📌 中文简介：Qwen-7B LoRA 微调模型（中文指令微调）

本模型基于阿里巴巴通义千问 Qwen-7B-Chat，采用 LoRA 技术，使用 Alpaca-Zh-51k 数据集进行了中文指令微调，适用于中文任务的理解与生成。

注：对Chat进行微调后效果反而变差了，或许对base版本微调会好一些

🧾 模型信息

基座模型：Qwen/Qwen-7B-Chat
微调方法：LoRA（使用 PEFT 库）
训练数据集：Alpaca-Zh-51k
训练脚本：train_qwen7b_lora.py
推理脚本：test_compare.py
⚠️ 本模型仅包含 LoRA adapter，不包含原始基座权重

🚀 使用示例

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

model_name = "Josh1207/qwen7b-alpaca-lora"

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen-7B-Chat", trust_remote_code=True)
model = PeftModel.from_pretrained(base_model, model_name)

prompt = "指令: 请介绍一下你自己。"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📌 English Overview: Qwen-7B LoRA Fine-tuned Model (Chinese Instruction Tuning)

This model is fine-tuned from Alibaba’s Qwen-7B-Chat using LoRA technique on the Alpaca-Zh-51k dataset. It is suitable for instruction-following tasks in Chinese.

(I found that after making adjustments to Chat model, the effect actually got worse. Perhaps making adjustments to the base version would be better)

🧾 Model Information

Base model: Qwen/Qwen-7B-Chat
Tuning method: LoRA (via peft)
Dataset: Alpaca-Zh-51k
Training script: train_qwen7b_lora.py
Inference script: test_compare.py
⚠️ This repository includes only LoRA adapter weights, not the original base model.

🚀 Usage Example

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

model_name = "Josh1207/qwen7b-alpaca-lora"

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen-7B-Chat", trust_remote_code=True)
model = PeftModel.from_pretrained(base_model, model_name)

prompt = "指令: 请介绍一下你自己。"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))