_params=8650752 +heads=16 +kv_groups=0 +kv_channels=128 +qk_layernorm=1 +shared_ffn=2816 +shared_expert_overlap=0 +mla=1 +dense_first_layers=1 +moe_layers=23 +rotary_base=10000 +rotary_scaling=40 +qkv_bias=0 +router_balancing=seq_aux_loss +aux_loss_coeff=1e-3 +router_pre_softmax=1 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-6 +seq_length=8192 +micro_batch_size=1 +global_batch_size=8 +num_microbatches=1 +assignments_per_rank=49152 +mean_tokens_per_expert=6144 +slot_policy=matched +slots_per_rank=8 +capacity_mode=static +capacity=1.0755208333333333 +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= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_recipe=mxfp8 diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/train.log b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/train.log new file mode 100644 index 0000000000000000000000000000000000000000..afabeb8eb0354ec4d0f4e99b072ef90afd4dafa9 --- /dev/null +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/train.log @@ -0,0 +1,1572 @@ +============================================================== + DSV2_LITE_L24 / echo (slots/rank=8 policy=matched) + E=64 K=6 H'=1408 m=8 + L=24 seq=8192 gbs=8 (1 microbatches) + cap_mode=static cap=1.0755208333333333 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0826 18:20:57.271000 492233 torch/distributed/run.py:766] +W0826 18:20:57.271000 492233 torch/distributed/run.py:766] ***************************************** +W0826 18:20:57.271000 492233 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. +W0826 18:20:57.271000 492233 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 ............................... False + 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 .............................. 24 + encoder_seq_length .............................. 8192 + 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 ................................. 10944 + 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 ............................... 8 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... False + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 2048 + 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 .................................... 512 + 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 .............................. 0.001 + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 1408 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ 64 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.0755208333333333 + 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 .................. seq_aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... True + moe_router_score_function ....................... softmax + moe_router_topk ................................. 6 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. 2816 + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + 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 .......................................... 0.707 + mscale_all_dim .................................. 0.707 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... True + 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-06 + normalization ................................... RMSNorm + num_attention_heads ............................. 16 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 24 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 1 + 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 .................................... True + 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 ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 40.0 + 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 ...................................... 8192 + 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 1 +> 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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 60 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77 + +[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 [Gloo] Rank 5 is connected to 77 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected 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 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 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 [Gloo] Rank 6 is connected to 57 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected 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 [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 +[rank4]:[W826 18:21:15.763846225 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. +[rank7]:[W826 18:21:15.765547635 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. +[rank2]:[W826 18:21:15.765678408 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]:[W826 18:21:15.765685392 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. +[rank3]:[W826 18:21:15.765722118 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]:[W826 18:21:15.765725161 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]:[W826 18:21:15.765760005 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-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.034 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]:[W826 18:21:15.928385033 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: 3.777 seconds +time to initialize megatron (seconds): 7.085 +[after megatron is initialized] datetime: 2026-08-26 18:21:21 +building GPT model ... +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + 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( +/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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 2596386816 +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: 20 +Params for bucket 1 (103022592 elements, 103022592 padded size): + module.output_layer.weight +Params for bucket 2 (48503296 elements, 48503296 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.23.pre_mlp_layernorm.weight + module.decoder.layers.23.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.23.self_attention.linear_kv_down_proj.weight + module.decoder.layers.23.self_attention.linear_q_proj.weight + module.decoder.layers.23.self_attention.linear_proj.weight + module.decoder.layers.23.input_layernorm.weight + module.decoder.layers.22.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.23.mlp.router.weight + module.decoder.layers.23.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.23.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.23.self_attention.linear_kv_up_proj.weight + module.decoder.layers.22.mlp.shared_experts.linear_fc2.weight +Params for bucket 3 (40901632 elements, 40901632 padded size): + module.decoder.layers.22.mlp.router.weight + module.decoder.layers.22.self_attention.linear_kv_up_proj.weight + module.decoder.layers.22.self_attention.linear_q_proj.weight + module.decoder.layers.22.self_attention.linear_proj.weight + module.decoder.layers.22.input_layernorm.weight + module.decoder.layers.21.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.21.mlp.router.weight + module.decoder.layers.21.self_attention.linear_kv_up_proj.weight + module.decoder.layers.22.self_attention.linear_kv_down_proj.weight + module.decoder.layers.22.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.22.pre_mlp_layernorm.weight + module.decoder.layers.21.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.21.pre_mlp_layernorm.weight + module.decoder.layers.21.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.21.self_attention.linear_kv_down_proj.weight + module.decoder.layers.21.self_attention.linear_q_proj.weight +Params for bucket 4 (41163264 elements, 41163264 padded size): + module.decoder.layers.20.mlp.router.weight + module.decoder.layers.20.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.20.self_attention.linear_kv_up_proj.weight + module.decoder.layers.20.self_attention.linear_q_proj.weight + module.decoder.layers.20.self_attention.linear_proj.weight + module.decoder.layers.20.input_layernorm.weight + module.decoder.layers.21.self_attention.linear_proj.weight + module.decoder.layers.20.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.21.input_layernorm.weight + module.decoder.layers.20.pre_mlp_layernorm.weight + module.decoder.layers.20.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.20.self_attention.linear_kv_down_proj.weight + module.decoder.layers.19.mlp.shared_experts.linear_fc2.weight +Params for bucket 5 (42734080 elements, 42734080 padded size): + module.decoder.layers.19.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.19.pre_mlp_layernorm.weight + module.decoder.layers.19.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.19.self_attention.linear_kv_down_proj.weight + module.decoder.layers.19.self_attention.linear_q_proj.weight + module.decoder.layers.19.self_attention.linear_proj.weight + module.decoder.layers.19.input_layernorm.weight + module.decoder.layers.18.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.19.self_attention.linear_kv_up_proj.weight + module.decoder.layers.19.mlp.router.weight + module.decoder.layers.18.mlp.shared_experts.linear_fc2.weight +Params for bucket 6 (40901632 elements, 40901632 padded size): + module.decoder.layers.18.self_attention.linear_kv_down_proj.weight + module.decoder.layers.18.pre_mlp_layernorm.weight + module.decoder.layers.18.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.18.self_attention.linear_q_proj.weight + module.decoder.layers.17.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.17.mlp.router.weight + module.decoder.layers.17.self_attention.linear_kv_up_proj.weight + module.decoder.layers.18.mlp.router.weight + module.decoder.layers.18.self_attention.linear_kv_up_proj.weight + module.decoder.layers.18.self_attention.linear_proj.weight + module.decoder.layers.18.input_layernorm.weight + module.decoder.layers.17.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.17.pre_mlp_layernorm.weight + module.decoder.layers.17.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.17.self_attention.linear_kv_down_proj.weight + module.decoder.layers.17.self_attention.linear_q_proj.weight +Params for bucket 7 (41163264 elements, 41163264 padded size): + module.decoder.layers.17.self_attention.linear_proj.weight + module.decoder.layers.17.input_layernorm.weight + module.decoder.layers.16.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.16.pre_mlp_layernorm.weight + module.decoder.layers.16.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.16.self_attention.linear_kv_down_proj.weight + module.decoder.layers.16.self_attention.linear_q_proj.weight + module.decoder.layers.16.self_attention.linear_proj.weight + module.decoder.layers.16.input_layernorm.weight + module.decoder.layers.16.mlp.router.weight + module.decoder.layers.16.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.16.self_attention.linear_kv_up_proj.weight + module.decoder.layers.15.mlp.shared_experts.linear_fc2.weight +Params for bucket 8 (42734080 elements, 42734080 padded size): + module.decoder.layers.15.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.15.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_kv_down_proj.weight + module.decoder.layers.15.self_attention.linear_proj.weight + module.decoder.layers.15.input_layernorm.weight + module.decoder.layers.15.self_attention.linear_kv_up_proj.weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_q_proj.weight + module.decoder.layers.14.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.14.mlp.shared_experts.linear_fc1.weight +Params for bucket 9 (40901632 elements, 40901632 padded size): + module.decoder.layers.14.self_attention.linear_kv_down_proj.weight + module.decoder.layers.14.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_q_proj.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.input_layernorm.weight + module.decoder.layers.13.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.self_attention.linear_kv_up_proj.weight + module.decoder.layers.13.self_attention.linear_q_proj.weight + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_kv_up_proj.weight + module.decoder.layers.13.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.13.pre_mlp_layernorm.weight + module.decoder.layers.13.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.13.self_attention.linear_kv_down_proj.weight +Params for bucket 10 (41163264 elements, 41163264 padded size): + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.input_layernorm.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_kv_up_proj.weight + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.12.input_layernorm.weight + module.decoder.layers.12.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.12.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.12.self_attention.linear_kv_down_proj.weight + module.decoder.layers.12.self_attention.linear_q_proj.weight + module.decoder.layers.11.mlp.shared_experts.linear_fc2.weight +Params for bucket 11 (42734080 elements, 42734080 padded size): + module.decoder.layers.11.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.11.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_kv_down_proj.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_kv_up_proj.weight + module.decoder.layers.11.self_attention.linear_q_proj.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.input_layernorm.weight + module.decoder.layers.10.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.10.mlp.shared_experts.linear_fc1.weight +Params for bucket 12 (40901632 elements, 40901632 padded size): + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.self_attention.linear_kv_down_proj.weight + module.decoder.layers.10.self_attention.linear_q_proj.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.9.self_attention.linear_kv_down_proj.weight + module.decoder.layers.9.self_attention.linear_q_proj.weight + module.decoder.layers.10.self_attention.linear_kv_up_proj.weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.input_layernorm.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_kv_up_proj.weight +Params for bucket 13 (41163264 elements, 41163264 padded size): + module.decoder.layers.9.input_layernorm.weight + module.decoder.layers.8.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.8.self_attention.linear_kv_down_proj.weight + module.decoder.layers.8.self_attention.linear_q_proj.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.8.self_attention.linear_kv_up_proj.weight + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.8.input_layernorm.weight +Params for bucket 14 (42734080 elements, 42734080 padded size): + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_kv_down_proj.weight + module.decoder.layers.7.input_layernorm.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.7.self_attention.linear_kv_up_proj.weight + module.decoder.layers.7.self_attention.linear_q_proj.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc1.weight +Params for bucket 15 (40901632 elements, 40901632 padded size): + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_kv_up_proj.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_kv_up_proj.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_kv_down_proj.weight + module.decoder.layers.6.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.6.self_attention.linear_q_proj.weight + module.decoder.layers.6.input_layernorm.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.5.self_attention.linear_kv_down_proj.weight + module.decoder.layers.5.self_attention.linear_q_proj.weight +Params for bucket 16 (41163264 elements, 41163264 padded size): + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.4.self_attention.linear_kv_up_proj.weight + module.decoder.layers.3.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.4.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.5.input_layernorm.weight + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.4.self_attention.linear_kv_down_proj.weight + module.decoder.layers.4.self_attention.linear_q_proj.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.input_layernorm.weight +Params for bucket 17 (42734080 elements, 42734080 padded size): + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_kv_up_proj.weight + module.decoder.layers.3.self_attention.linear_q_proj.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.3.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.3.self_attention.linear_kv_down_proj.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.input_layernorm.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc1.weight +Params for bucket 18 (40901632 elements, 40901632 padded size): + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_kv_up_proj.weight + module.decoder.layers.2.self_attention.linear_q_proj.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_kv_up_proj.weight + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_kv_down_proj.weight + module.decoder.layers.2.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.2.input_layernorm.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.1.self_attention.linear_kv_down_proj.weight + module.decoder.layers.1.self_attention.linear_q_proj.weight +Params for bucket 19 (71436288 elements, 71436288 padded size): + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.mlp.linear_fc2.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.layers.1.input_layernorm.weight +Params for bucket 20 (116789760 elements, 116789760 padded size): + module.decoder.layers.0.self_attention.linear_q_proj.weight + module.decoder.layers.0.self_attention.linear_kv_up_proj.weight + module.decoder.layers.0.input_layernorm.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_kv_down_proj.weight + module.decoder.layers.0.