See axolotl config

axolotl version: 0.4.1

yaml

1adapter: lora 2base_model: Qwen/Qwen2.5-Math-7B-Instruct 3bf16: true 4chat_template: llama3 5dataset_prepared_path: null 6datasets: 7- data_files: 8 - 62a9e2878ae88813_train_data.json 9 ds_type: json 10 format: custom 11 path: /workspace/input_data/62a9e2878ae88813_train_data.json 12 type: 13 field_instruction: question 14 field_output: answer 15 format: '{instruction}' 16 no_input_format: '{instruction}' 17 system_format: '{system}' 18 system_prompt: '' 19debug: null 20device_map: 21 ? '' 22 : 0,1,2,3,4,5,6,7 23early_stopping_patience: 2 24eval_max_new_tokens: 128 25eval_steps: 100 26eval_table_size: null 27flash_attention: true 28gradient_accumulation_steps: 8 29gradient_checkpointing: true 30group_by_length: false 31hub_model_id: Alphatao/b2f97e89-fe1c-4a1d-a135-f54c9da4f682 32hub_repo: null 33hub_strategy: null 34hub_token: null 35learning_rate: 0.0002 36load_best_model_at_end: true 37load_in_4bit: false 38load_in_8bit: false 39local_rank: null 40logging_steps: 1 41lora_alpha: 128 42lora_dropout: 0.1 43lora_fan_in_fan_out: null 44lora_model_dir: null 45lora_r: 64 46lora_target_linear: true 47lora_target_modules: 48- q_proj 49- k_proj 50- v_proj 51- o_proj 52lr_scheduler: cosine 53max_grad_norm: 1.0 54max_steps: 3168 55micro_batch_size: 4 56mlflow_experiment_name: /tmp/62a9e2878ae88813_train_data.json 57model_type: AutoModelForCausalLM 58num_epochs: 10 59optimizer: adamw_bnb_8bit 60output_dir: miner_id_24 61pad_to_sequence_len: true 62resume_from_checkpoint: null 63s2_attention: null 64sample_packing: false 65save_steps: 100 66sequence_len: 1024 67strict: false 68tf32: true 69tokenizer_type: AutoTokenizer 70train_on_inputs: false 71trust_remote_code: true 72val_set_size: 0.023880026745629956 73wandb_entity: null 74wandb_mode: online 75wandb_name: 4c2648cd-2f83-40bf-8de7-aebad4780f8e 76wandb_project: Gradients-On-Demand 77wandb_run: your_name 78wandb_runid: 4c2648cd-2f83-40bf-8de7-aebad4780f8e 79warmup_steps: 10 80weight_decay: 0.0 81xformers_attention: null 82

b2f97e89-fe1c-4a1d-a135-f54c9da4f682

• Loss: 1.8621

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 0.0002
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• training_steps: 3168

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