config.json: 0.00B [00:00, ?B/s] config.json: 4.62kB [00:00, 22.1MB/s] [2026-04-29 00:28:59,993] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:3345] baseline 0.000GB () [2026-04-29 00:28:59,994] [INFO] [axolotl.cli.config.load_cfg:341] [PID:3345] config: { "activation_offloading": true, "adapter": "lora", "axolotl_config_path": "Mura.yaml", "base_model": "google/gemma-4-31B-it", "base_model_config": "google/gemma-4-31B-it", "batch_size": 8, "bf16": true, "capabilities": { "bf16": true, "compute_capability": "sm_120", "fp8": true, "n_gpu": 1, "n_node": 1, "tf32": true }, "chat_template": "jinja", "chat_template_jinja": "{%- macro strip_thinking(text) -%}\n {%- set ns = namespace(result='') -%}\n {%- for part in text.split('') -%}\n {%- if '<|channel>' in part -%}\n {%- set ns.result = ns.result + part.split('<|channel>')[0] -%}\n {%- else -%}\n {%- set ns.result = ns.result + part -%}\n {%- endif -%}\n {%- endfor -%}\n {{- ns.result | trim -}}\n{%- endmacro -%}\n{%- set loop_messages = messages -%} {{ bos_token }}\n{#- Handle System Definitions Block -#} {%- if (enable_thinking is defined and enable_thinking) or messages[0]['role'] in ['system', 'developer'] -%}\n {{- '<|turn>system\\n' -}}\n\n {#- Inject Thinking token at the very top of the FIRST system turn -#}\n {%- if enable_thinking is defined and enable_thinking -%}\n {{- '<|think|>' -}}\n {%- endif -%}\n\n {%- if messages[0]['role'] in ['system', 'developer'] -%}\n {{- messages[0]['content'] | trim -}}\n {%- set loop_messages = messages[1:] -%}\n {%- endif -%}\n \n {{- '\\n' -}}\n{%- endif %}\n{#- Loop through messages -#} {%- for message in loop_messages -%}\n {%- set role = 'model' if message['role'] == 'assistant' else message['role'] -%}\n {{- '<|turn>' + role + '\\n' -}}\n\n {#- Flag to identify the final SFT turn -#}\n {%- set is_final_sft_turn = loop.last and not add_generation_prompt -%}\n\n {%- if message['content'] is string -%}\n {%- if role == 'model' -%}\n {%- if is_final_sft_turn and '<|channel>thought' not in message['content'] -%}\n {{- '<|channel>thought\\n' -}}\n {%- endif -%}\n {{- strip_thinking(message['content']) -}}\n {%- else -%}\n {{- message['content'] | trim -}}\n {%- endif -%}\n {%- elif message['content'] is sequence -%}\n {%- set ns = namespace(has_thinking=false) -%}\n {%- for item in message['content'] -%}\n {%- if item['type'] == 'text' and '<|channel>thought' in item['text'] -%}\n {%- set ns.has_thinking = true -%}\n {%- endif -%}\n {%- endfor -%}\n \n {%- if role == 'model' and is_final_sft_turn and not ns.has_thinking -%}\n {{- '<|channel>thought\\n' -}}\n {%- endif -%}\n\n {%- for item in message['content'] -%}\n {%- if item['type'] == 'text' -%}\n {%- if role == 'model' -%}\n {{- strip_thinking(item['text']) -}}\n {%- else -%}\n {{- item['text'] | trim -}}\n {%- endif -%}\n {%- endif -%}\n {%- endfor -%}\n {%- endif -%}\n\n {{- '\\n' -}}\n{%- endfor -%}\n{#- Generation Prompt handled as normal (serves as the final turn when true) -#} {%- if add_generation_prompt -%}\n {{- '<|turn>model\\n' -}}\n {%- if not enable_thinking | default(false) -%}\n {{- '<|channel>thought\\n' -}}\n {%- endif -%}\n{%- endif -%}\n", "context_parallel_size": 1, "cut_cross_entropy": true, "dataloader_num_workers": 1, "dataloader_pin_memory": true, "dataloader_prefetch_factor": 256, "dataset_num_proc": 16, "datasets": [ { "chat_template": "tokenizer_default", "message_property_mappings": { "content": "content", "role": "role" }, "path": "ConicCat/Mura_Books", "trust_remote_code": false, "type": "chat_template" } ], "ddp": false, "device": "cuda:0", "dion_rank_fraction": 1.0, "dion_rank_multiple_of": 1, "eaft_alpha": 1.0, "eaft_k": 20, "env_capabilities": { "torch_version": "2.8.0" }, "eot_tokens": [ "" ], "eval_batch_size": 2, "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": false, "fp16": false, "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": true, "gradient_checkpointing_kwargs": { "use_reentrant": true }, "hf_use_auth_token": true, "include_tkps": true, "is_multimodal": true, "layer_offloading": false, "learning_rate": 2.5e-05, "liger_glu_activation": true, "liger_layer_norm": true, "liger_rms_norm": true, "liger_rms_norm_gated": 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": 64, "lora_dropout": 0.0, "lora_mlp_kernel": false, "lora_o_kernel": false, "lora_qkv_kernel": false, "lora_r": 32, "lora_target_modules": "model.language_model.layers.