Instructions to use oplus7/Qwen2.5-3B-ps-LoRA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use oplus7/Qwen2.5-3B-ps-LoRA with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-3B-Instruct") model = PeftModel.from_pretrained(base_model, "oplus7/Qwen2.5-3B-ps-LoRA") - Notebooks
- Google Colab
- Kaggle
LoRA-ps-helper
基于 Qwen2.5-3B-Instruct 微调的 PowerShell 指令助手(LoRA 适配器)。
模型信息
- 基底模型:
Qwen/Qwen2.5-3B-Instruct - 微调方式:LoRA(
r=32,alpha=64,dropout=0.05, target=全部 proj 层),PEFT 0.20.0 - 任务:PowerShell 命令讲解、按需求生成命令、Win+R 指令讲解、命令纠错
使用
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-3B-Instruct")
model = PeftModel.from_pretrained(model, "Oplus7/LoRA-ps-helper")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-3B-Instruct")
量化版 GGUF:见 Oplus7/LoRA-ps-helper-GGUF。
评测结果
微调模型 vs 基座(qwen2.5:3b),30 条测试集,temperature=0:
| 指标 | 微调 | 基座 | 差值 |
|---|---|---|---|
| 命令/参数覆盖率 | 65.3% | 54.2% | +11.1pt |
| ROUGE-L | 46.4% | 28.5% | +17.9pt |
| 逐条胜负 | 29 胜 | 1 负 | — |
训练数据与评测脚本见 GitHub 仓库 Oplus7/LoRA-ps-helper。
许可证
Apache-2.0(跟随基底模型 Qwen2.5)。
- Downloads last month
- 6