| adapter: lora | |
| base_model: HuggingFaceTB/SmolLM2-135M | |
| bf16: auto | |
| dataset_prepared_path: ./last_run_prepared | |
| datasets: | |
| - path: mhenrichsen/alpaca_2k_test | |
| type: alpaca | |
| flash_attention: true | |
| gradient_accumulation_steps: 2 | |
| gradient_checkpointing: false | |
| learning_rate: 0.0003 | |
| logging_steps: 10 | |
| lora_alpha: 16 | |
| lora_dropout: 0.05 | |
| lora_r: 8 | |
| lora_target_modules: | |
| - q_proj | |
| - v_proj | |
| max_steps: 100 | |
| micro_batch_size: 2 | |
| optimizer: adamw_torch | |
| output_dir: ./outputs/admin_20251203_195913 | |
| saves_per_epoch: 1 | |
| sequence_len: 256 | |
| special_tokens: | |
| pad_token: <|endoftext|> | |
| val_set_size: 0.05 | |