| library_name: peft | |
| license: other | |
| base_model: google/gemma-3-4b-it | |
| tags: | |
| - llama-factory | |
| - lora | |
| - generated_from_trainer | |
| model-index: | |
| - name: gemma3-4b-it_eqa_lora_sft | |
| results: [] | |
| <!-- This model card has been generated automatically according to the information the Trainer had access to. You | |
| should probably proofread and complete it, then remove this comment. --> | |
| # gemma3-4b-it_eqa_lora_sft | |
| This model is a fine-tuned version of [google/gemma-3-4b-it](https://huggingface.co/google/gemma-3-4b-it) on the energy_qa_alpaca dataset. | |
| ## Model description | |
| More information needed | |
| ## Intended uses & limitations | |
| More information needed | |
| ## Training and evaluation data | |
| More information needed | |
| ## Training procedure | |
| ### Training hyperparameters | |
| The following hyperparameters were used during training: | |
| - learning_rate: 0.0001 | |
| - train_batch_size: 1 | |
| - eval_batch_size: 8 | |
| - seed: 42 | |
| - gradient_accumulation_steps: 8 | |
| - total_train_batch_size: 8 | |
| - optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments | |
| - lr_scheduler_type: cosine | |
| - lr_scheduler_warmup_ratio: 0.1 | |
| - num_epochs: 1.0 | |
| ### Training results | |
| ### Framework versions | |
| - PEFT 0.15.2 | |
| - Transformers 4.52.4 | |
| - Pytorch 2.7.0+cu126 | |
| - Datasets 3.6.0 | |
| - Tokenizers 0.21.1 |