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 40 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.23.mlp.experts.linear_fc2.weight7 + module.decoder.layers.23.mlp.experts.linear_fc2.weight3 + module.decoder.layers.23.mlp.experts.linear_fc2.weight5 + module.decoder.layers.23.mlp.experts.linear_fc2.weight1 + module.decoder.layers.23.mlp.experts.linear_fc1.weight7 + module.decoder.layers.23.mlp.experts.linear_fc1.weight5 + module.decoder.layers.23.mlp.experts.linear_fc2.weight6 + module.decoder.layers.23.mlp.experts.linear_fc2.weight4 + module.decoder.layers.23.mlp.experts.linear_fc2.weight2 + module.decoder.layers.23.mlp.experts.linear_fc2.weight0 + module.decoder.layers.23.mlp.experts.linear_fc1.weight6 +Params for bucket 2 (40370176 elements, 40370176 padded size): + module.decoder.layers.23.mlp.experts.linear_fc1.weight1 + module.decoder.layers.23.mlp.experts.linear_fc1.weight3 + module.decoder.layers.22.mlp.experts.linear_fc2.weight6 + module.decoder.layers.22.mlp.experts.linear_fc2.weight4 + module.decoder.layers.23.mlp.experts.linear_fc1.weight4 + module.decoder.layers.23.mlp.experts.linear_fc1.weight2 + module.decoder.layers.23.mlp.experts.linear_fc1.weight0 + module.decoder.layers.22.mlp.experts.linear_fc2.weight7 + module.decoder.layers.22.mlp.experts.linear_fc2.weight5 +Params for bucket 3 (40370176 elements, 40370176 padded size): + module.decoder.layers.22.mlp.experts.linear_fc2.weight0 + module.decoder.layers.22.mlp.experts.linear_fc2.weight2 + module.decoder.layers.22.mlp.experts.linear_fc1.weight6 + module.decoder.layers.22.mlp.experts.linear_fc1.weight4 + module.decoder.layers.22.mlp.experts.linear_fc2.weight3 + module.decoder.layers.22.mlp.experts.linear_fc2.weight1 + module.decoder.layers.22.mlp.experts.linear_fc1.weight7 + module.decoder.layers.22.mlp.experts.linear_fc1.weight5 + module.decoder.layers.22.mlp.experts.linear_fc1.weight3 +Params for bucket 4 (40370176 elements, 40370176 padded size): + module.decoder.layers.22.mlp.experts.linear_fc1.weight2 + module.decoder.layers.21.mlp.experts.linear_fc2.weight6 + module.decoder.layers.22.mlp.experts.linear_fc1.weight0 + module.decoder.layers.21.mlp.experts.linear_fc2.weight4 + module.decoder.layers.21.mlp.experts.linear_fc2.weight2 + module.decoder.layers.21.mlp.experts.linear_fc2.weight0 + module.decoder.layers.21.mlp.experts.linear_fc2.weight7 + module.decoder.layers.22.mlp.experts.linear_fc1.weight1 + module.decoder.layers.21.mlp.experts.linear_fc2.weight5 + module.decoder.layers.21.mlp.experts.linear_fc2.weight3 + module.decoder.layers.21.mlp.experts.linear_fc2.weight1 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.21.mlp.experts.linear_fc1.weight6 + module.decoder.layers.21.mlp.experts.linear_fc1.weight4 + module.decoder.layers.21.mlp.experts.linear_fc1.weight2 + module.decoder.layers.21.mlp.experts.linear_fc1.weight7 + module.decoder.layers.21.mlp.experts.linear_fc1.weight5 + module.decoder.layers.21.mlp.experts.linear_fc1.weight3 + module.decoder.layers.21.mlp.experts.linear_fc1.weight1 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.21.mlp.experts.linear_fc1.weight0 + module.decoder.layers.20.mlp.experts.linear_fc2.weight5 + module.decoder.layers.20.mlp.experts.linear_fc2.weight7 + module.decoder.layers.20.mlp.experts.linear_fc2.weight3 + module.decoder.layers.20.mlp.experts.linear_fc2.weight1 + module.decoder.layers.20.mlp.experts.linear_fc1.weight7 + module.decoder.layers.20.mlp.experts.linear_fc2.weight6 + module.decoder.layers.20.mlp.experts.linear_fc2.weight4 + module.decoder.layers.20.mlp.experts.linear_fc2.weight2 + module.decoder.layers.20.mlp.experts.linear_fc2.weight0 + module.decoder.layers.20.mlp.experts.linear_fc1.weight6 +Params for bucket 7 (40370176 elements, 40370176 padded size): + module.decoder.layers.20.mlp.experts.linear_fc1.weight5 + module.decoder.layers.20.mlp.experts.linear_fc1.weight3 + module.decoder.layers.20.mlp.experts.linear_fc1.weight1 + module.decoder.layers.19.mlp.experts.linear_fc2.weight6 + module.decoder.layers.20.mlp.experts.linear_fc1.weight4 + module.decoder.layers.20.mlp.experts.linear_fc1.weight2 + module.decoder.layers.20.mlp.experts.linear_fc1.weight0 + module.decoder.layers.19.mlp.experts.linear_fc2.weight7 +Params for bucket 8 (40370176 elements, 40370176 padded size): + module.decoder.layers.19.mlp.experts.linear_fc2.weight4 + module.decoder.layers.19.mlp.experts.linear_fc2.weight2 + module.decoder.layers.19.mlp.experts.linear_fc2.weight0 + module.decoder.layers.19.mlp.experts.linear_fc1.weight6 + module.decoder.layers.19.mlp.experts.linear_fc1.weight4 + module.decoder.layers.19.mlp.experts.linear_fc2.weight5 + module.decoder.layers.19.mlp.experts.linear_fc2.weight1 + module.decoder.layers.19.mlp.experts.linear_fc2.weight3 + module.decoder.layers.19.mlp.experts.linear_fc1.weight7 + module.decoder.layers.19.mlp.experts.linear_fc1.weight5 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.19.mlp.experts.linear_fc1.weight2 + module.decoder.layers.19.mlp.experts.linear_fc1.weight0 + module.decoder.layers.18.mlp.experts.linear_fc2.weight6 + module.decoder.layers.18.mlp.experts.linear_fc2.weight4 + module.decoder.layers.18.mlp.experts.linear_fc2.weight2 + module.decoder.layers.19.mlp.experts.linear_fc1.weight3 + module.decoder.layers.18.mlp.experts.linear_fc2.weight7 + module.decoder.layers.19.mlp.experts.linear_fc1.weight1 + module.decoder.layers.18.mlp.experts.linear_fc2.weight5 + module.decoder.layers.18.mlp.experts.linear_fc2.weight3 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.18.mlp.experts.linear_fc2.weight0 + module.decoder.layers.18.mlp.experts.linear_fc1.weight6 + module.decoder.layers.18.mlp.experts.linear_fc1.weight4 + module.decoder.layers.18.mlp.experts.linear_fc1.weight2 + module.decoder.layers.18.mlp.experts.linear_fc2.weight1 + module.decoder.layers.18.mlp.experts.linear_fc1.weight5 + module.decoder.layers.18.mlp.experts.linear_fc1.weight7 + module.decoder.layers.18.mlp.experts.linear_fc1.weight3 +Params for bucket 11 (40370176 elements, 40370176 padded size): + module.decoder.layers.17.mlp.experts.linear_fc2.weight6 + module.decoder.layers.18.mlp.experts.linear_fc1.weight0 + module.decoder.layers.17.mlp.experts.linear_fc2.weight4 + module.decoder.layers.17.mlp.experts.linear_fc2.weight2 + module.decoder.layers.17.mlp.experts.linear_fc2.weight0 + module.decoder.layers.17.mlp.experts.linear_fc1.weight7 + module.decoder.layers.18.mlp.experts.linear_fc1.weight1 + module.decoder.layers.17.mlp.experts.linear_fc2.weight7 + module.decoder.layers.17.mlp.experts.linear_fc2.weight5 + module.decoder.layers.17.mlp.experts.linear_fc2.weight3 + module.decoder.layers.17.mlp.experts.linear_fc2.weight1 +Params for bucket 12 (40370176 elements, 40370176 padded size): + module.decoder.layers.17.mlp.experts.linear_fc1.weight3 + module.decoder.layers.17.mlp.experts.linear_fc1.weight5 + module.decoder.layers.17.mlp.experts.linear_fc1.weight1 + module.decoder.layers.17.mlp.experts.linear_fc1.weight6 + module.decoder.layers.17.mlp.experts.linear_fc1.weight2 + module.decoder.layers.17.mlp.experts.linear_fc1.weight4 + module.decoder.layers.17.mlp.experts.linear_fc1.weight0 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.16.mlp.experts.linear_fc2.weight7 + module.decoder.layers.16.mlp.experts.linear_fc2.weight3 + module.decoder.layers.16.mlp.experts.linear_fc2.weight5 + module.decoder.layers.16.mlp.experts.linear_fc2.weight1 + module.decoder.layers.16.mlp.experts.linear_fc1.weight6 + module.decoder.layers.16.mlp.experts.linear_fc2.weight6 + module.decoder.layers.16.mlp.experts.linear_fc2.weight4 + module.decoder.layers.16.mlp.experts.linear_fc2.weight2 + module.decoder.layers.16.mlp.experts.linear_fc2.weight0 + module.decoder.layers.16.mlp.experts.linear_fc1.weight7 + module.decoder.layers.16.mlp.experts.linear_fc1.weight5 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.16.mlp.experts.linear_fc1.weight4 + module.decoder.layers.16.mlp.experts.linear_fc1.weight0 + module.decoder.layers.16.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.16.mlp.experts.linear_fc1.weight3 + module.decoder.layers.16.mlp.experts.linear_fc1.weight1 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 15 (40370176 elements, 40370176 padded size): + 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.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 + 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.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 +Params for bucket 16 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + 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 + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 + 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 +Params for bucket 17 (40370176 elements, 40370176 padded size): + 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 + 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 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + 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.weight6 + 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_fc1.weight7 +Params for bucket 19 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + 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.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 +Params for bucket 20 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 + module.decoder.layers.12.mlp.experts.linear_fc2.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight3 + 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_fc2.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight4 + module.decoder.layers.12.mlp.experts.linear_fc2.weight0 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 + 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.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 +Params for bucket 22 (40370176 elements, 40370176 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_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + 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_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 +Params for bucket 23 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight6 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight4 + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 + 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.weight1 +Params for bucket 24 (40370176 elements, 40370176 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.weight6 + 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 25 (40370176 elements, 40370176 padded size): + 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 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + 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.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 + 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.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 +Params for bucket 27 (40370176 elements, 40370176 padded size): + 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.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 +Params for bucket 28 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + 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.8.mlp.experts.linear_fc1.weight1 + 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 29 (40370176 elements, 40370176 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.weight1 + 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.weight2 +Params for bucket 30 (40370176 elements, 40370176 padded size): + 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_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + 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 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 +Params for bucket 31 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 +Params for bucket 32 (40370176 elements, 40370176 padded size): + 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.weight0 + 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_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 +Params for bucket 33 (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.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 + 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.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + 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.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 + 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 +Params for bucket 36 (40370176 elements, 40370176 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.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 +Params for bucket 37 (40370176 elements, 40370176 padded size): + 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 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + 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.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 +Params for bucket 38 (40370176 elements, 40370176 padded size): + 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.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + 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_fc2.weight3 + 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_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 +Params for bucket 40 (17301504 elements, 17301504 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 +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: 2026-08-26 18:21:22 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 96 + 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=(96, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=8192, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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 eadc3756447e39d609e8aa728a91f451-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from eadc3756447e39d609e8aa728a91f451-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from eadc3756447e39d609e8aa728a91f451-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 14536 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-26 18:21:22 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (628.93, 636.44) + train/valid/test-data-iterators-setup ..........: (59.21, 66.86) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-26 18:21:22 +/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: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: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: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: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: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: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.") + [2026-08-26 18:22:00] iteration 1/ 12 | consumed samples: 8 | elapsed time per iteration (ms): 38208.1 | throughput per GPU (TFLOP/s/GPU): 3.3 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.092139E+01 | loss scale: 1.0 | grad norm: 30.620 | number of skipped iterations: 0 | number of nan iterations: 0 |Number of parameters in transformer block in billions: 13.53 + +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 13.74 +Number of parameters in most loaded shard in billions: 13.7356 +Theoretical memory footprints: weight and optimizer=98244.96 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 40509.6416015625 | max allocated: 106239.89453125 | reserved: 109990.0 | max reserved: 109990.0 +/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: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: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: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.") + [2026-08-26 18:22:01] iteration 2/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 634.6 | throughput per GPU (TFLOP/s/GPU): 201.0 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.092003E+01 | loss scale: 1.0 | grad norm: 30.566 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:01] iteration 3/ 12 | consumed samples: 24 | elapsed time per iteration (ms): 642.5 | throughput per GPU (TFLOP/s/GPU): 198.6 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 9.952471E+00 | loss scale: 1.0 | grad norm: 42.237 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-26 18:22:03] iteration 4/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 1950.2 | throughput per GPU (TFLOP/s/GPU): 65.4 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 9.476898E+00 | loss scale: 1.0 | grad norm: 7.154 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:04] iteration 5/ 12 | consumed samples: 40 | elapsed time per iteration (ms): 584.1 | throughput per GPU (TFLOP/s/GPU): 218.4 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 1.114793E+01 | loss scale: 1.0 | grad norm: 400.648 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:05] iteration 6/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 589.8 | throughput per GPU (TFLOP/s/GPU): 216.3 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 9.368878E+00 | loss scale: 1.0 | grad norm: 5.139 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:05] iteration 7/ 12 | consumed samples: 56 | elapsed time per iteration (ms): 591.4 | throughput per GPU (TFLOP/s/GPU): 215.7 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 8.975649E+00 | loss scale: 1.0 | grad norm: 3.681 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:06] iteration 8/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 587.4 | throughput per GPU (TFLOP/s/GPU): 217.2 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.785368E+00 | loss scale: 1.0 | grad norm: 5.697 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:06] iteration 9/ 12 | consumed samples: 72 | elapsed time per iteration (ms): 585.4 | throughput per GPU (TFLOP/s/GPU): 217.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.613759E+00 | loss scale: 1.0 | grad norm: 3.002 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:07] iteration 10/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 585.4 | throughput per GPU (TFLOP/s/GPU): 217.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.899922E+00 | loss scale: 1.0 | grad norm: 1155.511 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:07] iteration 11/ 12 | consumed samples: 88 | elapsed time per iteration (ms): 585.2 | throughput per GPU (TFLOP/s/GPU): 218.0 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.563369E+00 | loss scale: 1.0 | grad norm: 14.007 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:22:08] iteration 12/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 585.5 | throughput per GPU (TFLOP/s/GPU): 217.