README.md

---
language: en
license: apache-2.0
tags:
  - medical
  - soap-notes
  - mistral
  - lora
  - fine-tuned
---

# 🏥 Mistral 7B Fine-Tuned for SOAP Note Generation

This model is a fine-tuned version of [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) 
specialized for generating clinical SOAP notes from doctor-patient conversations.

## Model Details
- **Base Model**: mistralai/Mistral-7B-v0.1
- **Fine-tuning Method**: LoRA (Low-Rank Adaptation)
- **LoRA Rank**: 16
- **Training**: 25 minutes on Google Colab T4 GPU
- **Output Format**: Structured JSON with Subjective, Objective, Assessment, Plan sections

## Evaluation Results (Groq Llama-3.3-70B Judge)
| Metric | Score |
|--------|-------|
| Answer Relevancy | 0.86 |
| Contextual Precision | 0.60 |
| Contextual Recall | 0.66 |
| Contextual Relevancy | 0.78 |
| Faithfulness | 0.70 |

## Usage

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

model = AutoModelForCausalLM.from_pretrained(
    "SaberaBanu/mistral-soap-notes",
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("SaberaBanu/mistral-soap-notes")

PROMPT = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
Generate a SOAP note from the clinical conversation. Output MUST be a valid JSON object.

### Input:
{conversation}

### Response:
"""

inputs = tokenizer(PROMPT.format(conversation=your_conversation), return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=600, do_sample=False)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Limitations
- Objective section may hallucinate vitals not mentioned in conversation
- Works best with clearly structured doctor-patient dialogues
- Not intended for real clinical use without human review