LoRA fine-tuned Gemma 4 E2B-it for fhir4px naming conventions.
This is the merged safetensors format. ONNX conversion is pending due to tooling
version conflicts (Unsloth requires transformers 5.5.0, but ONNX export needs
transformers 5.6+ for native Gemma 4 support).
Use HF's online Optimum converter or wait for clean conversion tooling:
https://huggingface.co/spaces/optimum/transformers-to-onnx