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.509861E+00 | loss scale: 1.0 | grad norm: 83.626 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-26 18:22:08 +[rank0]:[W826 18:22:09.693780828 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()) diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/mem_samples.csv b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..5024b82e4c03d1140a63ca8df0ac7a52f8d99987 --- /dev/null +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/mem_samples.csv @@ -0,0 +1,248 @@ +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, 4 +1, 4 +2, 27 +3, 4 +4, 4 +5, 4 +6, 26 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 821 +7, 821 +0, 1955 +1, 1955 +2, 1955 +3, 1955 +4, 1955 +5, 1955 +6, 1955 +7, 1955 +0, 26289 +1, 26377 +2, 26223 +3, 26223 +4, 26223 +5, 26223 +6, 26355 +7, 26377 +0, 32145 +1, 32231 +2, 32081 +3, 32079 +4, 32013 +5, 32077 +6, 32145 +7, 32231 +0, 36587 +1, 36651 +2, 36541 +3, 36529 +4, 36569 +5, 36537 +6, 36683 +7, 36701 +0, 37785 +1, 37869 +2, 37719 +3, 37717 +4, 37717 +5, 37715 +6, 37851 +7, 37869 +0, 37873 +1, 37957 +2, 37807 +3, 37805 +4, 37805 +5, 37803 +6, 37939 +7, 37957 +0, 38017 +1, 38061 +2, 37935 +3, 37941 +4, 37929 +5, 37927 +6, 38065 +7, 38067 +0, 38821 +1, 38905 +2, 38755 +3, 38755 +4, 38755 +5, 38751 +6, 38887 +7, 38905 +0, 38821 +1, 38905 +2, 38755 +3, 38755 +4, 38755 +5, 38751 +6, 38887 +7, 38905 +0, 38829 +1, 38913 +2, 38763 +3, 38763 +4, 38763 +5, 38759 +6, 38895 +7, 38913 +0, 38829 +1, 38913 +2, 38763 +3, 38763 +4, 38763 +5, 38759 +6, 38895 +7, 38913 +0, 111919 +1, 111999 +2, 111869 +3, 111833 +4, 111853 +5, 111851 +6, 111981 +7, 112005 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 112443 +1, 112529 +2, 112377 +3, 112377 +4, 112377 +5, 112375 +6, 112509 +7, 112529 +0, 115705 +1, 115791 +2, 115639 +3, 115639 +4, 115639 +5, 115637 +6, 115771 +7, 115791 +0, 116353 +1, 116439 +2, 116289 +3, 116287 +4, 116289 +5, 116287 +6, 116419 +7, 116439 +0, 116963 +1, 117117 +2, 116965 +3, 116987 +4, 116897 +5, 116959 +6, 117053 +7, 117067 +0, 92647 +1, 88353 +2, 92363 +3, 98993 +4, 88261 +5, 92315 +6, 92713 +7, 88369 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 151295 +1, 151305 +2, 151201 +3, 151201 +4, 151201 +5, 151153 +6, 151361 +7, 151305 +0, 150347 +1, 2058 +2, 778 +3, 1034 +4, 150253 +5, 2058 +6, 1034 +7, 150357 diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/run.env b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/run.env new file mode 100644 index 0000000000000000000000000000000000000000..186fc15cda04c26f2455c01b15c7ee8b67526e58 --- /dev/null +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/run.env @@ -0,0 +1,61 @@ +backend=echo +config=DSV2_LITE_L24 +family=tierB +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=24 +hidden=2048 +ffn_hidden=10944 +moe_ffn_hidden=1408 +num_experts=64 +topk=6 +experts_per_rank=8 +expert_params=8650752 +heads=16 +kv_groups=0 +kv_channels=128 +qk_layernorm=1 +shared_ffn=2816 +shared_expert_overlap=0 +mla=1 +dense_first_layers=1 +moe_layers=23 +rotary_base=10000 +rotary_scaling=40 +qkv_bias=0 +router_balancing=seq_aux_loss +aux_loss_coeff=1e-3 +router_pre_softmax=1 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-6 +seq_length=8192 +micro_batch_size=1 +global_batch_size=8 +num_microbatches=1 +assignments_per_rank=49152 +mean_tokens_per_expert=6144 +slot_policy=matched +slots_per_rank=8 +capacity_mode=static +capacity=1.0755208333333333 +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= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_recipe=mxfp8 diff --git a/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/train.log b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/train.log new file mode 100644 index 0000000000000000000000000000000000000000..733c63d34ef9a7ccd296b6482048a4943b748e07 --- /dev/null +++ b/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/train.log @@ -0,0 +1,1572 @@ +============================================================== + DSV2_LITE_L24 / echo (slots/rank=8 policy=matched) + E=64 K=6 H'=1408 m=8 + L=24 seq=8192 gbs=8 (1 microbatches) + cap_mode=static cap=1.0755208333333333 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0826 18:22:17.496000 495753 torch/distributed/run.py:766] +W0826 18:22:17.496000 495753 torch/distributed/run.py:766] ***************************************** +W0826 18:22:17.496000 495753 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. +W0826 18:22:17.496000 495753 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 ............................... False + 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 .............................. 24 + encoder_seq_length .............................. 8192 + 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 ................................. 10944 + 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 ............................... 8 + grad_reduce_in_bf16 ............................. False + gradient_accumulation_fusion .................... True + gradient_reduce_div_fusion ...................... True + group_query_attention ........................... False + head_lr_mult .................................... 1.0 + heterogeneous_layers_config_encoded_json ........ None + heterogeneous_layers_config_path ................ None + hidden_dropout .................................. 0.0 + hidden_size ..................................... 2048 + 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 .................................... 512 + 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 .............................. 0.001 + moe_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 1408 + moe_grouped_gemm ................................ True + moe_hybridep_num_sms ............................ 16 + moe_input_jitter_eps ............................ None + moe_layer_freq .................................. [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] + moe_layer_recompute ............................. False + moe_num_echo_experts ............................ 64 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.0755208333333333 + 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 .................. seq_aux_loss + moe_router_num_groups ........................... None + moe_router_padding_for_fp8 ...................... False + moe_router_pre_softmax .......................... True + moe_router_score_function ....................... softmax + moe_router_topk ................................. 6 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. 2816 + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + 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 .......................................... 0.707 + mscale_all_dim .................................. 0.707 + mtp_loss_scaling_factor ......................... 0.1 + mtp_num_layers .................................. None + multi_latent_attention .......................... True + 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-06 + normalization ................................... RMSNorm + num_attention_heads ............................. 16 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 24 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 1 + 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 .................................... True + 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 ..................................... 10000 + rotary_interleaved .............................. False + rotary_percent .................................. 1.0 + rotary_scaling_factor ........................... 40.0 + 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 ...................................... 8192 + 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 1 +> 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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 1 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 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 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 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 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 [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank 421356 is connected to is connected to is connected to is connected to is connected to is connected to 777777 peer ranks. peer ranks. peer ranks. 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 : Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 777777 + + + + + +[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]:[W826 18:22:35.516537475 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]:[W826 18:22:35.516578151 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]:[W826 18:22:35.516598451 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. +[rank7]:[W826 18:22:35.517322201 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. +[rank1]:[W826 18:22:35.518896279 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]:[W826 18:22:35.520857596 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]:[W826 18:22:35.522225971 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.034 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]:[W826 18:22:35.685576104 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: 4.061 seconds +time to initialize megatron (seconds): 7.367 +[after megatron is initialized] datetime: 2026-08-26 18:22:41 +building GPT model ... +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + warnings.warn( +/workspace/Megatron-MoE-Echo/megatron/core/transformer/transformer_config.py:1448: UserWarning: MLA config max_position_embeddings is overridden by customer input, please use the original_max_position_embeddings instead for YaRN!max_position_embeddings will be deprecated in the future.Assigned original_max_position_embeddings to max_position_embeddings if not set,and assigned max_position_embeddings back to the original value. + 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( +/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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 2596386816 +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: 20 +Params for bucket 1 (103022592 elements, 103022592 padded size): + module.output_layer.weight +Params for bucket 2 (48503296 elements, 48503296 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.23.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.23.pre_mlp_layernorm.weight + module.decoder.layers.23.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.23.self_attention.linear_kv_down_proj.weight + module.decoder.layers.23.self_attention.linear_q_proj.weight + module.decoder.layers.22.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.23.mlp.router.weight + module.decoder.layers.23.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.23.self_attention.linear_kv_up_proj.weight + module.decoder.layers.23.self_attention.linear_proj.weight + module.decoder.layers.23.input_layernorm.weight + module.decoder.layers.22.mlp.shared_experts.linear_fc1.weight +Params for bucket 3 (40901632 elements, 40901632 padded size): + module.decoder.layers.22.pre_mlp_layernorm.weight + module.decoder.layers.22.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.22.self_attention.linear_kv_down_proj.weight + module.decoder.layers.22.self_attention.linear_q_proj.weight + module.decoder.layers.22.self_attention.linear_proj.weight + module.decoder.layers.22.input_layernorm.weight + module.decoder.layers.21.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.21.mlp.router.weight + module.decoder.layers.21.self_attention.linear_kv_up_proj.weight + module.decoder.layers.22.mlp.router.weight + module.decoder.layers.22.self_attention.linear_kv_up_proj.weight + module.decoder.layers.21.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.21.pre_mlp_layernorm.weight + module.decoder.layers.21.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.21.self_attention.linear_kv_down_proj.weight + module.decoder.layers.21.self_attention.linear_q_proj.weight +Params for bucket 4 (41163264 elements, 41163264 padded size): + module.decoder.layers.21.self_attention.linear_proj.weight + module.decoder.layers.21.input_layernorm.weight + module.decoder.layers.20.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.20.pre_mlp_layernorm.weight + module.decoder.layers.20.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.20.self_attention.linear_kv_down_proj.weight + module.decoder.layers.20.self_attention.linear_q_proj.weight + module.decoder.layers.19.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.20.mlp.router.weight + module.decoder.layers.20.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.20.self_attention.linear_kv_up_proj.weight + module.decoder.layers.20.self_attention.linear_proj.weight + module.decoder.layers.20.input_layernorm.weight +Params for bucket 5 (42734080 elements, 42734080 padded size): + module.decoder.layers.19.mlp.router.weight + module.decoder.layers.19.self_attention.linear_kv_up_proj.weight + module.decoder.layers.19.self_attention.linear_proj.weight + module.decoder.layers.19.input_layernorm.weight + module.decoder.layers.18.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.19.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.19.pre_mlp_layernorm.weight + module.decoder.layers.19.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.19.self_attention.linear_kv_down_proj.weight + module.decoder.layers.19.self_attention.linear_q_proj.weight + module.decoder.layers.18.mlp.shared_experts.linear_fc2.weight +Params for bucket 6 (40901632 elements, 40901632 padded size): + module.decoder.layers.18.self_attention.linear_kv_down_proj.weight + module.decoder.layers.18.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.18.pre_mlp_layernorm.weight + module.decoder.layers.18.self_attention.linear_q_proj.weight + module.decoder.layers.17.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.17.mlp.router.weight + module.decoder.layers.17.self_attention.linear_kv_up_proj.weight + module.decoder.layers.17.self_attention.linear_q_proj.weight + module.decoder.layers.18.mlp.router.weight + module.decoder.layers.18.self_attention.linear_kv_up_proj.weight + module.decoder.layers.18.self_attention.linear_proj.weight + module.decoder.layers.18.input_layernorm.weight + module.decoder.layers.17.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.17.pre_mlp_layernorm.weight + module.decoder.layers.17.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.17.self_attention.linear_kv_down_proj.weight +Params for bucket 7 (41163264 elements, 41163264 padded size): + module.decoder.layers.16.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.16.pre_mlp_layernorm.weight + module.decoder.layers.16.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.16.self_attention.linear_proj.weight + module.decoder.layers.16.input_layernorm.weight + module.decoder.layers.17.self_attention.linear_proj.weight + module.decoder.layers.17.input_layernorm.weight + module.decoder.layers.16.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.16.mlp.router.weight + module.decoder.layers.16.self_attention.linear_kv_up_proj.weight + module.decoder.layers.16.self_attention.linear_kv_down_proj.weight + module.decoder.layers.16.self_attention.linear_q_proj.weight + module.decoder.layers.15.mlp.shared_experts.linear_fc2.weight +Params for bucket 8 (42734080 elements, 42734080 padded size): + module.decoder.layers.15.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.15.self_attention.linear_kv_up_proj.weight + module.decoder.layers.15.self_attention.linear_proj.weight + module.decoder.layers.15.input_layernorm.weight + module.decoder.layers.14.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_kv_down_proj.weight + module.decoder.layers.15.self_attention.linear_q_proj.weight + module.decoder.layers.14.mlp.shared_experts.linear_fc2.weight +Params for bucket 9 (40901632 elements, 40901632 padded size): + module.decoder.layers.14.self_attention.linear_kv_down_proj.weight + module.decoder.layers.14.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.13.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.pre_mlp_layernorm.weight + module.decoder.layers.13.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.13.self_attention.linear_kv_down_proj.weight + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_kv_up_proj.weight + module.decoder.layers.14.self_attention.linear_q_proj.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.input_layernorm.weight + module.decoder.layers.13.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.13.self_attention.linear_kv_up_proj.weight + module.decoder.layers.13.self_attention.linear_q_proj.weight +Params for bucket 10 (41163264 elements, 41163264 padded size): + module.decoder.layers.12.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.12.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.12.self_attention.linear_kv_down_proj.weight + module.decoder.layers.12.self_attention.linear_q_proj.weight + module.decoder.layers.11.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.input_layernorm.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_kv_up_proj.weight + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.12.input_layernorm.weight +Params for bucket 11 (42734080 elements, 42734080 padded size): + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_kv_up_proj.weight + module.decoder.layers.10.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.11.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.11.self_attention.linear_kv_down_proj.weight + module.decoder.layers.11.self_attention.linear_q_proj.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.input_layernorm.weight + module.decoder.layers.10.mlp.shared_experts.linear_fc1.weight +Params for bucket 12 (40901632 elements, 40901632 padded size): + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.self_attention.linear_kv_down_proj.weight + module.decoder.layers.10.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.10.self_attention.linear_q_proj.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.9.self_attention.linear_kv_down_proj.weight + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_kv_up_proj.weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.input_layernorm.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_kv_up_proj.weight + module.decoder.layers.9.self_attention.linear_q_proj.weight +Params for bucket 13 (41163264 elements, 41163264 padded size): + module.decoder.layers.8.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.8.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.8.self_attention.linear_kv_down_proj.weight + module.decoder.layers.8.self_attention.linear_q_proj.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.input_layernorm.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_kv_up_proj.weight + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.8.input_layernorm.weight +Params for bucket 14 (42734080 elements, 42734080 padded size): + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.7.self_attention.linear_kv_down_proj.