Text Generation
PEFT
Safetensors
Chinese
English
biomedical
proteomics
question-generation
lora
conversational
Instructions to use shikunpunk/ask-dao with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use shikunpunk/ask-dao with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("E:\\生成诗歌\\poetry-gen-train\\models\\Qwen2.5-3B-Instruct") model = PeftModel.from_pretrained(base_model, "shikunpunk/ask-dao") - Notebooks
- Google Colab
- Kaggle
Ask-Dao v0.2 (已修正 key 嵌套)
⚠️ 重要:本仓库的 adapter 历史上因
PeftModel.from_pretrained双重包装导致 key 嵌套错误。 旧版本里PeftModel.from_pretrained加载会"Found missing adapter keys"并静默丢弃全部权重, 模型实际退化为 base Qwen2.5-3B-Instruct。 现在本仓库的adapter_model.safetensors已重新打包为标准深度 8(504/504 keys 命中),可被正确加载。 验证方式:见shikunpunk/ask-dao-v0.3仓库里 "Holdout 泛化对比" 一节。
推荐使用最新版本:
👉 shikunpunk/ask-dao-v0.3 —— 从本版本续训 + judge-filtered 数据,正向迭代。
本版本训练
- 数据:154 条单句化问题(蛋白质组学 / 液体活检)
- 基础:
Qwen/Qwen2.5-3B-Instruct(4-bit QLoRA, r=16, alpha=32, all-linear) - 3 epochs,max_len=2048,assistant-only loss
- Downloads last month
- 16