JacobLinCool's picture
Train runtime tool-call MiniCPM5 LoRA adapter
25de69b verified
|
Raw
History Blame Contribute Delete
468 Bytes
# Hackathon Advisor MiniCPM5 LoRA
This PEFT LoRA adapter was trained for The Unwritten Almanac's runtime tool-call planner. It maps `system_prompt + render_context` chats to one XML function call.
## Recipe
- Base model: `openbmb/MiniCPM5-1B`
- Adapter repo: `build-small-hackathon/hackathon-advisor-minicpm5-lora`
- Task: runtime XML tool-call planning
- Method: LoRA SFT
- Examples: 9
- Max steps: 120
- LoRA rank: 16
- LoRA alpha: 32
- Publish status: published