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.7.input_layernorm.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_kv_up_proj.weight + module.decoder.layers.7.self_attention.linear_q_proj.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc1.weight +Params for bucket 15 (40901632 elements, 40901632 padded size): + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.6.self_attention.linear_kv_down_proj.weight + module.decoder.layers.6.self_attention.linear_q_proj.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_kv_up_proj.weight + module.decoder.layers.5.self_attention.linear_q_proj.weight + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_kv_up_proj.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.input_layernorm.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.5.self_attention.linear_kv_down_proj.weight +Params for bucket 16 (41163264 elements, 41163264 padded size): + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.4.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.4.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.5.input_layernorm.weight + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.4.self_attention.linear_kv_down_proj.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.input_layernorm.weight + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_kv_up_proj.weight + module.decoder.layers.4.self_attention.linear_q_proj.weight + module.decoder.layers.3.mlp.shared_experts.linear_fc2.weight +Params for bucket 17 (42734080 elements, 42734080 padded size): + module.decoder.layers.3.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.3.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_kv_down_proj.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.3.self_attention.linear_kv_up_proj.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_q_proj.weight + module.decoder.layers.3.input_layernorm.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc1.weight +Params for bucket 18 (40901632 elements, 40901632 padded size): + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_kv_down_proj.weight + module.decoder.layers.2.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.2.self_attention.linear_q_proj.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc1.weight + module.decoder.layers.1.self_attention.linear_kv_up_proj.weight + module.decoder.layers.1.self_attention.linear_kv_down_proj.weight + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_kv_up_proj.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.2.input_layernorm.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.1.self_attention.linear_q_proj.weight +Params for bucket 19 (71436288 elements, 71436288 padded size): + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.1.input_layernorm.weight + module.decoder.layers.0.mlp.linear_fc1.weight + module.decoder.layers.0.mlp.linear_fc2.weight +Params for bucket 20 (116789760 elements, 116789760 padded size): + module.decoder.layers.0.self_attention.linear_kv_up_proj.weight + module.embedding.word_embeddings.weight + module.decoder.layers.0.mlp.linear_fc1.layer_norm_weight + module.decoder.layers.0.self_attention.linear_kv_up_proj.layer_norm_weight + module.decoder.layers.0.self_attention.linear_kv_down_proj.weight + module.decoder.layers.0.self_attention.linear_q_proj.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.input_layernorm.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 40 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.23.mlp.experts.linear_fc2.weight6 + module.decoder.layers.23.mlp.experts.linear_fc2.weight4 + module.decoder.layers.23.mlp.experts.linear_fc2.weight2 + module.decoder.layers.23.mlp.experts.linear_fc2.weight0 + module.decoder.layers.23.mlp.experts.linear_fc1.weight7 + module.decoder.layers.23.mlp.experts.linear_fc1.weight5 + module.decoder.layers.23.mlp.experts.linear_fc2.weight7 + module.decoder.layers.23.mlp.experts.linear_fc2.weight5 + module.decoder.layers.23.mlp.experts.linear_fc2.weight3 + module.decoder.layers.23.mlp.experts.linear_fc2.weight1 + module.decoder.layers.23.mlp.experts.linear_fc1.weight6 +Params for bucket 2 (40370176 elements, 40370176 padded size): + module.decoder.layers.23.mlp.experts.linear_fc1.weight3 + module.decoder.layers.23.mlp.experts.linear_fc1.weight1 + module.decoder.layers.22.mlp.experts.linear_fc2.weight6 + module.decoder.layers.22.mlp.experts.linear_fc2.weight4 + module.decoder.layers.23.mlp.experts.linear_fc1.weight4 + module.decoder.layers.23.mlp.experts.linear_fc1.weight0 + module.decoder.layers.23.mlp.experts.linear_fc1.weight2 + module.decoder.layers.22.mlp.experts.linear_fc2.weight7 + module.decoder.layers.22.mlp.experts.linear_fc2.weight5 +Params for bucket 3 (40370176 elements, 40370176 padded size): + module.decoder.layers.22.mlp.experts.linear_fc2.weight0 + module.decoder.layers.22.mlp.experts.linear_fc1.weight7 + module.decoder.layers.22.mlp.experts.linear_fc2.weight2 + module.decoder.layers.22.mlp.experts.linear_fc1.weight5 + module.decoder.layers.22.mlp.experts.linear_fc1.weight3 + module.decoder.layers.22.mlp.experts.linear_fc2.weight3 + module.decoder.layers.22.mlp.experts.linear_fc2.weight1 + module.decoder.layers.22.mlp.experts.linear_fc1.weight6 + module.decoder.layers.22.mlp.experts.linear_fc1.weight4 +Params for bucket 4 (40370176 elements, 40370176 padded size): + module.decoder.layers.21.mlp.experts.linear_fc2.weight6 + module.decoder.layers.22.mlp.experts.linear_fc1.weight1 + module.decoder.layers.21.mlp.experts.linear_fc2.weight4 + module.decoder.layers.21.mlp.experts.linear_fc2.weight2 + module.decoder.layers.21.mlp.experts.linear_fc2.weight0 + module.decoder.layers.22.mlp.experts.linear_fc1.weight2 + module.decoder.layers.21.mlp.experts.linear_fc2.weight7 + module.decoder.layers.22.mlp.experts.linear_fc1.weight0 + module.decoder.layers.21.mlp.experts.linear_fc2.weight5 + module.decoder.layers.21.mlp.experts.linear_fc2.weight3 + module.decoder.layers.21.mlp.experts.linear_fc2.weight1 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.21.mlp.experts.linear_fc1.weight7 + module.decoder.layers.21.mlp.experts.linear_fc1.weight3 + module.decoder.layers.21.mlp.experts.linear_fc1.weight5 + module.decoder.layers.21.mlp.experts.linear_fc1.weight1 + module.decoder.layers.21.mlp.experts.linear_fc1.weight6 + module.decoder.layers.21.mlp.experts.linear_fc1.weight4 + module.decoder.layers.21.mlp.experts.linear_fc1.weight2 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.20.mlp.experts.linear_fc2.weight7 + module.decoder.layers.20.mlp.experts.linear_fc2.weight5 + module.decoder.layers.20.mlp.experts.linear_fc2.weight3 + module.decoder.layers.20.mlp.experts.linear_fc2.weight1 + module.decoder.layers.20.mlp.experts.linear_fc1.weight6 + module.decoder.layers.21.mlp.experts.linear_fc1.weight0 + module.decoder.layers.20.mlp.experts.linear_fc2.weight6 + module.decoder.layers.20.mlp.experts.linear_fc2.weight4 + module.decoder.layers.20.mlp.experts.linear_fc2.weight2 + module.decoder.layers.20.mlp.experts.linear_fc2.weight0 + module.decoder.layers.20.mlp.experts.linear_fc1.weight7 +Params for bucket 7 (40370176 elements, 40370176 padded size): + module.decoder.layers.20.mlp.experts.linear_fc1.weight2 + module.decoder.layers.20.mlp.experts.linear_fc1.weight4 + module.decoder.layers.20.mlp.experts.linear_fc1.weight0 + module.decoder.layers.19.mlp.experts.linear_fc2.weight6 + module.decoder.layers.20.mlp.experts.linear_fc1.weight5 + module.decoder.layers.20.mlp.experts.linear_fc1.weight3 + module.decoder.layers.20.mlp.experts.linear_fc1.weight1 + module.decoder.layers.19.mlp.experts.linear_fc2.weight7 +Params for bucket 8 (40370176 elements, 40370176 padded size): + module.decoder.layers.19.mlp.experts.linear_fc2.weight2 + module.decoder.layers.19.mlp.experts.linear_fc2.weight4 + module.decoder.layers.19.mlp.experts.linear_fc2.weight0 + module.decoder.layers.19.mlp.experts.linear_fc1.weight7 + module.decoder.layers.19.mlp.experts.linear_fc1.weight5 + module.decoder.layers.19.mlp.experts.linear_fc2.weight5 + module.decoder.layers.19.mlp.experts.linear_fc2.weight1 + module.decoder.layers.19.mlp.experts.linear_fc2.weight3 + module.decoder.layers.19.mlp.experts.linear_fc1.weight6 + module.decoder.layers.19.mlp.experts.linear_fc1.weight4 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.19.mlp.experts.linear_fc1.weight3 + module.decoder.layers.19.mlp.experts.linear_fc1.weight1 + module.decoder.layers.18.mlp.experts.linear_fc2.weight7 + module.decoder.layers.18.mlp.experts.linear_fc2.weight5 + module.decoder.layers.18.mlp.experts.linear_fc2.weight3 + module.decoder.layers.19.mlp.experts.linear_fc1.weight2 + module.decoder.layers.19.mlp.experts.linear_fc1.weight0 + module.decoder.layers.18.mlp.experts.linear_fc2.weight6 + module.decoder.layers.18.mlp.experts.linear_fc2.weight4 + module.decoder.layers.18.mlp.experts.linear_fc2.weight2 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.18.mlp.experts.linear_fc2.weight1 + module.decoder.layers.18.mlp.experts.linear_fc1.weight6 + module.decoder.layers.18.mlp.experts.linear_fc1.weight4 + module.decoder.layers.18.mlp.experts.linear_fc1.weight2 + module.decoder.layers.18.mlp.experts.linear_fc1.weight5 + module.decoder.layers.18.mlp.experts.linear_fc2.weight0 + module.decoder.layers.18.mlp.experts.linear_fc1.weight7 + module.decoder.layers.18.mlp.experts.linear_fc1.weight3 +Params for bucket 11 (40370176 elements, 40370176 padded size): + module.decoder.layers.17.mlp.experts.linear_fc2.weight6 + module.decoder.layers.18.mlp.experts.linear_fc1.weight0 + module.decoder.layers.17.mlp.experts.linear_fc2.weight4 + module.decoder.layers.17.mlp.experts.linear_fc2.weight2 + module.decoder.layers.17.mlp.experts.linear_fc2.weight0 + module.decoder.layers.17.mlp.experts.linear_fc1.weight7 + module.decoder.layers.18.mlp.experts.linear_fc1.weight1 + module.decoder.layers.17.mlp.experts.linear_fc2.weight7 + module.decoder.layers.17.mlp.experts.linear_fc2.weight5 + module.decoder.layers.17.mlp.experts.linear_fc2.weight3 + module.decoder.layers.17.mlp.experts.linear_fc2.weight1 +Params for bucket 12 (40370176 elements, 40370176 padded size): + module.decoder.layers.17.mlp.experts.linear_fc1.weight3 + module.decoder.layers.17.mlp.experts.linear_fc1.weight5 + module.decoder.layers.17.mlp.experts.linear_fc1.weight1 + module.decoder.layers.17.mlp.experts.linear_fc1.weight6 + module.decoder.layers.17.mlp.experts.linear_fc1.weight4 + module.decoder.layers.17.mlp.experts.linear_fc1.weight2 + module.decoder.layers.17.mlp.experts.linear_fc1.weight0 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.16.mlp.experts.linear_fc2.weight6 + module.decoder.layers.16.mlp.experts.linear_fc2.weight4 + module.decoder.layers.16.mlp.experts.linear_fc2.weight2 + module.decoder.layers.16.mlp.experts.linear_fc2.weight0 + module.decoder.layers.16.mlp.experts.linear_fc1.weight7 + module.decoder.layers.16.mlp.experts.linear_fc1.weight5 + module.decoder.layers.16.mlp.experts.linear_fc2.weight7 + module.decoder.layers.16.mlp.experts.linear_fc2.weight3 + module.decoder.layers.16.mlp.experts.linear_fc2.weight5 + module.decoder.layers.16.mlp.experts.linear_fc2.weight1 + module.decoder.layers.16.mlp.experts.linear_fc1.weight6 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.16.mlp.experts.linear_fc1.weight3 + module.decoder.layers.16.mlp.experts.linear_fc1.weight1 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 + module.decoder.layers.16.mlp.experts.linear_fc1.weight4 + module.decoder.layers.16.mlp.experts.linear_fc1.weight2 + module.decoder.layers.16.mlp.experts.linear_fc1.weight0 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 +Params for bucket 15 (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.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 +Params for bucket 16 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + 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.15.mlp.experts.linear_fc1.weight1 + 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 +Params for bucket 17 (40370176 elements, 40370176 padded size): + 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 + 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 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + 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_fc1.weight7 + 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_fc1.weight6 +Params for bucket 19 (40370176 elements, 40370176 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.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + 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.weight7 +Params for bucket 20 (40370176 elements, 40370176 padded size): + 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_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + 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_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 +Params for bucket 21 (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.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 + module.decoder.layers.11.mlp.experts.linear_fc2.weight2 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 + module.decoder.layers.11.mlp.experts.linear_fc2.weight3 +Params for bucket 22 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + 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.weight3 + 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_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 +Params for bucket 23 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 + 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.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + 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.weight2 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 +Params for bucket 24 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 + 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.weight6 + 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 25 (40370176 elements, 40370176 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.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight5 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight4 +Params for bucket 27 (40370176 elements, 40370176 padded size): + 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.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + 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.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 +Params for bucket 28 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + 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.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + 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 29 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + 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.weight2 +Params for bucket 30 (40370176 elements, 40370176 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.weight1 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + 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_fc1.weight7 +Params for bucket 31 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + 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.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 +Params for bucket 32 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + 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_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 +Params for bucket 33 (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.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 +Params for bucket 34 (40370176 elements, 40370176 padded size): + 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.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + 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 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +Params for bucket 36 (40370176 elements, 40370176 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.weight0 + 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 +Params for bucket 37 (40370176 elements, 40370176 padded size): + 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 + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + 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_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.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + 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.