mhylle/gemma3-270m-mist-9liner-bilingual-gguf

Gemma3 270M - MIST 9-Liner Bilingual (GGUF)

This is a GGUF format export of a fine-tuned Gemma3 270M model for medical dictation extraction.

Model Description

• Base Model: google/gemma-3-270m-it
• Training Method: Full parameter fine-tuning
• Languages: English + German (bilingual)
• Task: Extract structured MIST 9-liner data from medical dictations
• Format: GGUF F16 (for use with llama.cpp)

Performance Metrics

Metric	Score
JSON Valid	99%
Schema Compliant	96%
Avg Field Accuracy	68.7%

Training Details

• Epochs: 3
• Batch Size: 8
• Learning Rate: 2e-5
• Max Sequence Length: 1024
• Training Time: ~28 minutes

Intended Use

This model is designed for on-device deployment to extract structured MIST 9-liner information from medical dictation text. The MIST 9-liner format includes:

1. Mechanism of injury
2. Injury pattern
3. Signs and symptoms
4. Treatment given
5. Vital signs (5 fields)

Usage with llama.cpp

# Download the GGUF file
huggingface-cli download mhylle/gemma3-270m-mist-9liner-bilingual-gguf gemma3-270m-mist-bilingual-f16.gguf

# Run inference
./llama-cli -m gemma3-270m-mist-bilingual-f16.gguf -p "<prompt>"

Limitations

• Optimized for MIST 9-liner format specifically
• Best performance with clear, structured dictation input
• Trained on synthetic/curated medical dictation data

Model Card Contact

For questions or issues, please open a GitHub issue on the project repository.