[2026-03-12 21:41:44,720] [DEBUG] [axolotl.utils.config.resolve_dtype:66] [PID:4624] bf16 support detected, enabling for this configuration. [2026-03-12 21:41:44,952] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:4624] baseline 0.000GB () [2026-03-12 21:41:44,953] [INFO] [axolotl.cli.config.load_cfg:259] [PID:4624] config: { "activation_offloading": false, "auto_resume_from_checkpoints": true, "axolotl_config_path": "output/sft.yml", "base_model": "kajuma/DiffLlama-1B", "base_model_config": "kajuma/DiffLlama-1B", "batch_size": 32, "bf16": true, "capabilities": { "bf16": true, "compute_capability": "sm_120", "fp8": true, "n_gpu": 1, "n_node": 1 }, "chat_template": "tokenizer_default", "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": 24, "dataset_prepared_path": "./output/dataset", "datasets": [ { "chat_template": "tokenizer_default", "field_messages": "messages", "message_property_mappings": { "content": "content", "role": "role" }, "path": "kajuma/Zero_SFT_Ja_v3.5", "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": "2.8.0" }, "eval_batch_size": 4, "eval_causal_lm_metrics": [ "sacrebleu", "comet", "ter", "chrf" ], "eval_max_new_tokens": 128, "eval_sample_packing": false, "eval_steps": 100, "eval_table_size": 0, "experimental_skip_move_to_device": true, "flash_attention": false, "fp16": false, "gradient_accumulation_steps": 32, "gradient_checkpointing": false, "group_by_length": false, "hf_use_auth_token": true, "include_tkps": true, "is_falcon_derived_model": false, "is_llama_derived_model": true, "is_mistral_derived_model": false, "learning_rate": 0.0005, "liger_cross_entropy": false, "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, "loraplus_lr_embedding": 1e-06, "lr_scheduler": "cosine", "mean_resizing_embeddings": false, "micro_batch_size": 1, "model_config_type": "diffllama", "num_epochs": 1.0, "optimizer": "adamw_torch", "otel_metrics_host": "localhost", "otel_metrics_port": 8000, "output_dir": "./output/model", "pad_to_sequence_len": true, "plugins": [ "axolotl.integrations.liger.LigerPlugin" ], "pretrain_multipack_attn": true, "profiler_steps_start": 0, "qlora_sharded_model_loading": false, "ray_num_workers": 1, "remove_unused_columns": false, "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_steps": 100, "save_strategy": "steps", "save_total_limit": 1, "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, "tf32": false, "tiled_mlp_use_original_mlp": true, "tokenizer_config": "kajuma/DiffLlama-1B", "tokenizer_save_jinja_files": true, "tokenizer_type": "AutoTokenizer", "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 }, "type_of_model": "AutoModelForCausalLM", "use_otel_metrics": false, "use_ray": false, "use_wandb": true, "val_set_size": 0.002, "vllm": { "device": "auto", "dtype": "auto", "gpu_memory_utilization": 0.9, "host": "0.0.0.0", "port": 8000 }, "wandb_entity": "tepic", "wandb_name": "diffllama-sft-datapilot", "wandb_project": "diffllama", "warmup_steps": 20, "weight_decay": 0.01, "world_size": 1 } [2026-03-12 21:41:46,326] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:285] [PID:4624] EOS: 2 / [2026-03-12 21:41:46,326] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:286] [PID:4624] BOS: 1 / [2026-03-12 21:41:46,326] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:287] [PID:4624] PAD: 3 / [2026-03-12 21:41:46,326] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:288] [PID:4624] UNK: 0 / [2026-03-12 21:41:46,326] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:475] [PID:4624] Loading prepared dataset from disk at output/dataset/a693065c12ad716b550e60474f46c363... Loading dataset from disk: 0%| | 0/24 [00:00 [2026-03-12 21:41:52,144] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:286] [PID:4624] BOS: 1 / [2026-03-12 21:41:52,144] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:287] [PID:4624] PAD: 3 / [2026-03-12 21:41:52,144] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:288] [PID:4624] UNK: 0 / [2026-03-12 21:41:52,144] [DEBUG] [axolotl.train.setup_model_and_tokenizer:82] [PID:4624] Loading model [2026-03-12 21:41:52,347] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:87] [PID:4624] Patched Trainer.evaluation_loop with nanmean loss calculation [2026-03-12 21:41:52,347] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:138] [PID:4624] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation [2026-03-12 21:41:52,348] [INFO] [axolotl.loaders.patch_manager._apply_multipack_patches:344] [PID:4624] Applying multipack dataloader patch for sample packing... [2026-03-12 21:41:52,348] [INFO] [axolotl.loaders.patch_manager._patch_llama_sample_packing:473] [PID:4624] Patching llama _prepare_4d_causal_attention_mask*... [2026-03-12 21:41:52,409] [WARNING] [axolotl.integrations.liger.plugin.warning_once:46] [PID:4624] Applied ONLY liger_fused_linear_cross_entropy genericpatches for model type: diffllama [2026-03-12 21:41:52,410] [WARNING] [axolotl.integrations.liger.plugin.warning_once:46] [PID:4624] Liger + diffllama generic FLCE support is experimental and may not work as expected. [2026-03-12 21:41:53,474] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:4624] Memory usage after model load 0.000GB () [2026-03-12 21:41:57,259] [INFO] [axolotl.train.save_initial_configs:417] [PID:4624] Pre-saving tokenizer to ./output/model... [2026-03-12 21:41:57,273] [INFO] [axolotl.train.save_initial_configs:422] [PID:4624] Pre-saving model config to ./output/model... [2026-03-12 21:41:57,274] [INFO] [axolotl.train.execute_training:212] [PID:4624] Starting trainer... [2026-03-12 21:41:59,761] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:4624] generate_batches time: 0.7732937335968018 [2026-03-12 21:42:00,542] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:4624] generate_batches time: 0.7788197994232178 [2026-03-12 21:42:01,330] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:4624] generate_batches time: 0.7867546081542969 [2026-03-12 21:42:02,110] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:4624] generate_batches time: 0.7773358821868896 [2026-03-12 21:42:02,110] [INFO] [axolotl.utils.samplers.multipack.calc_min_len:438] [PID:4624] gather_len_batches: [18405] wandb: Currently logged in as: weak-kajuma (tepic) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin wandb: ⢿ Waiting for wandb.init()... m wandb: ⣻ Waiting for wandb.init()... m wandb: ⣽ setting up run arakm91f (0.2s) m wandb: ⣾ setting up run arakm91f (0.2s) m wandb: ⣷ setting up run arakm91f (0.2s) m wandb: ⣯ setting up run arakm91f (0.2s) m wandb: ⣟ setting up run arakm91f (0.2s) m wandb: ⡿ setting up run arakm91f (0.7s) m wandb: ⢿ setting up run arakm91f (0.7s) m wandb: ⣻ setting up run arakm91f (0.7s) m wandb: Tracking run with wandb version 0.23.1 wandb: Run data is saved locally in /workspace/axolotl/wandb/run-20260312_214202-arakm91f wandb: Run `wandb offline` to turn off syncing. wandb: Syncing run diffllama-sft-datapilot wandb: ⭐️ View project at https://wandb.ai/tepic/diffllama wandb: 🚀 View run at https://wandb.ai/tepic/diffllama/runs/arakm91f wandb: Detected [huggingface_hub.inference] in use. wandb: Use W&B Weave for improved LLM call tracing. Install Weave with `pip install weave` then add `import weave` to the top of your script. wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/ 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") wandb: WARNING Symlinked 1 file into the W&B run directory; call wandb.save again to sync new files. [2026-03-12 21:42:05,653] [INFO] [axolotl.utils.callbacks.on_train_begin:757] [PID:4624] The Axolotl config has been saved to the WandB run under files. 0%| | 0/575 [00:00 [2026-03-15 08:11:17,720] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:286] [PID:5120] BOS: 1 / [2026-03-15 08:11:17,720] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:287] [PID:5120] PAD: 3 / [2026-03-15 08:11:17,720] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:288] [PID:5120] UNK: 0 / [2026-03-15 08:11:17,720] [INFO] [axolotl.cli.utils.load.load_model_and_tokenizer:43] [PID:5120] loading model... [2026-03-15 08:11:17,725] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:87] [PID:5120] Patched Trainer.evaluation_loop with nanmean loss calculation [2026-03-15 08:11:17,726] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:138] [PID:5120] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation [2026-03-15 08:11:17,788] [WARNING] [axolotl.integrations.liger.plugin.warning_once:46] [PID:5120] Applied ONLY liger_fused_linear_cross_entropy genericpatches for model type: diffllama [2026-03-15 08:11:17,788] [WARNING] [axolotl.integrations.liger.plugin.warning_once:46] [PID:5120] Liger + diffllama generic FLCE support is experimental and may not work as expected. [2026-03-15 08:11:19,249] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:5120] Memory usage after model load 0.000GB () * Running on local URL: http://127.0.0.1:7860 * Running on public URL: https://48ecbc6e83cc897113.gradio.live This share link expires in 1 week. For free permanent hosting and GPU upgrades, run `gradio deploy` from the terminal in the working directory to deploy to Hugging Face Spaces (https://huggingface.co/spaces) Using existing dataset file at: .gradio/flagged/dataset1.csv Keyboard interruption in main thread... closing server. Killing tunnel 127.0.0.1:7860 <> https://48ecbc6e83cc897113.gradio.live