[\\d]+.(_checkpoint_wrapped_module.)?(mlp|self_attn).(up|down|gate|q|k|v|o)_proj", "loraplus_lr_embedding": 1e-06, "loraplus_lr_ratio": 16.0, "lr_scheduler": "cosine", "max_grad_norm": 1.0, "mean_resizing_embeddings": false, "merge_method": "memory_efficient", "micro_batch_size": 2, "model_config_type": "gemma4", "model_config_type_text": "gemma4_text", "num_epochs": 11.0, "num_generation_samples": 3, "optimizer": "paged_adamw_8bit", "otel_metrics_host": "localhost", "otel_metrics_port": 8000, "output_dir": "./Writer-Stage-2", "pad_to_sequence_len": true, "plugins": [ "axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin", "axolotl.integrations.liger.LigerPlugin" ], "pretrain_multipack_attn": true, "processor_config": "google/gemma-4-31B-it", "profiler_steps_start": 0, "push_dataset_to_hub": "ConicCat/Gemma4-Mura", "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, "save_strategy": "no", "sdp_attention": true, "seed": 42, "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, "tf32": true, "tiled_mlp_use_original_mlp": true, "tokenizer_config": "google/gemma-4-31B-it", "tokenizer_save_jinja_files": true, "torch_compile": false, "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_tensorboard": true, "val_set_size": 0.0, "vllm": { "device": "auto", "dtype": "auto", "gpu_memory_utilization": 0.9, "host": "0.0.0.0", "port": 8000 }, "warmup_ratio": 0.05, "weight_decay": 0.0, "world_size": 1 } tokenizer_config.json: 0.00B [00:00, ?B/s] tokenizer_config.json: 2.10kB [00:00, 9.15MB/s] tokenizer.json: 0%| | 0.00/32.2M [00:00 [2026-04-29 00:29:02,981] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:308] [PID:3345] BOS: 2 / [2026-04-29 00:29:02,981] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:309] [PID:3345] PAD: 0 / [2026-04-29 00:29:02,982] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:310] [PID:3345] UNK: 3 / [2026-04-29 00:29:02,987] [INFO] [axolotl.utils.data.shared.try_load_from_hub:491] [PID:3345] Attempting to load prepared dataset from HuggingFace Hub at ConicCat/Gemma4-Mura (version cd09e8d53ec6484b42196b1f3b13c4dd)... README.md: 0%| | 0.00/523 [00:00 [2026-04-29 00:29:09,311] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:308] [PID:3345] BOS: 2 / [2026-04-29 00:29:09,311] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:309] [PID:3345] PAD: 0 / [2026-04-29 00:29:09,311] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:310] [PID:3345] UNK: 3 / processor_config.json: 0.00B [00:00, ?B/s] processor_config.json: 1.69kB [00:00, 9.14MB/s] [2026-04-29 00:29:13,167] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:3345] Loading model [2026-04-29 00:29:13,230] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:3345] Patched OptimState8bit for torch.compile compatibility [2026-04-29 00:29:13,231] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:3345] Patched OptimState4bit for torch.compile compatibility [2026-04-29 00:29:13,231] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:3345] Patched OptimStateFp8 for torch.compile compatibility [2026-04-29 00:29:13,234] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:3345] Patched Trainer.evaluation_loop with nanmean loss calculation [2026-04-29 00:29:13,235] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:3345] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation [2026-04-29 00:29:13,236] [INFO] [axolotl.loaders.patch_manager._apply_multipack_patches:478] [PID:3345] Applying multipack dataloader patch for sample packing... [2026-04-29 00:29:13,249] [INFO] [axolotl.integrations.cut_cross_entropy.pre_model_load:94] [PID:3345] Applying Cut Cross Entropy to model type: gemma4 [2026-04-29 00:29:13,382] [WARNING] [axolotl.integrations.liger.plugin.pre_model_load:241] [PID:3345] Unsupported model config type: gemma4. Liger not applied. model.safetensors.index.json: 0.00B [00:00, ?B/s] model.safetensors.index.json: 120kB [00:00, 87.5MB/s] Downloading (incomplete total...): 0.00B [00:00, ?B/s] Fetching 2 files: 0%| | 0/2 [00:00