[2026-08-18 14:31:00,725] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:12035] baseline 0.000GB () [2026-08-18 14:31:00,726] [INFO] [axolotl.cli.config.load_cfg:333] [PID:12035] config: { "activation_offloading": false, "adapter": "lora", "attn_implementation": "sdpa", "attn_needs_dtype_cast": false, "attn_supports_packing": false, "attn_uses_flash_lib": false, "axolotl_config_path": "/workspace/wave/training/axolotl.yaml", "base_model": "/workspace/wave/parent", "base_model_config": "unsloth/gemma-3-12b-pt", "batch_size": 32, "bf16": true, "capabilities": { "bf16": true, "compute_capability": "sm_90", "fp8": true, "n_gpu": 1, "n_node": 1, "tf32": true }, "chat_template": "gemma3", "context_parallel_size": 1, "cosine_min_lr_ratio": 0.1, "dataloader_num_workers": 1, "dataloader_pin_memory": true, "dataloader_prefetch_factor": 256, "dataset_num_proc": 8, "dataset_prepared_path": "/workspace/wave/training/prepared", "datasets": [ { "chat_template": "tokenizer_default", "field_messages": "messages", "message_property_mappings": { "content": "content", "role": "role" }, "path": "/workspace/wave/data/datasets/aft_coin2.jsonl", "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": "2.12.1" }, "eot_tokens": [ "" ], "eval_batch_size": 16, "eval_causal_lm_metrics": [ "sacrebleu", "comet", "ter", "chrf" ], "eval_max_new_tokens": 128, "eval_table_size": 0, "experimental_skip_move_to_device": true, "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": 2, "gradient_checkpointing": true, "gradient_checkpointing_kwargs": { "use_reentrant": true }, "include_tkps": true, "is_multimodal": true, "layer_offloading": false, "learning_rate": 0.0001, "liger_fused_linear_cross_entropy": true, "liger_glu_activation": 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": 64, "lora_dropout": 0.05, "lora_embedding_kernel": true, "lora_mlp_kernel": true, "lora_o_kernel": true, "lora_qkv_kernel": true, "lora_r": 32, "lora_target_modules": [ "q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj" ], "loraplus_lr_embedding": 1e-06, "lr_scheduler": "cosine", "max_grad_norm": 1.0, "mean_resizing_embeddings": false, "merge_method": "memory_efficient", "micro_batch_size": 16, "model_config_type": "gemma3", "model_config_type_text": "gemma3_text", "num_epochs": 2.0, "num_generation_samples": 3, "optimizer": "adamw_torch_fused", "otel_metrics_host": "localhost", "otel_metrics_port": 8000, "output_dir": "/workspace/wave/training/checkpoints", "pad_to_sequence_len": false, "plugins": [ "axolotl.integrations.liger.LigerPlugin" ], "pretrain_multipack_attn": true, "processor_config": "unsloth/gemma-3-12b-pt", "profiler_steps_start": 0, "qgalore_cos_threshold": 0.4, "qgalore_gamma_proj": 2, "qgalore_proj_bits": 4, "qgalore_proj_group_size": 256, "qgalore_proj_quant": true, "qgalore_proj_type": "std", "qgalore_queue_size": 5, "qgalore_rank": 256, "qgalore_scale": 0.25, "qgalore_update_proj_gap": 200, "qlora_sharded_model_loading": false, "quantize_moe_experts": false, "ray_num_workers": 1, "relora_prune_method": "magnitude", "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_steps": 32, "save_strategy": "steps", "save_total_limit": 20, "seed": 42, "sequence_len": 1280, "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": "unsloth/gemma-3-12b-pt", "tokenizer_save_jinja_files": true, "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 }, "trust_remote_code": false, "use_otel_metrics": false, "use_ray": false, "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": 0.01, "world_size": 1 } [2026-08-18 14:31:00,837] [DEBUG] [axolotl.loaders.utils.check_model_config:88] [PID:12035] Loaded image size: 896 from model config [2026-08-18 14:31:02,584] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:12035] EOS: 1 / [2026-08-18 14:31:02,584] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:12035] BOS: 2 / [2026-08-18 