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + 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_fc2.weight3 + 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_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 +Params for bucket 40 (17301504 elements, 17301504 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 +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: 2026-08-26 18:22:42 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 96 + 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=(96, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=8192, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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 eadc3756447e39d609e8aa728a91f451-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from eadc3756447e39d609e8aa728a91f451-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from eadc3756447e39d609e8aa728a91f451-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 14536 +> finished creating GPT datasets ... +[after dataloaders are built] datetime: 2026-08-26 18:22:42 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (621.97, 634.77) + train/valid/test-data-iterators-setup ..........: (61.61, 72.04) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-26 18:22:42 +/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: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: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: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: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: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: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.") + [2026-08-26 18:23:20] iteration 1/ 12 | consumed samples: 8 | elapsed time per iteration (ms): 38367.7 | throughput per GPU (TFLOP/s/GPU): 3.3 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.092139E+01 | loss scale: 1.0 | grad norm: 30.620 | number of skipped iterations: 0 | number of nan iterations: 0 |Number of parameters in transformer block in billions: 13.53 + +Number of parameters in embedding layers in billions: 0.21 +Total number of parameters in billions: 13.74 +Number of parameters in most loaded shard in billions: 13.7356 +Theoretical memory footprints: weight and optimizer=98244.96 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 40509.6416015625 | max allocated: 106239.89453125 | reserved: 109990.0 | max reserved: 109990.0 +/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.") + [2026-08-26 18:23:21] iteration 2/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 633.8 | throughput per GPU (TFLOP/s/GPU): 201.3 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.092003E+01 | loss scale: 1.0 | grad norm: 30.566 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:22] iteration 3/ 12 | consumed samples: 24 | elapsed time per iteration (ms): 634.4 | throughput per GPU (TFLOP/s/GPU): 201.1 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 9.952471E+00 | loss scale: 1.0 | grad norm: 42.237 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-26 18:23:24] iteration 4/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 1951.4 | throughput per GPU (TFLOP/s/GPU): 65.4 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 9.476898E+00 | loss scale: 1.0 | grad norm: 7.154 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:24] iteration 5/ 12 | consumed samples: 40 | elapsed time per iteration (ms): 584.2 | throughput per GPU (TFLOP/s/GPU): 218.4 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 1.114793E+01 | loss scale: 1.0 | grad norm: 400.648 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:25] iteration 6/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 589.5 | throughput per GPU (TFLOP/s/GPU): 216.4 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 9.368878E+00 | loss scale: 1.0 | grad norm: 5.139 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:25] iteration 7/ 12 | consumed samples: 56 | elapsed time per iteration (ms): 589.1 | throughput per GPU (TFLOP/s/GPU): 216.6 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 8.975649E+00 | loss scale: 1.0 | grad norm: 3.681 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:26] iteration 8/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 587.3 | throughput per GPU (TFLOP/s/GPU): 217.2 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.785368E+00 | loss scale: 1.0 | grad norm: 5.697 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:26] iteration 9/ 12 | consumed samples: 72 | elapsed time per iteration (ms): 589.6 | throughput per GPU (TFLOP/s/GPU): 216.4 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.613759E+00 | loss scale: 1.0 | grad norm: 3.002 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:27] iteration 10/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 588.9 | throughput per GPU (TFLOP/s/GPU): 216.6 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.899922E+00 | loss scale: 1.0 | grad norm: 1155.511 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:28] iteration 11/ 12 | consumed samples: 88 | elapsed time per iteration (ms): 585.7 | throughput per GPU (TFLOP/s/GPU): 217.8 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.563369E+00 | loss scale: 1.0 | grad norm: 14.007 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:23:28] iteration 12/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 588.2 | throughput per GPU (TFLOP/s/GPU): 216.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.509861E+00 | loss scale: 1.0 | grad norm: 83.626 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-26 18:23:28 +[rank0]:[W826 18:23:29.981402789 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()) diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/mem_samples.csv b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..5a72900af8a8ebd834f45337655230dab3df4e8e --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/mem_samples.csv @@ -0,0 +1,240 @@ +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, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 821 +7, 821 +0, 1803 +1, 1803 +2, 1803 +3, 1803 +4, 1803 +5, 1803 +6, 1803 +7, 1803 +0, 41093 +1, 41093 +2, 41093 +3, 41093 +4, 41093 +5, 41093 +6, 41093 +7, 41093 +0, 43905 +1, 43335 +2, 43335 +3, 44035 +4, 44035 +5, 44035 +6, 44035 +7, 44035 +0, 44559 +1, 43991 +2, 43991 +3, 44691 +4, 44691 +5, 44691 +6, 44691 +7, 44691 +0, 47239 +1, 46671 +2, 46671 +3, 47371 +4, 47371 +5, 47371 +6, 47371 +7, 47371 +0, 47519 +1, 46951 +2, 46951 +3, 47651 +4, 47651 +5, 47651 +6, 47651 +7, 47651 +0, 47917 +1, 47349 +2, 47349 +3, 48049 +4, 48049 +5, 48049 +6, 48043 +7, 48049 +0, 48553 +1, 47985 +2, 47985 +3, 48685 +4, 48685 +5, 48685 +6, 48685 +7, 48685 +0, 48799 +1, 48231 +2, 48231 +3, 48931 +4, 48931 +5, 48931 +6, 48931 +7, 48931 +0, 49933 +1, 49365 +2, 49365 +3, 50065 +4, 50065 +5, 50065 +6, 50065 +7, 50065 +0, 49933 +1, 49365 +2, 49365 +3, 50065 +4, 50065 +5, 50065 +6, 50065 +7, 50065 +0, 77603 +1, 77619 +2, 77649 +3, 77629 +4, 77631 +5, 77601 +6, 77611 +7, 77645 +0, 78077 +1, 78061 +2, 78059 +3, 78081 +4, 78081 +5, 78081 +6, 78081 +7, 78081 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 79529 +1, 79513 +2, 79511 +3, 79533 +4, 79545 +5, 79533 +6, 79533 +7, 79533 +0, 82709 +1, 81573 +2, 82691 +3, 82713 +4, 82993 +5, 82993 +6, 81873 +7, 81593 +0, 90017 +1, 88743 +2, 89937 +3, 89943 +4, 90169 +5, 90187 +6, 89065 +7, 88841 +0, 109745 +1, 103473 +2, 104119 +3, 105879 +4, 108941 +5, 105755 +6, 109781 +7, 103825 +0, 112291 +1, 111969 +2, 112055 +3, 112327 +4, 112327 +5, 112327 +6, 112327 +7, 112327 +0, 111397 +1, 111075 +2, 3918 +3, 111433 +4, 111433 +5, 111433 +6, 111433 +7, 111433 +0, 3662 +1, 2382 +2, 2382 +3, 2382 +4, 2382 +5, 2382 +6, 2638 +7, 111433 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/run.env b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/run.env new file mode 100644 index 0000000000000000000000000000000000000000..4d9a90354475d3bad1a46a58ada48191e142a80b --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/run.env @@ -0,0 +1,61 @@ +backend=echo +config=QWEN2_57B +family=tierB +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=10 +hidden=3584 +ffn_hidden=2560 +moe_ffn_hidden=2560 +num_experts=64 +topk=8 +experts_per_rank=8 +expert_params=27525120 +heads=28 +kv_groups=4 +kv_channels= +qk_layernorm=0 +shared_ffn=20480 +shared_expert_overlap=0 +mla=0 +dense_first_layers=0 +moe_layers=10 +rotary_base=1000000 +rotary_scaling= +qkv_bias=1 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=1 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-6 +seq_length=4096 +micro_batch_size=1 +global_batch_size=8 +num_microbatches=1 +assignments_per_rank=32768 +mean_tokens_per_expert=4096 +slot_policy=matched +slots_per_rank=8 +capacity_mode=static +capacity=1.08984375 +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= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_recipe=mxfp8 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/train.log b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/train.log new file mode 100644 index 0000000000000000000000000000000000000000..45a1293edb1d490c76b9d68682f6d640e09d5091 --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/train.log @@ -0,0 +1,1199 @@ +============================================================== + QWEN2_57B / echo (slots/rank=8 policy=matched) + E=64 K=8 H'=2560 m=8 + L=10 seq=4096 gbs=8 (1 microbatches) + cap_mode=static cap=1.08984375 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0826 18:23:37.850000 499283 torch/distributed/run.py:766] +W0826 18:23:37.850000 499283 torch/distributed/run.py:766] ***************************************** +W0826 18:23:37.850000 499283 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. +W0826 18:23:37.850000 499283 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 ------------------------ +/workspace/Megatron-MoE-Echo/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") + 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 .................................... True + 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 .............................. 10 + 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 ................................. 2560 + 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 ............................... 8 + 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 ..................................... 3584 + 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_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 2560 + 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 ............................ 64 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.08984375 + 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 .......................... True + moe_router_score_function ....................... softmax + moe_router_topk ................................. 8 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. 20480 + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + 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-06 + normalization ................................... RMSNorm + num_attention_heads ............................. 28 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 10 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 4 + 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 1 +> 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-MoE-Echo/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 2 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 [Gloo] Rank [Gloo] Rank [Gloo] Rank 3165 is connected to is connected to is connected to is connected to 7777 peer ranks. 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 : Expected number of connected peer ranks is : 7777 + + + +[Gloo] Rank [Gloo] Rank 74 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77 + +[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 [Gloo] Rank 5 is connected to 77 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is : [Gloo] Rank +76 + 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. [Gloo] Rank Expected number of connected peer ranks is : 00 + 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 + +[rank5]:[W826 18:23:55.923229216 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]:[W826 18:23:55.924199894 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]:[W826 18:23:55.924318434 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]:[W826 18:23:55.924371588 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]:[W826 18:23:55.926705726 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. +[rank7]:[W826 18:23:55.926744842 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]:[W826 18:23:55.926751638 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.038 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]:[W826 18:23:56.090693413 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: 3.626 seconds +time to initialize megatron (seconds): 6.914 +[after megatron is initialized] datetime: 2026-08-26 18:24:01 +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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 5060614656 +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: 22 +Params for bucket 1 (180289536 elements, 180289536 padded size): + module.output_layer.weight +Params for bucket 2 (73403904 elements, 73403904 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc2.weight +Params for bucket 3 (146800640 elements, 146800640 padded size): + module.decoder.layers.9.mlp.shared_experts.linear_fc1.weight +Params for bucket 4 (103001600 elements, 103001600 padded size): + 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_qkv.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_qkv.bias + module.decoder.layers.8.mlp.shared_experts.linear_fc2.weight +Params for bucket 5 (146800640 elements, 146800640 padded size): + module.decoder.layers.8.mlp.shared_experts.linear_fc1.weight +Params for bucket 6 (103001600 elements, 103001600 padded size): + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_qkv.bias + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_qkv.weight +Params for bucket 7 (146800640 elements, 146800640 padded size): + module.decoder.layers.7.mlp.shared_experts.linear_fc1.weight +Params for bucket 8 (103001600 elements, 103001600 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_qkv.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_qkv.bias +Params for bucket 9 (146800640 elements, 146800640 padded size): + module.decoder.layers.6.mlp.shared_experts.linear_fc1.weight +Params for bucket 10 (103001600 elements, 103001600 padded size): + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_qkv.bias + module.decoder.layers.5.mlp.shared_experts.linear_fc2.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_qkv.weight + module.decoder.layers.6.self_attention.linear_proj.weight +Params for bucket 11 (146800640 elements, 146800640 padded size): + module.decoder.layers.5.mlp.shared_experts.linear_fc1.weight +Params for bucket 12 (103001600 elements, 103001600 padded size): + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.self_attention.linear_qkv.bias + module.decoder.layers.4.mlp.shared_experts.linear_fc2.weight +Params for bucket 13 (146800640 elements, 146800640 padded size): + module.decoder.layers.4.mlp.shared_experts.linear_fc1.weight +Params for bucket 14 (103001600 elements, 103001600 padded size): + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.3.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_qkv.bias + module.decoder.layers.4.pre_mlp_layernorm.weight +Params for bucket 15 (146800640 elements, 146800640 padded size): + module.decoder.layers.3.mlp.shared_experts.linear_fc1.weight +Params for bucket 16 (103001600 elements, 103001600 padded size): + module.decoder.layers.3.self_attention.linear_qkv.bias + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc2.weight +Params for bucket 17 (146800640 elements, 146800640 padded size): + module.decoder.layers.2.mlp.shared_experts.linear_fc1.weight +Params for bucket 18 (103001600 elements, 103001600 padded size): + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.2.self_attention.linear_qkv.bias + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.2.self_attention.linear_qkv.weight +Params for bucket 19 (146800640 elements, 146800640 padded size): + module.decoder.layers.1.mlp.shared_experts.linear_fc1.weight +Params for bucket 20 (103001600 elements, 103001600 padded size): + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.1.pre_mlp_layernorm.weight +Params for bucket 21 (146800640 elements, 146800640 padded size): + module.decoder.layers.0.mlp.shared_experts.linear_fc1.weight +Params for bucket 22 (209890816 elements, 209890816 padded size): + module.decoder.layers.0.self_attention.linear_qkv.bias + module.decoder.layers.0.pre_mlp_layernorm.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 + module.decoder.layers.0.self_attention.linear_proj.weight + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 44 +Params for bucket 1 (45875200 elements, 45875200 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 +Params for bucket 2 (45875200 elements, 45875200 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 +Params for bucket 3 (55050240 elements, 55050240 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.weight5 +Params for bucket 4 (55050240 elements, 55050240 padded size): + 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.weight2 +Params for bucket 5 (45875200 elements, 45875200 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 +Params for bucket 6 (45875200 elements, 45875200 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + 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 +Params for bucket 7 (55050240 elements, 55050240 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 +Params for bucket 8 (55050240 elements, 55050240 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 +Params for bucket 9 (45875200 elements, 45875200 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 +Params for bucket 10 (45875200 elements, 45875200 padded size): + 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.weight4 + module.decoder.layers.7.mlp.experts.linear_fc2.weight3 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 +Params for bucket 11 (55050240 elements, 55050240 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 12 (55050240 elements, 55050240 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 13 (55050240 elements, 55050240 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 14 (45875200 elements, 45875200 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 +Params for bucket 15 (45875200 elements, 45875200 padded size): + 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_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight1 +Params for bucket 16 (55050240 elements, 55050240 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 17 (55050240 elements, 55050240 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 +Params for bucket 18 (45875200 elements, 45875200 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight5 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 +Params for bucket 19 (45875200 elements, 45875200 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + 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.weight0 +Params for bucket 20 (55050240 elements, 55050240 padded size): + 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.weight6 +Params for bucket 21 (55050240 elements, 55050240 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 +Params for bucket 22 (45875200 elements, 45875200 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 +Params for bucket 23 (45875200 elements, 45875200 padded size): + 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 24 (55050240 elements, 55050240 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 +Params for bucket 25 (55050240 elements, 55050240 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 26 (55050240 elements, 55050240 padded size): + 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.weight1 +Params for bucket 27 (45875200 elements, 45875200 padded size): + 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.