See axolotl config

axolotl version: 0.4.1

yaml

1adapter: lora 2base_model: unsloth/SmolLM-360M-Instruct 3bf16: true 4chat_template: llama3 5dataset_prepared_path: null 6datasets: 7- data_files: 8 - 824ef92913906b02_train_data.json 9 ds_type: json 10 format: custom 11 path: /workspace/input_data/824ef92913906b02_train_data.json 12 type: 13 field_input: gt_answer 14 field_instruction: question 15 field_output: answer 16 format: '{instruction} {input}' 17 no_input_format: '{instruction}' 18 system_format: '{system}' 19 system_prompt: '' 20debug: null 21deepspeed: null 22device_map: auto 23do_eval: true 24early_stopping_patience: 2 25eval_batch_size: 4 26eval_max_new_tokens: 128 27eval_steps: 150 28eval_table_size: null 29evals_per_epoch: null 30flash_attention: true 31fp16: false 32fsdp: null 33fsdp_config: null 34gradient_accumulation_steps: 4 35gradient_checkpointing: true 36group_by_length: true 37hub_model_id: aleegis12/5118e892-01d0-47ca-9a51-1aef83696d99 38hub_private_repo: false 39hub_repo: null 40hub_strategy: checkpoint 41hub_token: null 42learning_rate: 0.0002 43load_in_4bit: false 44load_in_8bit: false 45local_rank: null 46logging_steps: 50 47lora_alpha: 128 48lora_dropout: 0.3 49lora_fan_in_fan_out: null 50lora_model_dir: null 51lora_r: 64 52lora_target_linear: true 53lr_scheduler: cosine 54max_grad_norm: 1.0 55max_memory: 56 0: 75GB 57max_steps: 3000 58micro_batch_size: 4 59mlflow_experiment_name: /tmp/824ef92913906b02_train_data.json 60model_type: AutoModelForCausalLM 61num_epochs: 10 62optim_args: 63 adam_beta1: 0.9 64 adam_beta2: 0.95 65 adam_epsilon: 1e-5 66optimizer: adamw_bnb_8bit 67output_dir: miner_id_24 68pad_to_sequence_len: true 69resume_from_checkpoint: null 70s2_attention: null 71sample_packing: false 72save_steps: 150 73saves_per_epoch: null 74sequence_len: 512 75strict: false 76tf32: true 77tokenizer_type: AutoTokenizer 78train_on_inputs: false 79trust_remote_code: true 80val_set_size: 0.05 81wandb_entity: null 82wandb_mode: online 83wandb_name: edbc27c0-2567-4ef1-849f-694fac00cd13 84wandb_project: Gradients-On-Demand 85wandb_run: your_name 86wandb_runid: edbc27c0-2567-4ef1-849f-694fac00cd13 87warmup_steps: 50 88weight_decay: 0.0 89xformers_attention: null 90

5118e892-01d0-47ca-9a51-1aef83696d99

• Loss: 0.2958

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: 4
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_BNB 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: 50
• training_steps: 3000

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