See axolotl config

axolotl version: 0.4.1

yaml

1adapter: lora 2base_model: unsloth/SmolLM-360M 3bf16: auto 4chat_template: llama3 5dataloader_num_workers: 6 6dataset_prepared_path: null 7datasets: 8- data_files: 9 - ee17b4fb2e938020_train_data.json 10 ds_type: json 11 format: custom 12 path: /workspace/input_data/ee17b4fb2e938020_train_data.json 13 type: 14 field_instruction: instruction 15 field_output: golden_answer 16 format: '{instruction}' 17 no_input_format: '{instruction}' 18 system_format: '{system}' 19 system_prompt: '' 20debug: null 21deepspeed: null 22early_stopping: 23 metric: eval_loss 24 mode: min 25 patience: 3 26eval_max_new_tokens: 128 27eval_steps: 20 28eval_table_size: null 29flash_attention: true 30fp16: null 31fsdp: null 32fsdp_config: null 33gradient_accumulation_steps: 8 34gradient_checkpointing: false 35group_by_length: true 36hub_model_id: error577/6c790710-3b99-471c-b257-247f79be4895 37hub_repo: null 38hub_strategy: checkpoint 39hub_token: null 40learning_rate: 0.0002 41load_in_4bit: false 42load_in_8bit: true 43local_rank: null 44logging_steps: 1 45lora_alpha: 16 46lora_dropout: 0.3 47lora_fan_in_fan_out: null 48lora_model_dir: null 49lora_r: 8 50lora_target_linear: true 51lr_scheduler: cosine 52max_grad_norm: 1.0 53max_steps: 500 54micro_batch_size: 2 55mlflow_experiment_name: /tmp/ee17b4fb2e938020_train_data.json 56model_type: AutoModelForCausalLM 57num_epochs: 4 58optimizer: adamw_bnb_8bit 59output_dir: miner_id_24 60pad_to_sequence_len: true 61resume_from_checkpoint: null 62s2_attention: null 63sample_packing: false 64save_steps: 20 65sequence_len: 512 66strict: false 67tf32: false 68tokenizer_type: AutoTokenizer 69train_on_inputs: false 70trust_remote_code: true 71val_set_size: 0.05 72wandb_entity: null 73wandb_mode: online 74wandb_name: 0555365f-c421-44d7-895c-918401dea66a 75wandb_project: Gradients-On-Demand 76wandb_run: your_name 77wandb_runid: 0555365f-c421-44d7-895c-918401dea66a 78warmup_steps: 10 79weight_decay: 0.01 80xformers_attention: null 81

6c790710-3b99-471c-b257-247f79be4895

• Loss: 1.5436

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• 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: 10
• training_steps: 500

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