[2025-12-13 20:39:02,352] [DEBUG] [axolotl.utils.config.resolve_dtype:66] [PID:1121389] bf16 support detected, enabling for this configuration. [2025-12-13 20:39:02,354] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:1121389] baseline 0.000GB () [2025-12-13 20:39:02,354] [INFO] [axolotl.cli.config.load_cfg:248] [PID:1121389] config: { "activation_offloading": false, "axolotl_config_path": "train.lima.yml", "base_model": "/home/alex/Workspace/sllama/out_5/checkpoint-1722000", "base_model_config": "/home/alex/Workspace/sllama/out_5/checkpoint-1722000", "batch_size": 1, "bf16": true, "capabilities": { "bf16": true, "compute_capability": "sm_120", "fp8": false, "n_gpu": 1, "n_node": 1 }, "context_parallel_size": 1, "cosine_constant_lr_ratio": 0.1, "cut_cross_entropy": true, "dataloader_num_workers": 0, "dataloader_pin_memory": true, "dataset_prepared_path": "data_prep", "dataset_processes": 12, "datasets": [ { "chat_template": "tokenizer_default", "message_property_mappings": { "content": "content", "role": "role" }, "path": "lima.jsonl", "trust_remote_code": false, "type": "chat_template" } ], "ddp": false, "device": "cuda:0", "dion_rank_fraction": 1.0, "dion_rank_multiple_of": 1, "env_capabilities": { "torch_version": "2.7.1" }, "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, "fp8": true, "gradient_accumulation_steps": 1, "gradient_checkpointing": false, "gradient_checkpointing_kwargs": { "use_reentrant": false }, "group_by_length": false, "include_tkps": true, "is_llama_derived_model": true, "learning_rate": 0.1, "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": 10, "lora_dropout": 0.0, "loraplus_lr_embedding": 1e-06, "lr_scheduler": "cosine", "max_grad_norm": 1.0, "mean_resizing_embeddings": false, "micro_batch_size": 1, "model_config_type": "llama", "num_epochs": 4.0, "optim_args": { "momentum": 0.98 }, "optimizer": "sgd", "output_dir": "./out_6_lima", "pad_to_sequence_len": true, "plugins": [ "axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin", "axolotl.integrations.liger.LigerPlugin" ], "pretrain_multipack_attn": true, "profiler_steps_start": 0, "qlora_sharded_model_loading": false, "ray_num_workers": 1, "resize_token_embeddings_to_32x": true, "resources_per_worker": { "GPU": 1 }, "sample_packing": true, "sample_packing_bin_size": 200, "sample_packing_group_size": 100000, "save_only_model": true, "save_safetensors": true, "save_steps": 100, "save_total_limit": 2, "sequence_len": 2048, "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": "/home/alex/Workspace/sllama/out_5/checkpoint-1722000", "tokenizer_save_jinja_files": true, "torch_compile": true, "torch_compile_backend": "inductor", "torch_compile_mode": "default", "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 }, "trust_remote_code": true, "unfrozen_parameters": [ "^(?![\\s\\S]*embed_tokens)[\\s\\S]+$" ], "use_ray": false, "use_tensorboard": true, "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_project": "sllama", "warmup_ratio": 0.05, "weight_decay": 0.01, "world_size": 1 } [2025-12-13 20:39:02,909] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:278] [PID:1121389] EOS: 128009 / <|eot_id|> [2025-12-13 20:39:02,909] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:279] [PID:1121389] BOS: 128000 / <|begin_of_text|> [2025-12-13 20:39:02,909] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:280] [PID:1121389] PAD: 128004 / <|finetune_right_pad_id|> [2025-12-13 20:39:02,909] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:281] [PID:1121389] UNK: None / None [2025-12-13 20:39:02,909] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:470] [PID:1121389] Loading prepared dataset from disk at data_prep/0ed3928757e6a5973cddee77ac8c62e0... [2025-12-13 20:39:02,913] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:404] [PID:1121389] total_num_tokens: 640_837 [2025-12-13 20:39:02,917] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:422] [PID:1121389] `total_supervised_tokens: 445_575` [2025-12-13 20:39:03,640] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.36017513275146484 [2025-12-13 20:39:03,998] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.35738229751586914 [2025-12-13 20:39:04,353] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.35471415519714355 [2025-12-13 20:39:04,713] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.3598747253417969 [2025-12-13 20:39:04,724] [INFO] [axolotl.utils.samplers.multipack.calc_min_len:434] [PID:1121389] gather_len_batches: [322] [2025-12-13 20:39:04,724] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:481] [PID:1121389] data_loader_len: 322 [2025-12-13 20:39:04,725] [INFO] [axolotl.utils.trainer.calc_sample_packing_eff_est:497] [PID:1121389] sample_packing_eff_est across ranks: [0.9687575585332817] [2025-12-13 20:39:04,725] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:509] [PID:1121389] sample_packing_eff_est: 0.97 [2025-12-13 20:39:04,725] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:520] [PID:1121389] total_num_steps: 1288 [2025-12-13 20:39:04,725] [INFO] [axolotl.utils.data.sft._prepare_standard_dataset:121] [PID:1121389] Maximum number of steps set at 1288 [2025-12-13 20:39:04,740] [DEBUG] [axolotl.train.setup_model_and_tokenizer:70] [PID:1121389] Loading tokenizer... /home/alex/Workspace/sllama/out_5/checkpoint-1722000 [2025-12-13 20:39:04,952] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:278] [PID:1121389] EOS: 128009 / <|eot_id|> [2025-12-13 20:39:04,952] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:279] [PID:1121389] BOS: 128000 / <|begin_of_text|> [2025-12-13 20:39:04,952] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:280] [PID:1121389] PAD: 128004 / <|finetune_right_pad_id|> [2025-12-13 20:39:04,952] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:281] [PID:1121389] UNK: None / None [2025-12-13 20:39:04,952] [DEBUG] [axolotl.train.setup_model_and_tokenizer:79] [PID:1121389] Loading model [2025-12-13 20:39:04,955] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:87] [PID:1121389] Patched Trainer.evaluation_loop with nanmean loss calculation [2025-12-13 20:39:04,956] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:138] [PID:1121389] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation [2025-12-13 20:39:04,957] [INFO] [axolotl.monkeypatch.trainer_accelerator_args.patch_create_accelerate_code_for_fp8:80] [PID:1121389] patching create_accelerator_and_postprocess to allow for overrides [2025-12-13 20:39:04,957] [INFO] [axolotl.loaders.patch_manager._apply_multipack_patches:294] [PID:1121389] Applying multipack dataloader patch for sample packing... [2025-12-13 20:39:05,015] [INFO] [axolotl.integrations.cut_cross_entropy.pre_model_load:94] [PID:1121389] Applying Cut Cross Entropy to model type: llama [2025-12-13 20:39:05,020] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:71] [PID:1121389] Applying LIGER to llama with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': None, 'rms_norm': True, 'swiglu': True} [2025-12-13 20:39:05,890] [INFO] [axolotl.loaders.model._configure_embedding_dtypes:345] [PID:1121389] Converting modules to torch.bfloat16 [2025-12-13 20:39:06,358] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:1121389] Memory usage after model load 0.000GB () [2025-12-13 20:39:06,858] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.self_attn.q_proj.weight [2025-12-13 20:39:06,858] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.self_attn.k_proj.weight [2025-12-13 20:39:06,858] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.self_attn.v_proj.weight [2025-12-13 20:39:06,858] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.self_attn.o_proj.weight [2025-12-13 20:39:06,859] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.mlp.gate_proj.weight [2025-12-13 20:39:06,859] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.mlp.up_proj.weight [2025-12-13 20:39:06,859] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.mlp.down_proj.weight [2025-12-13 20:39:06,859] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.input_layernorm.weight [2025-12-13 20:39:06,859] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.layers.0.post_attention_layernorm.weight [2025-12-13 20:39:06,859] [DEBUG] [axolotl.utils.freeze.freeze_layers_except:56] [PID:1121389] Unfrozen model.norm.weight [2025-12-13 20:39:07,912] [INFO] [axolotl.train.save_initial_configs:412] [PID:1121389] Pre-saving tokenizer to ./out_6_lima... [2025-12-13 20:39:07,973] [INFO] [axolotl.train.save_initial_configs:417] [PID:1121389] Pre-saving model config to ./out_6_lima... [2025-12-13 20:39:07,975] [INFO] [axolotl.train.execute_training:203] [PID:1121389] Starting trainer... [2025-12-13 20:39:09,007] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.37635207176208496 [2025-12-13 20:39:09,392] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.3846709728240967 [2025-12-13 20:39:09,774] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.38227200508117676 [2025-12-13 20:39:10,157] [DEBUG] [axolotl.utils.samplers.multipack.__len__:458] [PID:1121389] generate_batches time: 0.3822507858276367 [2025-12-13 20:39:10,157] [INFO] [axolotl.utils.samplers.multipack.calc_min_len:434] [PID:1121389] gather_len_batches: [323] wandb: Currently logged in as: alex-ht to https://api.wandb.ai. Use `wandb login --relogin` to force relogin wandb: Tracking run with wandb version 0.21.1 wandb: Run data is saved locally in /home/alex/Workspace/sllama/wandb/run-20251213_203911-00ktlhat wandb: Run `wandb offline` to turn off syncing. wandb: Syncing run step6_lima wandb: ⭐️ View project at https://wandb.ai/alex-ht/sllama wandb: 🚀 View run at https://wandb.ai/alex-ht/sllama/runs/00ktlhat wandb: WARNING Saving files without folders. If you want to preserve subdirectories pass base_path to wandb.save, i.e. wandb.save("/mnt/folder/file.h5", base_path="/mnt") [2025-12-13 20:39:13,730] [INFO] [axolotl.utils.callbacks.on_train_begin:793] [PID:1121389] The Axolotl config has been saved to the WandB run under files. 0%| | 0/1288 [00:00