WizardLM

• Use LLaMA 1 7B
• 250K instructions

LLaMA

LLaMA2

• Using smaller LR for bigger models

• LLaMA-chat

  • SFT

    • 27,540 annotations

    

    

  • RLHF