14:31:02,584] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:12035] PAD: 0 / [2026-08-18 14:31:02,584] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:12035] UNK: 3 / [2026-08-18 14:31:02,585] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:482] [PID:12035] Unable to find prepared dataset in /workspace/wave/training/prepared/190a1e7151a6ac84f3e1d4ab8702c044 [2026-08-18 14:31:02,585] [INFO] [axolotl.utils.data.sft._load_raw_datasets:320] [PID:12035] Loading raw datasets... [2026-08-18 14:31:02,585] [WARNING] [axolotl.utils.data.sft._load_raw_datasets:322] [PID:12035] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`. [2026-08-18 14:31:02,757] [INFO] [axolotl.utils.data.wrappers.get_dataset_wrapper:87] [PID:12035] Loading dataset: /workspace/wave/data/datasets/aft_coin2.jsonl with base_type: chat_template and prompt_style: None [2026-08-18 14:31:02,759] [INFO] [axolotl.prompt_strategies.chat_template.__call__:1209] [PID:12035] Using chat template: --- {{ bos_token }} {%- if messages[0]['role'] == 'system' -%} {%- if messages[0]['content'] is string -%} {%- set first_user_prefix = messages[0]['content'] + ' ' -%} {%- else -%} {%- set first_user_prefix = messages[0]['content'][0]['text'] + ' ' -%} {%- endif -%} {%- set loop_messages = messages[1:] -%} {%- else -%} {%- set first_user_prefix = "" -%} {%- set loop_messages = messages -%} {%- endif -%} {%- for message in loop_messages -%} {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%} {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }} {%- endif -%} {%- if (message['role'] == 'assistant') -%} {%- set role = "model" -%} {%- else -%} {%- set role = message['role'] -%} {%- endif -%} {{ '' + role + ' ' + (first_user_prefix if loop.first else "") }} {%- if message['content'] is string -%} {{ message['content'] | trim }} {%- elif message['content'] is iterable -%} {%- for item in message['content'] -%} {%- if item['type'] == 'image' -%} {{ '' }} {%- elif item['type'] == 'text' -%} {{ item['text'] | trim }} {%- endif -%} {%- endfor -%} {%- else -%} {{ raise_exception("Invalid content type") }} {%- endif -%} {{ ' ' }} {%- endfor -%} {%- if add_generation_prompt -%} {{'model '}} {%- endif -%} --- [2026-08-18 14:31:14,861] [INFO] [axolotl.utils.data.utils._log_dataset_stats:212] [PID:12035] min_input_len: 446 [2026-08-18 14:31:14,861] [INFO] [axolotl.utils.data.utils._log_dataset_stats:213] [PID:12035] max_input_len: 967 Saving the dataset (0/8 shards): 0%| | 0/8192 [00:00 [2026-08-18 14:31:26,745] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:12035] BOS: 2 / [2026-08-18 14:31:26,745] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:12035] PAD: 0 / [2026-08-18 14:31:26,746] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:12035] UNK: 3 / [2026-08-18 14:31:30,236] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:12035] Loading model [2026-08-18 14:31:30,296] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:12035] Patched OptimState8bit for torch.compile compatibility [2026-08-18 14:31:30,296] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:12035] Patched OptimState4bit for torch.compile compatibility [2026-08-18 14:31:30,296] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:12035] Patched OptimStateFp8 for torch.compile compatibility [2026-08-18 14:31:30,301] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:12035] Patched Trainer.evaluation_loop with nanmean loss calculation [2026-08-18 14:31:30,302] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:12035] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation [2026-08-18 14:31:30,302] [WARNING] [axolotl.loaders.patch_manager._apply_self_attention_lora_patch:662] [PID:12035] Cannot patch self-attention - requires no dropout [2026-08-18 14:31:31,388] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:117] [PID:12035] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True} Loading weights: 0%| | 0/1066 [00:00