---
title: DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA
canonical_url: "https://www.modelscope.cn/models/Tierney/DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA"
md_url: "https://www.modelscope.cn/models/Tierney/DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA.md"
repository: Tierney/DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA
chinese_name: "基于DS14B模型的医疗微调LoRA"
last_updated: 2025-03-05
parameters: 68.8M
tensor_type:
  - F32
library_name:
  - transformer
  - safetensors
  - pytorch
frameworks:
  - Pytorch
downloads: 43
stars: 1
---

# DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA

> DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA - Tierney 在 ModelScope 开源的模型。DeepSeek-R1-Distill-Qwen-14B LoRA Adapter

- **Repository**: Tierney/DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA
- **Parameters**: 68.8M
- **Downloads**: 43
- **Stars**: 1
- **Last updated**: 2025-03-05

Source: https://www.modelscope.cn/models/Tierney/DeepSeek-Qwen-14B-R1-Medical-Doctor-LoRA

---

# DeepSeek-R1-Distill-Qwen-14B LoRA Adapter

## 📌 模型简介
本 LoRA 适配器基于 **DeepSeek-R1-Distill-Qwen-14B** 进行微调，主要优化医学领域的问答和推理能力。

- 🔹 **基座模型**: [deepseek-ai/DeepSeek-R1-Distill-Qwen-14B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B)
- 🔹 **微调方法**: LoRA（使用 [Unsloth](https://github.com/unslothai/unsloth) 进行优化）
- 🔹 **适用场景**: 医学文本问答、医学知识增强

---

## 📂 使用方法

### 🔄 加载 LoRA 适配器
要使用本 LoRA 适配器，你需要加载原始 DeepSeek-R1-14B 模型，并应用 LoRA 权重：
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B"
lora_model = "your-huggingface-username/DeepSeek-R1-Distill-Qwen-14B-lora-med"

tokenizer = AutoTokenizer.from_pretrained(base_model)
model = AutoModelForCausalLM.from_pretrained(base_model, torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(model, lora_model)
```

### 🚀 推理示例
```python
input_text = "请问阿司匹林的主要适应症是什么？"
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_length=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

---

## 🏗️ 训练信息
- **训练环境**: RTX 4090, CUDA 12.6, WSL Ubuntu
- **训练框架**: `transformers` + `peft` + `unsloth`
- **训练参数**:
  - LoRA Rank: 16
  - Alpha: 32
  - Dropout: 0.05
  - Max Seq Length: 4096

---

## 📜 许可证
本 LoRA 适配器基于 **DeepSeek-R1-Distill-Qwen-14B**，请遵守其[官方许可证](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B)。

---

## 📞 联系方式
如果你有任何问题或建议，可以在讨论区留言，或者联系我！