README.md

---
license: mit
---

# 🧠 AlphaMed

This is the official model checkpoint for the paper:  
**[AlphaMed: Incentivizing Medical Reasoning with minimalist Rule-Based RL](https://www.arxiv.org/abs/2505.17952)**  
AlphaMed is a medical large language model trained **without supervised fine-tuning on chain-of-thought (CoT) data**, 
relying solely on reinforcement learning to elicit step-by-step reasoning in complex medical tasks.

## 🚀 Usage

To use the model, format your input prompt as:

> **Question:** [your medical question here]  
> **Please reason step by step, and put the final answer in \boxed{}**

### 🔬 Example

```python
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# Load model and tokenizer
model_id = "che111/AlphaMed-3B-instruct-rl"  # Replace with actual repo path
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Format question
prompt = (
    "Question: A 45-year-old patient presents with chest pain radiating to the left arm and elevated troponin levels. "
    "What is the most likely diagnosis?\n"
    "Please reason step by step, and put the final answer in \\boxed{}"
)

# Generate output
max_new_tokens=8196
output = pipe(prompt, max_new_tokens=max_new_tokens, do_sample=False)[0]["generated_text"]
print(output)