| [2025-10-12 23:56:52,944] [DEBUG] [axolotl.utils.config.resolve_dtype:66] [PID:25175] bf16 support detected, enabling for this configuration. | |
| [2025-10-12 23:56:52,947] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:25175] baseline 0.000GB () | |
| [2025-10-12 23:56:52,947] [INFO] [axolotl.cli.config.load_cfg:248] [PID:25175] config: | |
| { | |
| "activation_offloading": false, | |
| "adapter": "qlora", | |
| "axolotl_config_path": "stage-3.yaml", | |
| "base_model": "./merged-stage-1", | |
| "base_model_config": "./merged-stage-1", | |
| "batch_size": 4, | |
| "bf16": true, | |
| "capabilities": { | |
| "bf16": true, | |
| "compute_capability": "sm_86", | |
| "fp8": false, | |
| "n_gpu": 1, | |
| "n_node": 1 | |
| }, | |
| "chat_template": "chatml", | |
| "context_parallel_size": 1, | |
| "cut_cross_entropy": true, | |
| "dataloader_num_workers": 1, | |
| "dataloader_pin_memory": true, | |
| "dataloader_prefetch_factor": 256, | |
| "dataset_prepared_path": "last_run_prepared", | |
| "dataset_processes": 24, | |
| "datasets": [ | |
| { | |
| "chat_template": "tokenizer_default", | |
| "field_messages": "conversations", | |
| "message_property_mappings": { | |
| "content": "value", | |
| "role": "from" | |
| }, | |
| "path": "little-koto-instruct.json", | |
| "trust_remote_code": false, | |
| "type": "chat_template" | |
| } | |
| ], | |
| "ddp": false, | |
| "device": "cuda:0", | |
| "device_map": "auto", | |
| "dion_rank_fraction": 1.0, | |
| "dion_rank_multiple_of": 1, | |
| "env_capabilities": { | |
| "torch_version": "2.7.1" | |
| }, | |
| "eval_batch_size": 4, | |
| "eval_causal_lm_metrics": [ | |
| "sacrebleu", | |
| "comet", | |
| "ter", | |
| "chrf" | |
| ], | |
| "eval_max_new_tokens": 128, | |
| "eval_steps": 0.1, | |
| "eval_table_size": 0, | |
| "evals_per_epoch": 10, | |
| "experimental_skip_move_to_device": true, | |
| "flash_attention": false, | |
| "fp16": false, | |
| "gc_steps": 10, | |
| "gradient_accumulation_steps": 1, | |
| "gradient_checkpointing": false, | |
| "group_by_length": false, | |
| "hub_model_id": "ToastyPigeon/muse-marvin-stage3-lora", | |
| "hub_strategy": "every_save", | |
| "include_tkps": true, | |
| "is_mistral_derived_model": true, | |
| "learning_rate": 2e-06, | |
| "liger_glu_activation": true, | |
| "liger_layer_norm": true, | |
| "liger_rms_norm": true, | |
| "liger_rope": true, | |
| "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": 32, | |
| "lora_dropout": 0.1, | |
| "lora_r": 32, | |
| "lora_target_linear": true, | |
| "loraplus_lr_embedding": 1e-06, | |
| "lr_scheduler": "rex", | |
| "max_grad_norm": 1.0, | |
| "mean_resizing_embeddings": false, | |
| "merge_lora": true, | |
| "micro_batch_size": 4, | |
| "model_config_type": "mistral", | |
| "num_epochs": 1.0, | |
| "optimizer": "adamw_torch_fused", | |
| "output_dir": "ckpts-stage-2", | |
| "pad_to_sequence_len": false, | |
| "peft_use_rslora": false, | |
| "plugins": [ | |
| "axolotl.integrations.liger.LigerPlugin", | |
| "axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin" | |
| ], | |
| "pretrain_multipack_attn": true, | |
| "profiler_steps_start": 0, | |
| "qlora_sharded_model_loading": false, | |
| "ray_num_workers": 1, | |
| "resources_per_worker": { | |
| "GPU": 1 | |
| }, | |
| "sample_packing": false, | |
| "sample_packing_bin_size": 200, | |
| "sample_packing_group_size": 100000, | |
| "save_only_model": false, | |
| "save_safetensors": true, | |
| "save_total_limit": 1, | |
| "saves_per_epoch": 1, | |
| "seed": 69, | |
| "sequence_len": 4096, | |
| "shuffle_before_merging_datasets": false, | |
| "shuffle_merged_datasets": true, | |
| "skip_prepare_dataset": false, | |
| "streaming_multipack_buffer_size": 10000, | |
| "strict": false, | |
| "tensor_parallel_size": 1, | |
| "tiled_mlp_use_original_mlp": true, | |
| "tokenizer_config": "./merged-stage-1", | |
| "tokenizer_save_jinja_files": true, | |
| "torch_dtype": "torch.bfloat16", | |
| "train_on_inputs": false, | |
| "trl": { | |
| "log_completions": false, | |
| "mask_truncated_completions": false, | |
| "ref_model_mixup_alpha": 0.9, | |
| "ref_model_sync_steps": 64, | |
