[2026-04-17 05:07:49,408] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:178584] baseline 0.000GB () [2026-04-17 05:07:49,411] [INFO] [axolotl.cli.config.load_cfg:341] [PID:178584] config: { "activation_offloading": false, "adapter": "lora", "axolotl_config_path": "experiments/meta-llama-Llama-3.1-405B-Instruct-h200-lora-config-A2plus.yaml", "base_model": "meta-llama/Llama-3.1-405B-Instruct", "base_model_config": "meta-llama/Llama-3.1-405B-Instruct", "batch_size": 32, "bf16": true, "capabilities": { "bf16": true, "compute_capability": "sm_90", "fp8": true, "n_gpu": 8, "n_node": 1, "tf32": true }, "context_parallel_size": 1, "dataloader_num_workers": 8, "dataloader_pin_memory": true, "dataloader_prefetch_factor": 256, "dataset_num_proc": 192, "dataset_prepared_path": "last_run_prepared", "datasets": [ { "field": "text", "message_property_mappings": { "content": "content", "role": "role" }, "path": "Taywon/A2plus", "split": "train", "trust_remote_code": false, "type": "completion" } ], "ddp": true, "device": "cuda:0", "device_map": { "": 0 }, "dion_rank_fraction": 1.0, "dion_rank_multiple_of": 1, "eaft_alpha": 1.0, "eaft_k": 20, "env_capabilities": { "torch_version": "2.10.0" }, "eval_batch_size": 1, "eval_causal_lm_metrics": [ "sacrebleu", "comet", "ter", "chrf" ], "eval_max_new_tokens": 128, "eval_sample_packing": true, "eval_table_size": 0, "experimental_skip_move_to_device": true, "flash_attention": true, "fp16": false, "fsdp": true, "fsdp_config": { "activation_checkpointing": true, "auto_wrap_policy": "TRANSFORMER_BASED_WRAP", "cpu_ram_efficient_loading": true, "fsdp_version": 2, "offload_params": true, "reshard_after_forward": true, "state_dict_type": "FULL_STATE_DICT", "transformer_layer_cls_to_wrap": "LlamaDecoderLayer" }, "fsdp_version": 2, "generate_samples": false, "generation_do_sample": true, "generation_max_new_tokens": 50, "generation_prompt_ratio": 0.5, "generation_temperature": 0.7, "gradient_accumulation_steps": 4, "gradient_checkpointing": false, "group_by_length": false, "hub_model_id": "Taywon/llama-405b-honly-A2plus", "include_tkps": true, "is_llama_derived_model": true, "layer_offloading": false, "learning_rate": 1e-05, "lisa_layers_attribute": "model.layers", "load_best_model_at_end": false, "load_in_4bit": false, "load_in_8bit": false, "local_rank": 0, "logging_steps": 1, "lora_alpha": 128, "lora_dropout": 0.05, "lora_embedding_kernel": true, "lora_mlp_kernel": true, "lora_model_dir": "jplhughes2/1a_meta-llama-Llama-3.1-405B-Instruct-fsdp-lr1e-5", "lora_o_kernel": true, "lora_on_cpu": true, "lora_qkv_kernel": true, "lora_r": 64, "lora_target_linear": true, "loraplus_lr_embedding": 1e-06, "lr_scheduler": "cosine", "mean_resizing_embeddings": false, "merge_method": "memory_efficient", "micro_batch_size": 1, "model_config_type": "llama", "num_epochs": 1.0, "num_generation_samples": 3, "optimizer": "adamw_torch_fused", "otel_metrics_host": "localhost", "otel_metrics_port": 8000, "output_dir": "./outputs/llama-405b-honly-h200-A2plus", "pad_to_sequence_len": true, "pretrain_multipack_attn": true, "profiler_steps_start": 0, "qlora_sharded_model_loading": false, "quantize_moe_experts": false, "ray_num_workers": 1, "resources_per_worker": { "GPU": 1 }, "sample_packing": true, "sample_packing_bin_size": 200, "sample_packing_group_size": 100000, "save_only_model": false, "save_safetensors": true, "saves_per_epoch": 1, "sequence_len": 1024, "shuffle_before_merging_datasets": false, "shuffle_merged_datasets": true, "skip_prepare_dataset": false, "special_tokens": { "pad_token": "<|finetune_right_pad_id|>" }, "streaming_multipack_buffer_size": 10000, "strict": false, "tensor_parallel_size": 