Gemma3-270M-EU-Medical-CandidateSelector-GGUF

Gemma3-270M-EU-Medical-CandidateSelector-GGUF

GGUF exports of Gemma3-270M-EU-Medical-CandidateSelector-LoRA.

This is a merged GGUF model for local/edge inference with llama.cpp or Ollama-compatible runtimes.

Files

• Q8_0/gemma3-270m-candidate-selector-Q8_0.gguf
• Q4_K_M/gemma3-270m-candidate-selector-Q4_K_M.gguf

An F16 GGUF may also be included under F16/ for archival/reference use.

Intended use

Candidate-conditioned German/EU medical terminology selection.

The model should receive deterministic candidate lists and select, reject, or flag ambiguity.

It is not a diagnostic model and must not be used for treatment, prescribing, dosage, triage, or clinical decision-making.

Full held-out test results from LoRA model

Evaluated on 8,311 rows:

• JSON parse rate: 0.9998
• ICD selected-code exact rate: 0.9956
• Selected-from-candidate-set rate: 1.0000
• Invalid out-of-candidate code rate: 0.0000
• ICD no-match null rate: 1.0000
• ICD ambiguity code exact rate: 0.9970
• EMA medicine selection exact rate: 1.0000
• EMA active-substance exact rate: 0.9702
• Safety refusal rate: 1.0000

Production rule

Use this model only as a selector. Final facts must be verified from SQLite/EMA source tables.