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 +Params for bucket 28 (45875200 elements, 45875200 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + 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_fc2.weight1 +Params for bucket 29 (55050240 elements, 55050240 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.weight5 +Params for bucket 30 (55050240 elements, 55050240 padded size): + 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.weight2 +Params for bucket 31 (45875200 elements, 45875200 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 +Params for bucket 32 (45875200 elements, 45875200 padded size): + 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_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 +Params for bucket 33 (55050240 elements, 55050240 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 +Params for bucket 34 (55050240 elements, 55050240 padded size): + 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.weight3 +Params for bucket 35 (45875200 elements, 45875200 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 +Params for bucket 36 (45875200 elements, 45875200 padded size): + 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.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 +Params for bucket 37 (55050240 elements, 55050240 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 +Params for bucket 38 (55050240 elements, 55050240 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 39 (55050240 elements, 55050240 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 40 (45875200 elements, 45875200 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 +Params for bucket 41 (45875200 elements, 45875200 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 +Params for bucket 42 (55050240 elements, 55050240 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 +Params for bucket 43 (55050240 elements, 55050240 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.weight1 +Params for bucket 44 (18350080 elements, 18350080 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 +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: 2026-08-26 18:24:02 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 96 + 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=(96, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 466b63c71b42dae59ebdc25db6336051-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: 2026-08-26 18:24:02 +done with setup ... +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (370.39, 376.11) + train/valid/test-data-iterators-setup ..........: (61.27, 78.86) +[before the start of training step] datetime: 2026-08-26 18:24:02 +/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: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: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.") + [2026-08-26 18:24:39] iteration 1/ 12 | consumed samples: 8 | elapsed time per iteration (ms): 37654.1 | throughput per GPU (TFLOP/s/GPU): 3.3 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.097931E+01 | loss scale: 1.0 | grad norm: 29.879 | number of skipped iterations: 0 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 20.11 +Number of parameters in embedding layers in billions: 0.36 +Total number of parameters in billions: 20.47 +Number of parameters in most loaded shard in billions: 20.4724 +Theoretical memory footprints: weight and optimizer=146430.14 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 68733.94384765625 | max allocated: 70397.458984375 | reserved: 81416.0 | max reserved: 81416.0 +/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.") + [2026-08-26 18:24:40] iteration 2/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 430.9 | throughput per GPU (TFLOP/s/GPU): 286.6 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.097906E+01 | loss scale: 1.0 | grad norm: 30.625 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:40] iteration 3/ 12 | consumed samples: 24 | elapsed time per iteration (ms): 227.6 | throughput per GPU (TFLOP/s/GPU): 542.5 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 9.538181E+00 | loss scale: 1.0 | grad norm: 32.388 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-26 18:24:42] iteration 4/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 1760.2 | throughput per GPU (TFLOP/s/GPU): 70.2 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 1.109149E+01 | loss scale: 1.0 | grad norm: 107.126 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:42] iteration 5/ 12 | consumed samples: 40 | elapsed time per iteration (ms): 181.6 | throughput per GPU (TFLOP/s/GPU): 680.0 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 9.997242E+00 | loss scale: 1.0 | grad norm: 26.217 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:42] iteration 6/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 180.5 | throughput per GPU (TFLOP/s/GPU): 684.2 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 1.049742E+01 | loss scale: 1.0 | grad norm: 8.721 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:42] iteration 7/ 12 | consumed samples: 56 | elapsed time per iteration (ms): 179.4 | throughput per GPU (TFLOP/s/GPU): 688.4 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 9.440645E+00 | loss scale: 1.0 | grad norm: 5.389 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:42] iteration 8/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 178.5 | throughput per GPU (TFLOP/s/GPU): 691.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.962455E+00 | loss scale: 1.0 | grad norm: 4.550 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:43] iteration 9/ 12 | consumed samples: 72 | elapsed time per iteration (ms): 178.3 | throughput per GPU (TFLOP/s/GPU): 692.5 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.746603E+00 | loss scale: 1.0 | grad norm: 4.297 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:43] iteration 10/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 178.7 | throughput per GPU (TFLOP/s/GPU): 690.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.686147E+00 | loss scale: 1.0 | grad norm: 4.422 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:43] iteration 11/ 12 | consumed samples: 88 | elapsed time per iteration (ms): 177.4 | throughput per GPU (TFLOP/s/GPU): 696.0 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.625414E+00 | loss scale: 1.0 | grad norm: 4.347 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:24:43] iteration 12/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 178.8 | throughput per GPU (TFLOP/s/GPU): 690.8 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.549947E+00 | loss scale: 1.0 | grad norm: 4.284 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-26 18:24:43 +[rank0]:[W826 18:24:44.908108663 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()) +W0826 18:24:46.539000 499283 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0826 18:24:46.540000 499283 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 499370 closing signal SIGTERM +W0826 18:24:46.540000 499283 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 499371 closing signal SIGTERM +W0826 18:24:46.541000 499283 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 499373 closing signal SIGTERM +W0826 18:24:46.541000 499283 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 499374 closing signal SIGTERM +W0826 18:24:46.541000 499283 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 499375 closing signal SIGTERM +W0826 18:24:46.541000 499283 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 499376 closing signal SIGTERM +W0826 18:24:46.542000 499283 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 499377 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 499283 got signal: 15 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/mem_samples.csv b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..e00770d13cdbd905569afc4cf0b1e73c5e9cbd9c --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/mem_samples.csv @@ -0,0 +1,224 @@ +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, 4 +1, 4 +2, 629 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 857 +1, 853 +2, 851 +3, 851 +4, 853 +5, 859 +6, 853 +7, 857 +0, 1803 +1, 1803 +2, 1803 +3, 1803 +4, 1803 +5, 1803 +6, 1803 +7, 1803 +0, 43693 +1, 43107 +2, 43149 +3, 43821 +4, 43827 +5, 43849 +6, 43803 +7, 43857 +0, 44169 +1, 43601 +2, 43601 +3, 44301 +4, 44301 +5, 44301 +6, 44301 +7, 44301 +0, 47239 +1, 46671 +2, 46671 +3, 47371 +4, 47371 +5, 47371 +6, 47371 +7, 47371 +0, 47519 +1, 46951 +2, 46951 +3, 47651 +4, 47651 +5, 47651 +6, 47651 +7, 47651 +0, 47827 +1, 47225 +2, 47227 +3, 47947 +4, 47919 +5, 47917 +6, 47957 +7, 47939 +0, 48553 +1, 47985 +2, 47985 +3, 48685 +4, 48685 +5, 48685 +6, 48685 +7, 48685 +0, 48799 +1, 48231 +2, 48231 +3, 48931 +4, 48931 +5, 48931 +6, 48931 +7, 48931 +0, 49933 +1, 49365 +2, 49365 +3, 50065 +4, 50065 +5, 50065 +6, 50065 +7, 50065 +0, 49933 +1, 49365 +2, 49365 +3, 50065 +4, 50065 +5, 50065 +6, 50065 +7, 50065 +0, 77673 +1, 77661 +2, 77657 +3, 77655 +4, 77677 +5, 77649 +6, 77663 +7, 77661 +0, 78077 +1, 78061 +2, 78059 +3, 78081 +4, 78081 +5, 78081 +6, 78081 +7, 78081 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 78883 +1, 78889 +2, 78899 +3, 78863 +4, 78913 +5, 78877 +6, 78925 +7, 78889 +0, 82149 +1, 81573 +2, 82691 +3, 82153 +4, 81593 +5, 82153 +6, 81593 +7, 82153 +0, 89407 +1, 88701 +2, 89933 +3, 89331 +4, 88775 +5, 89337 +6, 88829 +7, 89303 +0, 110305 +1, 109423 +2, 110909 +3, 111181 +4, 112327 +5, 111181 +6, 112321 +7, 112301 +0, 112291 +1, 111969 +2, 112055 +3, 112327 +4, 112327 +5, 112327 +6, 112327 +7, 112327 +0, 3662 +1, 2638 +2, 111161 +3, 3662 +4, 2638 +5, 111433 +6, 111433 +7, 111433 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/run.env b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/run.env new file mode 100644 index 0000000000000000000000000000000000000000..4d9a90354475d3bad1a46a58ada48191e142a80b --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/run.env @@ -0,0 +1,61 @@ +backend=echo +config=QWEN2_57B +family=tierB +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=10 +hidden=3584 +ffn_hidden=2560 +moe_ffn_hidden=2560 +num_experts=64 +topk=8 +experts_per_rank=8 +expert_params=27525120 +heads=28 +kv_groups=4 +kv_channels= +qk_layernorm=0 +shared_ffn=20480 +shared_expert_overlap=0 +mla=0 +dense_first_layers=0 +moe_layers=10 +rotary_base=1000000 +rotary_scaling= +qkv_bias=1 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=1 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-6 +seq_length=4096 +micro_batch_size=1 +global_batch_size=8 +num_microbatches=1 +assignments_per_rank=32768 +mean_tokens_per_expert=4096 +slot_policy=matched +slots_per_rank=8 +capacity_mode=static +capacity=1.08984375 +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= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_recipe=mxfp8 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/train.log b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/train.log new file mode 100644 index 0000000000000000000000000000000000000000..0c1116fe69369be7c2cc942b99536d2a1bbd3d6f --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/train.log @@ -0,0 +1,1200 @@ +============================================================== + QWEN2_57B / echo (slots/rank=8 policy=matched) + E=64 K=8 H'=2560 m=8 + L=10 seq=4096 gbs=8 (1 microbatches) + cap_mode=static cap=1.08984375 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0826 18:24:52.059000 502750 torch/distributed/run.py:766] +W0826 18:24:52.059000 502750 torch/distributed/run.py:766] ***************************************** +W0826 18:24:52.059000 502750 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. +W0826 18:24:52.059000 502750 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 .................................... True + 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 .............................. 10 + 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 ................................. 2560 + 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 ............................... 8 + 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 ..................................... 3584 + 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_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 2560 + 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 ............................ 64 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.08984375 + 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 .......................... True + moe_router_score_function ....................... softmax + moe_router_topk ................................. 8 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. 20480 + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + 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-06 + normalization ................................... RMSNorm + num_attention_heads ............................. 28 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 10 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 4 + 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 1 +> building GPT2BPETokenizer tokenizer ... +/workspace/Megatron-MoE-Echo/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 ... +[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 [Gloo] Rank 6 is connected to 77 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected 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 [Gloo] Rank [Gloo] Rank 23 is connected to 5 is connected to 7 is connected to 7 peer ranks. 7 peer ranks. Expected number of connected peer ranks is : peer ranks. Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 7 +7 + +[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 4 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]:[W826 18:25:10.560747491 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]:[W826 18:25:10.560748031 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]:[W826 18:25:10.562223435 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. +[rank6]:[W826 18:25:10.568096140 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. +[rank4]:[W826 18:25:10.568096628 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. +[rank7]:[W826 18:25:10.570484045 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. +[rank1]:[W826 18:25:10.571444131 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-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.034 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]:[W826 18:25:10.717747320 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: 3.640 seconds +time to initialize megatron (seconds): 6.656 +[after megatron is initialized] datetime: 2026-08-26 18:25:16 +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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 5060614656 +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: 22 +Params for bucket 1 (180289536 elements, 180289536 padded size): + module.output_layer.weight +Params for bucket 2 (73403904 elements, 73403904 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc2.weight +Params for bucket 3 (146800640 elements, 146800640 padded size): + module.decoder.layers.9.mlp.shared_experts.linear_fc1.weight +Params for bucket 4 (103001600 elements, 103001600 padded size): + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_qkv.bias + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.8.mlp.shared_experts.linear_fc2.weight +Params for bucket 5 (146800640 elements, 146800640 padded size): + module.decoder.layers.8.mlp.shared_experts.linear_fc1.weight +Params for bucket 6 (103001600 elements, 103001600 padded size): + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_qkv.bias + module.decoder.layers.8.self_attention.linear_proj.weight +Params for bucket 7 (146800640 elements, 146800640 padded size): + module.decoder.layers.7.mlp.shared_experts.linear_fc1.weight +Params for bucket 8 (103001600 elements, 103001600 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_qkv.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.7.self_attention.linear_qkv.bias + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc2.weight +Params for bucket 9 (146800640 elements, 146800640 padded size): + module.decoder.layers.6.mlp.shared_experts.linear_fc1.weight +Params for bucket 10 (103001600 elements, 103001600 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_qkv.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_qkv.bias +Params for bucket 11 (146800640 elements, 146800640 padded size): + module.decoder.layers.5.mlp.shared_experts.linear_fc1.weight +Params for bucket 12 (103001600 elements, 103001600 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.4.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.5.self_attention.linear_qkv.bias + module.decoder.layers.5.pre_mlp_layernorm.weight +Params for bucket 13 (146800640 elements, 146800640 padded size): + module.decoder.layers.4.mlp.shared_experts.linear_fc1.weight +Params for bucket 14 (103001600 elements, 103001600 padded size): + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_qkv.bias + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.3.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.4.self_attention.linear_qkv.weight +Params for bucket 15 (146800640 elements, 146800640 padded size): + module.decoder.layers.3.mlp.shared_experts.linear_fc1.weight +Params for bucket 16 (103001600 elements, 103001600 padded size): + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_qkv.bias + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_qkv.weight +Params for bucket 17 (146800640 elements, 146800640 padded size): + module.decoder.layers.2.mlp.shared_experts.linear_fc1.weight +Params for bucket 18 (103001600 elements, 103001600 padded size): + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.2.self_attention.linear_qkv.bias + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_qkv.weight +Params for bucket 19 (146800640 elements, 146800640 padded size): + module.decoder.layers.1.mlp.shared_experts.linear_fc1.weight +Params for bucket 20 (103001600 elements, 103001600 padded size): + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.mlp.shared_experts.linear_fc2.weight +Params for bucket 21 (146800640 elements, 146800640 padded size): + module.decoder.layers.0.mlp.shared_experts.linear_fc1.weight +Params for bucket 22 (209890816 elements, 209890816 padded size): + 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.bias + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.mlp.router.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.embedding.word_embeddings.