---
library_name: transformers
tags:
  - medical
  - spanish
  - llm
  - fine-tuning
  - lora
  - qlora
  - oswestry
  - functional-assessment
  - kaggle
  - mistral
---

# Model Card for oswestry-mistral-finetuned

This is a fine-tuned version of the `Mistral-7B-Instruct-v0.2` model specialized in scoring functional disability interviews using the Oswestry Disability Index (ODI) scale in Spanish. The model is designed to transform clinical-style interview transcripts into structured scores, demonstrating improved performance compared to the base model.

---

## Model Details

### Model Description

- **Developed by:** [Alejandro M.L]
- **Model type:** Causal Decoder-Only Transformer (LLM)
- **Language(s):** Spanish (with clinical vocabulary)
- **License:** apache-2.0 
- **Fine-tuned from model:** mistral-7b-instruct-v0.2

### Model Sources

- **Repository:** [Github](https://github.com/ALM-Bloom/TFG_Alejandro)
- **Paper [Transformación de Informes Médicos en escalas funcionales]:** 

---

## Uses

### Direct Use

The model takes as input a clinical interview transcript in Spanish (following a structured instruction format) and returns a text output containing the scores of each item in the Oswestry scale.

### Downstream Use

Can be integrated into tools that support:
- Preliminary functional assessment in telemedicine
- Research pipelines for NLP in healthcare
- Spanish-language LLM benchmarking on medical tasks

### Out-of-Scope Use

- Not suitable for general-purpose chat applications
- Should not be used for real clinical decisions without expert supervision
- Not intended for languages other than Spanish

---

## Bias, Risks, and Limitations

- The model was fine-tuned on synthetic data, which may limit generalizability.
- Outputs might include hallucinations if the input format is not followed.
- It reflects the biases of the base model and the prompt structure used.

### Recommendations

- Use only for research and non-critical applications.
- Always validate outputs against clinical judgment.
- Further training with real anonymized clinical data is highly recommended.

---

## How to Get Started with the Model

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("DrAleML/Oswestry-Instruct")
tokenizer = AutoTokenizer.from_pretrained("DrAleML/Oswestry-Instruct")

prompt = "Entrevista:\nPaciente refiere dolor lumbar que aumenta al estar de pie...\n\nResponde con puntuación Oswestry:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=300)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))