A fine-tuned variant of Gemma-3-1B-IT optimized for structured medical data extraction from prescription-style doctor dialogues.
This model outputs a valid JSON object containing:
- medicines with dosages, frequencies, and timing
- diseases and symptoms
- diagnostic tests and their timings
- lifestyle or follow-up instructions
Fine-tuned using the Unsloth framework on a NVIDIA GeForce RTX 3050 (6GB VRAM) with LoRA (Rank=8, α=16, dropout=0.05).
Dataset: Synthetic Indian-style medical conversations generated using Claude 3.5 Sonnet, validated for JSON structural accuracy.
Stop token: `AAA`
