See axolotl config

axolotl version: 0.4.1

yaml

1adapter: lora 2base_model: unsloth/Llama-3.2-3B-Instruct 3bf16: auto 4chat_template: llama3 5cosine_min_lr_ratio: 0.1 6data_processes: 16 7dataset_prepared_path: null 8datasets: 9- format: custom 10 path: lavita/ChatDoctor-HealthCareMagic-100k 11 type: 12 field_input: input 13 field_instruction: instruction 14 field_output: output 15 format: '{instruction} {input}' 16 no_input_format: '{instruction}' 17 system_format: '{system}' 18 system_prompt: '' 19debug: null 20deepspeed: null 21device_map: '{'''':torch.cuda.current_device()}' 22do_eval: true 23early_stopping_patience: 1 24eval_batch_size: 1 25eval_sample_packing: false 26eval_steps: 25 27evaluation_strategy: steps 28flash_attention: true 29fp16: null 30fsdp: null 31fsdp_config: null 32gradient_accumulation_steps: 32 33gradient_checkpointing: true 34group_by_length: true 35hub_model_id: cwaud/ea258833f53-4c8a-4b1a-9abf-59da4fa11e18 36hub_repo: stevemonite 37hub_strategy: checkpoint 38hub_token: null 39learning_rate: 0.0002 40load_in_4bit: false 41load_in_8bit: false 42local_rank: null 43logging_steps: 1 44lora_alpha: 32 45lora_dropout: 0.05 46lora_fan_in_fan_out: null 47lora_model_dir: null 48lora_r: 16 49lora_target_linear: true 50lora_target_modules: 51- q_proj 52- v_proj 53lr_scheduler: cosine 54max_grad_norm: 1.0 55max_memory: 56 0: 70GiB 57 1: 70GiB 58 2: 70GiB 59 3: 70GiB 60max_steps: 888 61micro_batch_size: 1 62mlflow_experiment_name: lavita/ChatDoctor-HealthCareMagic-100k 63model_type: AutoModelForCausalLM 64num_epochs: 3 65optim_args: 66 adam_beta1: 0.9 67 adam_beta2: 0.95 68 adam_epsilon: 1e-5 69optimizer: adamw_torch 70output_dir: miner_id_24 71pad_to_sequence_len: true 72resume_from_checkpoint: null 73s2_attention: null 74sample_packing: false 75save_steps: 50 76save_strategy: steps 77sequence_len: 2048 78strict: false 79tf32: false 80tokenizer_type: AutoTokenizer 81torch_compile: false 82train_on_inputs: false 83trust_remote_code: true 84val_set_size: 50 85wandb_entity: null 86wandb_mode: online 87wandb_project: Public_TuningSN 88wandb_run: miner_id_24 89wandb_runid: 153c7d6f 90warmup_raio: 0.03 91warmup_ratio: 0.04 92weight_decay: 0.01 93xformers_attention: null 94

ea258833f53-4c8a-4b1a-9abf-59da4fa11e18

• Loss: 2.1079

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 0.0002
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 32
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=adam_beta1=0.9,adam_beta2=0.95,adam_epsilon=1e-5
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 35
• training_steps: 888

Training results

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1