See axolotl config

axolotl version: 0.4.1

yaml

1adapter: lora 2base_model: Qwen/Qwen2-1.5B-Instruct 3bf16: true 4chat_template: llama3 5dataloader_num_workers: 8 6dataloader_pin_memory: true 7dataset_prepared_path: null 8datasets: 9- data_files: 10 - 7376a9b16ca78e0e_train_data.json 11 ds_type: json 12 format: custom 13 path: /workspace/input_data/ 14 type: 15 field_instruction: instruct 16 field_output: output 17 format: '{instruction}' 18 no_input_format: '{instruction}' 19 system_format: '{system}' 20 system_prompt: '' 21debug: null 22deepspeed: null 23device_map: auto 24dynamic_lora_per_layer: true 25early_stopping_patience: 3 26eval_max_new_tokens: 128 27eval_steps: 100 28eval_table_size: null 29evaluation_strategy: steps 30flash_attention: true 31fp16: false 32fsdp: null 33fsdp_config: null 34gradient_accumulation_steps: 2 35gradient_checkpointing: true 36group_by_length: false 37hub_model_id: JoshMe1/74ef9b35-3518-48d8-9d91-f06849ea9261 38hub_repo: null 39hub_strategy: checkpoint 40hub_token: null 41learning_rate: 5.0e-05 42load_in_4bit: false 43load_in_8bit: true 44local_rank: null 45logging_steps: 10 46lora_alpha: 128 47lora_dropout: 0.1 48lora_fan_in_fan_out: null 49lora_model_dir: null 50lora_r: 64 51lora_target_linear: true 52lr_finder: true 53lr_scheduler: cosine 54lr_scheduler_args: [] 55max_grad_norm: 1.0 56max_memory: 57 0: 130GB 58max_steps: 3972 59micro_batch_size: 8 60mixed_precision: bf16 61mlflow_experiment_name: /tmp/7376a9b16ca78e0e_train_data.json 62model_type: AutoModelForCausalLM 63num_epochs: 10 64optimizer: adamw_bnb_8bit 65output_dir: miner_id_24 66pad_to_sequence_len: true 67resume_from_checkpoint: null 68s2_attention: null 69sample_packing: false 70save_steps: 100 71save_strategy: steps 72save_total_limit: 3 73scheduler: 74 factor: 0.5 75 monitor: eval_loss 76 patience: 1 77 threshold: 0.01 78 type: ReduceLROnPlateau 79sequence_len: 512 80strict: false 81tf32: false 82tokenizer_type: AutoTokenizer 83train_on_inputs: false 84training_stages: 85- learning_rate: 5.0e-05 86 name: warmup 87 num_train_epochs: 1 88- learning_rate: 5.0e-06 89 name: main 90trl: 91 ema: true 92 ema_decay: 0.999 93trust_remote_code: true 94val_set_size: 0.05 95wandb_entity: null 96wandb_mode: online 97wandb_name: fe9696d5-94a1-410e-8793-2418f6c2af21 98wandb_project: Gradients-On-Demand 99wandb_run: your_name 100wandb_runid: fe9696d5-94a1-410e-8793-2418f6c2af21 101warmup_steps: 397 102weight_decay: 0.01 103xformers_attention: true 104

74ef9b35-3518-48d8-9d91-f06849ea9261

• Loss: 1.5638

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 397
• training_steps: 3972

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