---
license: gemma
base_model: google/gemma-3-270m-it
tags:
- medical
- 9-liner
- medevac
- fine-tuned
- full-fine-tuning
language:
- en
datasets:
- custom
pipeline_tag: text-generation
---

# Gemma 3 270M - MIST 9-Liner Fine-tuned

This model is a full parameter fine-tuned version of [google/gemma-3-270m-it](https://huggingface.co/google/gemma-3-270m-it) trained on MIST 9-liner (medical evacuation request) data.

## Model Details

- **Base Model:** google/gemma-3-270m-it
- **Training Type:** Full parameter fine-tuning (not LoRA)
- **Parameters:** 268M (100% trainable)
- **Training Data:** 9,500 MIST 9-liner examples
- **Epochs:** 3
- **Final Loss:** 0.191
- **Token Accuracy:** 92.5%

## Training Configuration

- Learning Rate: 2e-5
- Batch Size: 8
- Gradient Accumulation: 2
- Max Sequence Length: 1024
- Optimizer: AdamW
- Precision: bfloat16

## Intended Use

This model is designed for parsing and generating MIST 9-liner medical evacuation requests. The 9-liner format is a standardized military medical evacuation request format.

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("mhylle/gemma3-270m-9liner")
tokenizer = AutoTokenizer.from_pretrained("mhylle/gemma3-270m-9liner")

# Example usage
prompt = "Convert this 9-liner to a medical record: ..."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0]))
```

## Limitations

- Trained specifically on 9-liner format data
- May not generalize well to other medical documentation formats
- Should be validated before use in real medical applications

## License

This model inherits the Gemma license from the base model.