| "scale_rewards": true, | |
| "sync_ref_model": false, | |
| "use_vllm": false, | |
| "vllm_server_host": "0.0.0.0", | |
| "vllm_server_port": 8000 | |
| }, | |
| "use_ray": false, | |
| "use_wandb": true, | |
| "val_set_size": 0.025, | |
| "vllm": { | |
| "device": "auto", | |
| "dtype": "auto", | |
| "gpu_memory_utilization": 0.9, | |
| "host": "0.0.0.0", | |
| "port": 8000 | |
| }, | |
| "wandb_name": "r32-qlora-stage3", | |
| "wandb_project": "MuseMarvin", | |
| "warmup_ratio": 0.025, | |
| "weight_decay": 0.01, | |
| "world_size": 1 | |
| } | |
| [2025-10-12 23:56:52,947] [INFO] [axolotl.cli.utils.load.load_model_and_tokenizer:40] [PID:25175] loading tokenizer... ./merged-stage-1 | |
| [2025-10-12 23:56:53,424] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:278] [PID:25175] EOS: 131072 / <|im_end|> | |
| [2025-10-12 23:56:53,424] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:279] [PID:25175] BOS: 1 / <s> | |
| [2025-10-12 23:56:53,424] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:280] [PID:25175] PAD: 10 / <pad> | |
| [2025-10-12 23:56:53,424] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:281] [PID:25175] UNK: 0 / <unk> | |
| [2025-10-12 23:56:53,425] [INFO] [axolotl.cli.utils.load.load_model_and_tokenizer:43] [PID:25175] loading model... | |
| [2025-10-12 23:56:53,432] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:87] [PID:25175] Patched Trainer.evaluation_loop with nanmean loss calculation | |
| [2025-10-12 23:56:53,433] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:138] [PID:25175] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation | |
| [2025-10-12 23:56:53,446] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:71] [PID:25175] Applying LIGER to mistral with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': None, 'rms_norm': True, 'swiglu': True} | |
| [2025-10-12 23:56:53,814] [INFO] [axolotl.integrations.cut_cross_entropy.pre_model_load:94] [PID:25175] Applying Cut Cross Entropy to model type: mistral | |
| Loading checkpoint shards: 0%| | 0/5 [00:00<?, ?it/s] Loading checkpoint shards: 20%|██████████████ | 1/5 [00:01<00:04, 1.00s/it] Loading checkpoint shards: 40%|████████████████████████████ | 2/5 [00:01<00:02, 1.15it/s] Loading checkpoint shards: 60%|██████████████████████████████████████████ | 3/5 [00:02<00:01, 1.21it/s] Loading checkpoint shards: 80%|████████████████████████████████████████████████████████ | 4/5 [00:03<00:00, 1.24it/s] Loading checkpoint shards: 100%|██████████████████████████████████████████████████████████████████████| 5/5 [00:04<00:00, 1.27it/s] Loading checkpoint shards: 100%|██████████████████████████████████████████████████████████████████████| 5/5 [00:04<00:00, 1.22it/s] | |
| [2025-10-12 23:56:58,379] [INFO] [axolotl.loaders.model._configure_embedding_dtypes:345] [PID:25175] Converting modules to torch.bfloat16 | |
| [2025-10-12 23:56:58,382] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:25175] Memory usage after model load 12.891GB (+12.891GB allocated, +12.904GB reserved) | |
| [2025-10-12 23:56:58,383] [INFO] [axolotl.loaders.adapter.load_lora:80] [PID:25175] found linear modules: ['down_proj', 'gate_proj', 'k_proj', 'o_proj', 'q_proj', 'up_proj', 'v_proj'] | |
| [2025-10-12 23:56:58,383] [DEBUG] [axolotl.loaders.adapter.load_lora:143] [PID:25175] Loading pretrained PEFT - LoRA | |
| trainable params: 114,032,640 || all params: 12,361,835,520 || trainable%: 0.9225 | |
| [2025-10-12 23:56:59,786] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:25175] after adapters 11.007GB (+11.007GB allocated, +13.242GB reserved) | |
| [2025-10-12 23:57:00,457] [INFO] [axolotl.cli.merge_lora.do_merge_lora:27] [PID:25175] Running merge of LoRA with base model... | |
| Unloading and merging model: 0%| | 0/767 [00:00<?, ?it/s] Unloading and merging model: 16%|█████████▌ | 119/767 [00:00<00:00, 1189.49it/s] Unloading and merging model: 100%|██████████████████████████████████████████████████████████████| 767/767 [00:00<00:00, 4667.34it/s] | |
| [2025-10-12 23:57:00,628] [INFO] [axolotl.cli.merge_lora.do_merge_lora:40] [PID:25175] Saving merged model to: ckpts-stage-2/merged... | |