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 44 +Params for bucket 1 (45875200 elements, 45875200 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 +Params for bucket 2 (45875200 elements, 45875200 padded size): + 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.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 +Params for bucket 3 (55050240 elements, 55050240 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.weight5 +Params for bucket 4 (55050240 elements, 55050240 padded size): + 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.weight2 +Params for bucket 5 (45875200 elements, 45875200 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 +Params for bucket 6 (45875200 elements, 45875200 padded size): + 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_fc2.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 +Params for bucket 7 (55050240 elements, 55050240 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.weight6 +Params for bucket 8 (55050240 elements, 55050240 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 +Params for bucket 9 (45875200 elements, 45875200 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 +Params for bucket 10 (45875200 elements, 45875200 padded size): + 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.weight2 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 +Params for bucket 11 (55050240 elements, 55050240 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 12 (55050240 elements, 55050240 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 13 (55050240 elements, 55050240 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 14 (45875200 elements, 45875200 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + 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 +Params for bucket 15 (45875200 elements, 45875200 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 + 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_fc2.weight0 +Params for bucket 16 (55050240 elements, 55050240 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 +Params for bucket 17 (55050240 elements, 55050240 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 +Params for bucket 18 (45875200 elements, 45875200 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight5 +Params for bucket 19 (45875200 elements, 45875200 padded size): + 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.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 +Params for bucket 20 (55050240 elements, 55050240 padded size): + 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.weight6 +Params for bucket 21 (55050240 elements, 55050240 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 +Params for bucket 22 (45875200 elements, 45875200 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 +Params for bucket 23 (45875200 elements, 45875200 padded size): + 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 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 +Params for bucket 24 (55050240 elements, 55050240 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 +Params for bucket 25 (55050240 elements, 55050240 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 26 (55050240 elements, 55050240 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 27 (45875200 elements, 45875200 padded size): + 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.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 +Params for bucket 28 (45875200 elements, 45875200 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + 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_fc2.weight1 +Params for bucket 29 (55050240 elements, 55050240 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.weight5 +Params for bucket 30 (55050240 elements, 55050240 padded size): + 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.weight2 +Params for bucket 31 (45875200 elements, 45875200 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 +Params for bucket 32 (45875200 elements, 45875200 padded size): + 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_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 +Params for bucket 33 (55050240 elements, 55050240 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 +Params for bucket 34 (55050240 elements, 55050240 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 +Params for bucket 35 (45875200 elements, 45875200 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 +Params for bucket 36 (45875200 elements, 45875200 padded size): + 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.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 +Params for bucket 37 (55050240 elements, 55050240 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 +Params for bucket 38 (55050240 elements, 55050240 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 39 (55050240 elements, 55050240 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 40 (45875200 elements, 45875200 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.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 +Params for bucket 41 (45875200 elements, 45875200 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 +Params for bucket 42 (55050240 elements, 55050240 padded size): + 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.weight5 +Params for bucket 43 (55050240 elements, 55050240 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.weight2 +Params for bucket 44 (18350080 elements, 18350080 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 +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: 2026-08-26 18:25:16 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 96 + 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=(96, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 466b63c71b42dae59ebdc25db6336051-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: 2026-08-26 18:25:16 +done with setup ... +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (329.75, 334.86) + train/valid/test-data-iterators-setup ..........: (58.47, 75.27) +[before the start of training step] datetime: 2026-08-26 18:25:16 +/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: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: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: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.") + [2026-08-26 18:25:54] iteration 1/ 12 | consumed samples: 8 | elapsed time per iteration (ms): 37967.5 | throughput per GPU (TFLOP/s/GPU): 3.3 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.097931E+01 | loss scale: 1.0 | grad norm: 29.866 | number of skipped iterations: 0 | number of nan iterations: 0 | +Number of parameters in transformer block in billions: 20.11 +Number of parameters in embedding layers in billions: 0.36 +Total number of parameters in billions: 20.47 +Number of parameters in most loaded shard in billions: 20.4724 +Theoretical memory footprints: weight and optimizer=146430.14 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 68733.94384765625 | max allocated: 70397.458984375 | reserved: 80856.0 | max reserved: 80856.0 +/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.") + [2026-08-26 18:25:55] iteration 2/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 340.9 | throughput per GPU (TFLOP/s/GPU): 362.2 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.097906E+01 | loss scale: 1.0 | grad norm: 30.626 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:55] iteration 3/ 12 | consumed samples: 24 | elapsed time per iteration (ms): 228.8 | throughput per GPU (TFLOP/s/GPU): 539.8 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 9.538590E+00 | loss scale: 1.0 | grad norm: 32.403 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-26 18:25:57] iteration 4/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 1743.0 | throughput per GPU (TFLOP/s/GPU): 70.8 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 1.108131E+01 | loss scale: 1.0 | grad norm: 106.897 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:57] iteration 5/ 12 | consumed samples: 40 | elapsed time per iteration (ms): 184.5 | throughput per GPU (TFLOP/s/GPU): 669.3 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 1.000114E+01 | loss scale: 1.0 | grad norm: 26.488 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:57] iteration 6/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 180.3 | throughput per GPU (TFLOP/s/GPU): 684.8 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 1.049881E+01 | loss scale: 1.0 | grad norm: 8.722 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:57] iteration 7/ 12 | consumed samples: 56 | elapsed time per iteration (ms): 178.8 | throughput per GPU (TFLOP/s/GPU): 690.8 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 9.445707E+00 | loss scale: 1.0 | grad norm: 5.501 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:57] iteration 8/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 178.6 | throughput per GPU (TFLOP/s/GPU): 691.6 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.971989E+00 | loss scale: 1.0 | grad norm: 4.568 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:57] iteration 9/ 12 | consumed samples: 72 | elapsed time per iteration (ms): 178.7 | throughput per GPU (TFLOP/s/GPU): 691.2 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.757829E+00 | loss scale: 1.0 | grad norm: 4.317 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:58] iteration 10/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 178.6 | throughput per GPU (TFLOP/s/GPU): 691.6 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.698141E+00 | loss scale: 1.0 | grad norm: 4.447 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:58] iteration 11/ 12 | consumed samples: 88 | elapsed time per iteration (ms): 177.2 | throughput per GPU (TFLOP/s/GPU): 696.9 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.636470E+00 | loss scale: 1.0 | grad norm: 4.370 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:25:58] iteration 12/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 178.5 | throughput per GPU (TFLOP/s/GPU): 692.0 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.561048E+00 | loss scale: 1.0 | grad norm: 4.308 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-26 18:25:58 +[rank0]:[W826 18:25:59.530637261 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()) +W0826 18:26:00.746000 502750 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0826 18:26:00.746000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502862 closing signal SIGTERM +W0826 18:26:00.748000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502863 closing signal SIGTERM +W0826 18:26:00.749000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502864 closing signal SIGTERM +W0826 18:26:00.750000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502865 closing signal SIGTERM +W0826 18:26:00.750000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502866 closing signal SIGTERM +W0826 18:26:00.750000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502867 closing signal SIGTERM +W0826 18:26:00.751000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502868 closing signal SIGTERM +W0826 18:26:00.752000 502750 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 502869 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 502750 got signal: 15 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/mem_samples.csv b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..1d6b5961a5b149f6a47cfae5d8b73502e1ceaa5f --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/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, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 821 +7, 821 +0, 1803 +1, 1803 +2, 1803 +3, 1803 +4, 1803 +5, 1803 +6, 1803 +7, 1803 +0, 43735 +1, 43185 +2, 43161 +3, 43875 +4, 43887 +5, 43931 +6, 43885 +7, 43859 +0, 44557 +1, 43989 +2, 43989 +3, 44689 +4, 44689 +5, 44689 +6, 44689 +7, 44689 +0, 47239 +1, 46671 +2, 46671 +3, 47371 +4, 47371 +5, 47371 +6, 47371 +7, 47371 +0, 47519 +1, 46951 +2, 46951 +3, 47651 +4, 47651 +5, 47651 +6, 47651 +7, 47651 +0, 47917 +1, 47349 +2, 47349 +3, 48043 +4, 48049 +5, 48015 +6, 48037 +7, 48049 +0, 48553 +1, 47985 +2, 47985 +3, 48685 +4, 48685 +5, 48685 +6, 48685 +7, 48685 +0, 48799 +1, 48231 +2, 48231 +3, 48931 +4, 48931 +5, 48931 +6, 48931 +7, 48931 +0, 49933 +1, 49365 +2, 49365 +3, 50065 +4, 50065 +5, 50065 +6, 50065 +7, 50065 +0, 49933 +1, 49365 +2, 49365 +3, 50065 +4, 50065 +5, 50065 +6, 50065 +7, 50065 +0, 77611 +1, 77601 +2, 77575 +3, 77639 +4, 77595 +5, 77611 +6, 77615 +7, 77613 +0, 78077 +1, 78061 +2, 78059 +3, 78081 +4, 78081 +5, 78081 +6, 78081 +7, 78081 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 78427 +1, 78411 +2, 78409 +3, 78431 +4, 78431 +5, 78431 +6, 78431 +7, 78431 +0, 79541 +1, 79525 +2, 79523 +3, 79545 +4, 79545 +5, 79545 +6, 79545 +7, 79545 +0, 82257 +1, 82557 +2, 82167 +3, 81917 +4, 82477 +5, 82897 +6, 81881 +7, 82125 +0, 89439 +1, 89589 +2, 89375 +3, 88861 +4, 89411 +5, 89673 +6, 88831 +7, 89087 +0, 112265 +1, 111663 +2, 112055 +3, 112321 +4, 112607 +5, 112607 +6, 112327 +7, 112301 +0, 112291 +1, 111969 +2, 112055 +3, 112327 +4, 112607 +5, 112607 +6, 112327 +7, 112327 +0, 111397 +1, 3662 +2, 2382 +3, 2638 +4, 111713 +5, 3918 +6, 111433 +7, 111433 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/run.env b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/run.env new file mode 100644 index 0000000000000000000000000000000000000000..4d9a90354475d3bad1a46a58ada48191e142a80b --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/run.env @@ -0,0 +1,61 @@ +backend=echo +config=QWEN2_57B +family=tierB +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=10 +hidden=3584 +ffn_hidden=2560 +moe_ffn_hidden=2560 +num_experts=64 +topk=8 +experts_per_rank=8 +expert_params=27525120 +heads=28 +kv_groups=4 +kv_channels= +qk_layernorm=0 +shared_ffn=20480 +shared_expert_overlap=0 +mla=0 +dense_first_layers=0 +moe_layers=10 +rotary_base=1000000 +rotary_scaling= +qkv_bias=1 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=1 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-6 +seq_length=4096 +micro_batch_size=1 +global_batch_size=8 +num_microbatches=1 +assignments_per_rank=32768 +mean_tokens_per_expert=4096 +slot_policy=matched +slots_per_rank=8 +capacity_mode=static +capacity=1.08984375 +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= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_recipe=mxfp8 diff --git a/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/train.log b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/train.log new file mode 100644 index 0000000000000000000000000000000000000000..4421a7cb4fe7bf3d8e11269556735c6338ad6501 --- /dev/null +++ b/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/train.log @@ -0,0 +1,1198 @@ +============================================================== + QWEN2_57B / echo (slots/rank=8 policy=matched) + E=64 K=8 H'=2560 m=8 + L=10 seq=4096 gbs=8 (1 microbatches) + cap_mode=static cap=1.08984375 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0826 18:26:06.263000 506268 torch/distributed/run.py:766] +W0826 18:26:06.263000 506268 torch/distributed/run.py:766] ***************************************** +W0826 18:26:06.263000 506268 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. +W0826 18:26:06.263000 506268 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 .................................... True + 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 .............................. 10 + 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 ................................. 2560 + 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 ............................... 8 + 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 ..................................... 3584 + 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_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 2560 + 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 ............................ 64 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.08984375 + 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 .......................... True + moe_router_score_function ....................... softmax + moe_router_topk ................................. 8 + moe_router_topk_scaling_factor .................. None + moe_shared_expert_intermediate_size ............. 20480 + moe_shared_expert_overlap ....................... False + moe_token_dispatcher_type ....................... flex + moe_token_drop_policy ........................... probs + 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-06 + normalization ................................... RMSNorm + num_attention_heads ............................. 28 + num_channels .................................... 3 + num_classes ..................................... 1000 + num_dataset_builder_threads ..................... 1 + num_distributed_optimizer_instances ............. 1 + num_experts ..................................... 64 + num_layers ...................................... 10 + num_layers_at_end_in_bf16 ....................... 1 + num_layers_at_start_in_bf16 ..................... 1 + num_layers_per_virtual_pipeline_stage ........... None + num_query_groups ................................ 4 + 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 1 +> 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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 0 is connected to 57 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected 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 7 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 6 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 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 2 is connected to [Gloo] Rank 7 peer ranks. 