LoRA Fine-Tune

Using LoRA to fine-tune LLaMA-3-8B for [Mask] guessing task.

Dataset

Mask Wiki Documents from TriviaQA_Test (165052 Samples after filting)

Dataset Generation Prompt

{
  "instruction": "Your Task is to GUESS the complete sentence to fill in the [Mask] section of the Wiki Document.",
  "input": wiki_title + '\n' + masked_wiki_document,
  "output": mask_sentence,
}

References

LLaMA3-8B-Instruct Lora 微调

深入浅出 LoRA