|
|
--- |
|
|
base_model: deepseek-ai/DeepSeek-R1-Distill-Llama-8B |
|
|
license: mit |
|
|
pipeline_tag: question-answering |
|
|
tags: |
|
|
- medical |
|
|
- chain-of-thought |
|
|
- fine-tuning |
|
|
- qlora |
|
|
- unsloth |
|
|
--- |
|
|
|
|
|
# DeepSeek-R1-Medical-CoT |
|
|
|
|
|
🚀 **Fine-tuning DeepSeek R1 for Medical Chain-of-Thought Reasoning** |
|
|
|
|
|
This model is a fine-tuned version of `deepseek-ai/DeepSeek-R1-Distill-Llama-8B`, specifically designed to enhance medical reasoning through **Chain-of-Thought (CoT) prompting**. It is trained using **QLoRA** with **Unsloth optimization**, allowing efficient fine-tuning on limited hardware resources. |
|
|
|
|
|
--- |
|
|
|
|
|
## 📌 Model Details |
|
|
|
|
|
### **Model Description** |
|
|
- **Developed by:** [Your Name or Organization] |
|
|
- **Fine-tuned from:** `deepseek-ai/DeepSeek-R1-Distill-Llama-8B` |
|
|
- **Language(s):** English, with a focus on medical terminology |
|
|
- **Training Data:** Medical reasoning dataset (`medical-o1-reasoning-SFT`) |
|
|
- **Fine-tuning Method:** QLoRA (4-bit adapters), later merged into 16-bit weights |
|
|
- **Optimization:** Unsloth (2x faster fine-tuning with lower memory usage) |
|
|
|
|
|
### **Model Sources** |
|
|
- **Repository:** [Your Hugging Face Model Repo URL] |
|
|
- **Paper (if applicable):** [Link] |
|
|
- **Demo (if applicable):** [Link] |
|
|
|
|
|
--- |
|
|
|
|
|
## 🛠 **How to Use the Model** |
|
|
|
|
|
### 1️⃣ Load the Model |
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
|
|
repo_name = "your-huggingface-username/DeepSeek-R1-Medical-CoT" |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(repo_name) |
|
|
model = AutoModelForCausalLM.from_pretrained(repo_name) |
|
|
|
|
|
model.eval() |
|
|
``` |
|
|
### 2️⃣ Run inference |
|
|
|
|
|
```python |
|
|
prompt = "What are the early symptoms of diabetes?" |
|
|
inputs = tokenizer(prompt, return_tensors="pt").to(model.device) |
|
|
|
|
|
with torch.no_grad(): |
|
|
output = model.generate(**inputs, max_new_tokens=200) |
|
|
|
|
|
response = tokenizer.decode(output[0], skip_special_tokens=True) |
|
|
print("Model Response:", response) |
|
|
``` |
|
|
### 📢 Acknowledgments |
|
|
- **DeepSeek-AI for releasing DeepSeek-R1** |
|
|
- **Unsloth for optimized LoRA fine-tuning** |
|
|
- **Hugging Face for hosting the models** |