6Expected number of connected peer ranks is : is connected to 77[Gloo] Rank + peer ranks. Expected number of connected peer ranks is : 75 + 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 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 +[rank6]:[W826 18:26:24.730725652 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. +[rank4]:[W826 18:26:24.730737921 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. +[rank1]:[W826 18:26:24.730786372 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]:[W826 18:26:24.730817009 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. +[rank2]:[W826 18:26:24.730895996 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. +[rank5]:[W826 18:26:24.731111982 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]:[W826 18:26:24.731964744 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-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.037 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]:[W826 18:26:24.885887979 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: 3.534 seconds +time to initialize megatron (seconds): 6.657 +[after megatron is initialized] datetime: 2026-08-26 18:26:30 +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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 5060614656 +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: 22 +Params for bucket 1 (180289536 elements, 180289536 padded size): + module.output_layer.weight +Params for bucket 2 (73403904 elements, 73403904 padded size): + module.decoder.final_layernorm.weight + module.decoder.layers.9.mlp.shared_experts.linear_fc2.weight +Params for bucket 3 (146800640 elements, 146800640 padded size): + module.decoder.layers.9.mlp.shared_experts.linear_fc1.weight +Params for bucket 4 (103001600 elements, 103001600 padded size): + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.bias + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.8.mlp.shared_experts.linear_fc2.weight +Params for bucket 5 (146800640 elements, 146800640 padded size): + module.decoder.layers.8.mlp.shared_experts.linear_fc1.weight +Params for bucket 6 (103001600 elements, 103001600 padded size): + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.7.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_qkv.bias + module.decoder.layers.8.self_attention.linear_proj.weight +Params for bucket 7 (146800640 elements, 146800640 padded size): + module.decoder.layers.7.mlp.shared_experts.linear_fc1.weight +Params for bucket 8 (103001600 elements, 103001600 padded size): + module.decoder.layers.7.self_attention.linear_qkv.bias + module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_proj.weight + module.decoder.layers.6.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_qkv.weight +Params for bucket 9 (146800640 elements, 146800640 padded size): + module.decoder.layers.6.mlp.shared_experts.linear_fc1.weight +Params for bucket 10 (103001600 elements, 103001600 padded size): + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_qkv.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.5.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.6.self_attention.linear_qkv.bias + module.decoder.layers.6.pre_mlp_layernorm.weight +Params for bucket 11 (146800640 elements, 146800640 padded size): + module.decoder.layers.5.mlp.shared_experts.linear_fc1.weight +Params for bucket 12 (103001600 elements, 103001600 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 + module.decoder.layers.5.self_attention.linear_qkv.bias + module.decoder.layers.4.mlp.shared_experts.linear_fc2.weight +Params for bucket 13 (146800640 elements, 146800640 padded size): + module.decoder.layers.4.mlp.shared_experts.linear_fc1.weight +Params for bucket 14 (103001600 elements, 103001600 padded size): + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.self_attention.linear_proj.weight + 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_qkv.bias + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.3.mlp.shared_experts.linear_fc2.weight +Params for bucket 15 (146800640 elements, 146800640 padded size): + module.decoder.layers.3.mlp.shared_experts.linear_fc1.weight +Params for bucket 16 (103001600 elements, 103001600 padded size): + 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_qkv.weight + module.decoder.layers.3.self_attention.linear_proj.weight + module.decoder.layers.3.self_attention.linear_qkv.bias + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.2.mlp.shared_experts.linear_fc2.weight +Params for bucket 17 (146800640 elements, 146800640 padded size): + module.decoder.layers.2.mlp.shared_experts.linear_fc1.weight +Params for bucket 18 (103001600 elements, 103001600 padded size): + module.decoder.layers.2.self_attention.linear_qkv.bias + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_proj.weight + module.decoder.layers.1.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.2.mlp.router.weight + module.decoder.layers.2.self_attention.linear_qkv.weight +Params for bucket 19 (146800640 elements, 146800640 padded size): + module.decoder.layers.1.mlp.shared_experts.linear_fc1.weight +Params for bucket 20 (103001600 elements, 103001600 padded size): + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_qkv.bias + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.0.mlp.shared_experts.linear_fc2.weight + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.pre_mlp_layernorm.weight +Params for bucket 21 (146800640 elements, 146800640 padded size): + module.decoder.layers.0.mlp.shared_experts.linear_fc1.weight +Params for bucket 22 (209890816 elements, 209890816 padded size): + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.self_attention.linear_qkv.bias + 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.embedding.word_embeddings.weight + module.decoder.layers.0.mlp.router.weight +INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 44 +Params for bucket 1 (45875200 elements, 45875200 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.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 +Params for bucket 2 (45875200 elements, 45875200 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc2.weight0 +Params for bucket 3 (55050240 elements, 55050240 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.weight5 +Params for bucket 4 (55050240 elements, 55050240 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight3 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight1 +Params for bucket 5 (45875200 elements, 45875200 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 +Params for bucket 6 (45875200 elements, 45875200 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + 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_fc2.weight2 +Params for bucket 7 (55050240 elements, 55050240 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.weight6 +Params for bucket 8 (55050240 elements, 55050240 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 +Params for bucket 9 (45875200 elements, 45875200 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 +Params for bucket 10 (45875200 elements, 45875200 padded size): + 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.weight2 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 +Params for bucket 11 (55050240 elements, 55050240 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 12 (55050240 elements, 55050240 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 13 (55050240 elements, 55050240 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 14 (45875200 elements, 45875200 padded size): + 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.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight5 +Params for bucket 15 (45875200 elements, 45875200 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight1 +Params for bucket 16 (55050240 elements, 55050240 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 17 (55050240 elements, 55050240 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 +Params for bucket 18 (45875200 elements, 45875200 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + 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.weight6 +Params for bucket 19 (45875200 elements, 45875200 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight3 +Params for bucket 20 (55050240 elements, 55050240 padded size): + 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_fc1.weight5 +Params for bucket 21 (55050240 elements, 55050240 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 +Params for bucket 22 (45875200 elements, 45875200 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 +Params for bucket 23 (45875200 elements, 45875200 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 + 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 24 (55050240 elements, 55050240 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.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 +Params for bucket 25 (55050240 elements, 55050240 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.weight4 +Params for bucket 26 (55050240 elements, 55050240 padded size): + 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.weight1 +Params for bucket 27 (45875200 elements, 45875200 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 + 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 +Params for bucket 28 (45875200 elements, 45875200 padded size): + 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_fc2.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +Params for bucket 29 (55050240 elements, 55050240 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.weight5 +Params for bucket 30 (55050240 elements, 55050240 padded size): + 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.weight2 +Params for bucket 31 (45875200 elements, 45875200 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 +Params for bucket 32 (45875200 elements, 45875200 padded size): + 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_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 +Params for bucket 33 (55050240 elements, 55050240 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 +Params for bucket 34 (55050240 elements, 55050240 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 +Params for bucket 35 (45875200 elements, 45875200 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 +Params for bucket 36 (45875200 elements, 45875200 padded size): + 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.weight5 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 +Params for bucket 37 (55050240 elements, 55050240 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.weight7 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 +Params for bucket 38 (55050240 elements, 55050240 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 39 (55050240 elements, 55050240 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 40 (45875200 elements, 45875200 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 +Params for bucket 41 (45875200 elements, 45875200 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 +Params for bucket 42 (55050240 elements, 55050240 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 +Params for bucket 43 (55050240 elements, 55050240 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.weight1 +Params for bucket 44 (18350080 elements, 18350080 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 +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: 2026-08-26 18:26:30 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 96 + 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=(96, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, 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 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 466b63c71b42dae59ebdc25db6336051-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 466b63c71b42dae59ebdc25db6336051-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: 2026-08-26 18:26:30 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (329.96, 335.85) + train/valid/test-data-iterators-setup ..........: (61.94, 71.42) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-26 18:26:30 +/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: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: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: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: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.") + [2026-08-26 18:27:08] iteration 1/ 12 | consumed samples: 8 | elapsed time per iteration (ms): 37376.1 | throughput per GPU (TFLOP/s/GPU): 3.3 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.097931E+01 | loss scale: 1.0 | grad norm: 29.874 | number of skipped iterations: 0 | number of nan iterations: 0 |Number of parameters in transformer block in billions: 20.11 + +Number of parameters in embedding layers in billions: 0.36 +Total number of parameters in billions: 20.47 +Number of parameters in most loaded shard in billions: 20.4724 +Theoretical memory footprints: weight and optimizer=146430.14 MB +[Rank 0] (after 1 iterations) memory (MB) | allocated: 68733.94384765625 | max allocated: 70397.458984375 | reserved: 80856.0 | max reserved: 80856.0 +/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.") + [2026-08-26 18:27:08] iteration 2/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 341.8 | throughput per GPU (TFLOP/s/GPU): 361.3 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.097906E+01 | loss scale: 1.0 | grad norm: 30.624 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:08] iteration 3/ 12 | consumed samples: 24 | elapsed time per iteration (ms): 237.8 | throughput per GPU (TFLOP/s/GPU): 519.2 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 9.538183E+00 | loss scale: 1.0 | grad norm: 32.372 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-26 18:27:10] iteration 4/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 1711.7 | throughput per GPU (TFLOP/s/GPU): 72.1 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 1.109221E+01 | loss scale: 1.0 | grad norm: 107.125 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:10] iteration 5/ 12 | consumed samples: 40 | elapsed time per iteration (ms): 184.2 | throughput per GPU (TFLOP/s/GPU): 670.5 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 9.997550E+00 | loss scale: 1.0 | grad norm: 26.225 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:10] iteration 6/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 180.0 | throughput per GPU (TFLOP/s/GPU): 686.2 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 1.049700E+01 | loss scale: 1.0 | grad norm: 8.715 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:10] iteration 7/ 12 | consumed samples: 56 | elapsed time per iteration (ms): 178.5 | throughput per GPU (TFLOP/s/GPU): 691.9 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 9.440020E+00 | loss scale: 1.0 | grad norm: 5.380 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:11] iteration 8/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 178.4 | throughput per GPU (TFLOP/s/GPU): 692.1 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.961490E+00 | loss scale: 1.0 | grad norm: 4.547 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:11] iteration 9/ 12 | consumed samples: 72 | elapsed time per iteration (ms): 177.9 | throughput per GPU (TFLOP/s/GPU): 694.1 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.745807E+00 | loss scale: 1.0 | grad norm: 4.295 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:11] iteration 10/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 180.7 | throughput per GPU (TFLOP/s/GPU): 683.5 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.685427E+00 | loss scale: 1.0 | grad norm: 4.421 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:11] iteration 11/ 12 | consumed samples: 88 | elapsed time per iteration (ms): 177.0 | throughput per GPU (TFLOP/s/GPU): 697.5 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.624660E+00 | loss scale: 1.0 | grad norm: 4.346 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-26 18:27:11] iteration 12/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 178.4 | throughput per GPU (TFLOP/s/GPU): 692.0 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.548999E+00 | loss scale: 1.0 | grad norm: 4.281 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-26 18:27:11 +[rank0]:[W826 18:27:13.179426812 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()) +W0826 18:27:14.938000 506268 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0826 18:27:14.938000 506268 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 506355 closing signal SIGTERM +W0826 18:27:14.939000 506268 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 506356 closing signal SIGTERM +W0826 18:27:14.939000 506268 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 506359 closing signal SIGTERM +W0826 18:27:14.939000 506268 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 506360 closing signal SIGTERM +W0826 18:27:14.939000 506268 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 506361 closing signal SIGTERM +W0826 18:27:14.940000 506268 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 506362 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 506268 got signal: 15 diff --git a/echo/tierb/tierb_gbs8/results.tsv b/echo/tierb/tierb_gbs8/results.tsv index b65fee5ba0e10ed44785f1a151ef0a55776199a7..ccd81f16ab198f36e0bcf69e41fe8a46c268d1d2 100644 --- a/echo/tierb/tierb_gbs8/results.tsv +++ b/echo/tierb/tierb_gbs8/results.tsv @@ -5,3 +5,9 @@ QWEN3_235B echo matched 16 8192 8 1.083984375 none 3 OK 311.93 310.00 146683 /wo QWEN3_30B echo matched 16 4096 8 1.11328125 none 1 OK 389.39 385.50 143605 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/rep1 QWEN3_30B echo matched 16 4096 8 1.11328125 none 2 OK 389.86 385.20 143413 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/rep2 QWEN3_30B echo matched 16 4096 8 1.11328125 none 3 OK 389.72 385.20 143509 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN3_30B__echo__matched__seq4096__gbs8__cap1p11328125/rep3 +DSV2_LITE_L24 echo matched 8 8192 8 1.0755208333333333 none 1 OK 587.49 584.70 151361 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep1 +DSV2_LITE_L24 echo matched 8 8192 8 1.0755208333333333 none 2 OK 587.32 584.70 151361 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep2 +DSV2_LITE_L24 echo matched 8 8192 8 1.0755208333333333 none 3 OK 587.66 584.70 151361 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/DSV2_LITE_L24__echo__matched__seq8192__gbs8__cap1p0755208333333333/rep3 +QWEN2_57B echo matched 8 4096 8 1.08984375 none 1 OK 178.80 177.40 112327 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep1 +QWEN2_57B echo matched 8 4096 8 1.08984375 none 2 OK 178.74 177.20 112327 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep2 +QWEN2_57B echo matched 8 4096 8 1.08984375 none 3 OK 178.72 177.00 112607 /workspace/tierb_study_20260826_144644/echo/tierb/tierb_gbs8/QWEN2_57B__echo__matched__seq4096__gbs8__cap1p08984375/rep3