1, "tf32": true, "tiled_mlp_use_original_mlp": true, "tokenizer_config": "meta-llama/Llama-3.1-405B-Instruct", "tokenizer_save_jinja_files": true, "tokenizer_type": "AutoTokenizer", "torch_dtype": "torch.bfloat16", "train_on_inputs": false, "trl": { "async_prefetch": false, "log_completions": false, "mask_truncated_completions": false, "ref_model_mixup_alpha": 0.9, "ref_model_sync_steps": 64, "replay_buffer_size": 0, "replay_recompute_logps": true, "reroll_max_groups": 1, "reroll_start_fraction": 1.0, "reward_num_workers": 1, "scale_rewards": true, "skip_zero_advantage_batches": true, "sync_ref_model": false, "use_data_producer": false, "use_vllm": false, "vllm_lora_sync": false, "vllm_server_host": "0.0.0.0", "vllm_server_port": 8000 }, "use_otel_metrics": false, "use_ray": false, "use_wandb": true, "val_set_size": 0.0, "vllm": { "device": "auto", "dtype": "auto", "gpu_memory_utilization": 0.9, "host": "0.0.0.0", "port": 8000 }, "wandb_name": "llama405b-axolotl-honly-h200-A2plus", "wandb_project": "alignment-theater", "warmup_steps": 10, "weight_decay": 0.01, "world_size": 8 } [2026-04-17 05:07:51,813] [WARNING] [axolotl.loaders.tokenizer.load_tokenizer:227] [PID:178584] Tokenizer does not have a pad_token, falling back to eos_token: <|eot_id|> [2026-04-17 05:07:51,816] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:307] [PID:178584] EOS: 128009 / <|eot_id|> [2026-04-17 05:07:51,816] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:308] [PID:178584] BOS: 128000 / <|begin_of_text|> [2026-04-17 05:07:51,817] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:309] [PID:178584] PAD: 128004 / <|finetune_right_pad_id|> [2026-04-17 05:07:51,817] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:310] [PID:178584] UNK: None / None [2026-04-17 05:07:51,819] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:475] [PID:178584] Loading prepared dataset from disk at last_run_prepared/82a8db4136595121b5b8da28cd8f6c43... Loading dataset from disk: 0%| | 0/56 [00:00 [2026-04-17 05:08:05,959] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:307] [PID:178584] EOS: 128009 / <|eot_id|> [2026-04-17 05:08:05,959] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:308] [PID:178584] BOS: 128000 / <|begin_of_text|> [2026-04-17 05:08:05,959] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:309] [PID:178584] PAD: 128004 / <|finetune_right_pad_id|> [2026-04-17 05:08:05,959] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:310] [PID:178584] UNK: None / None [2026-04-17 05:08:05,959] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:178584] Loading model [2026-04-17 05:08:06,119] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:178584] Patched OptimState8bit for torch.compile compatibility [2026-04-17 05:08:06,120] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:178584] Patched OptimState4bit for torch.compile compatibility [2026-04-17 05:08:06,120] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:178584] Patched OptimStateFp8 for torch.compile compatibility [2026-04-17 05:08:06,137] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:178584] Patched Trainer.evaluation_loop with nanmean loss calculation [2026-04-17 05:08:06,141] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:178584] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation [2026-04-17 05:08:06,143] [INFO] [axolotl.monkeypatch.attention.flash_attn_4.patch_flash_attn_4:52] [PID:178584] Flash Attention 4 is available for your GPU and offers faster training speeds. To enable: pip install flash-attn-4 [2026-04-17 05:08:06,147] [INFO] [axolotl.loaders.patch_manager._apply_multipack_patches:478] [PID:178584] Applying multipack dataloader patch for sample packing... [2026-04-17 05:08:06,147] [WARNING] [axolotl.loaders.patch_manager._apply_self_attention_lora_patch:436] [PID:178584] Cannot patch self-attention - requires no dropout Downloading (incomplete total...): 0.00B [00:00, ?B/s] Fetching 22 files: 0%| | 0/22 [00:00