Skip to content

Model Registry

OpenMed ships a curated registry (openmed.core.model_registry.OPENMED_MODELS) that annotates every official checkpoint with metadata such as category, specialization, recommended confidence, and Hugging Face IDs. Use it to pick the right model, surface dropdowns in UIs, or validate incoming requests.

Exploring the registry

from openmed.core.model_registry import (
    get_all_models,
    list_model_categories,
    get_models_by_category,
    get_model_info,
    get_model_suggestions,
)

print(list_model_categories())

oncology_models = get_models_by_category("Oncology")
for info in oncology_models:
    print(info.display_name, info.model_id, info.recommended_confidence)

info = get_model_info("disease_detection_superclinical")
print(info.description, info.entity_types)

suggestions = get_model_suggestions("Metastatic breast cancer on paclitaxel.")
for model_key, info, reason in suggestions:
    print(model_key, info.display_name, reason)
  • ModelInfo objects include display_name, category, entity_types, size hints, and a default confidence threshold.
  • get_model_suggestions leans on lightweight heuristics to recommend models based on text snippets or hints (disease, pharma, oncology, etc.).

CLI parity

openmed models list              # registry + remote (HF) by default
openmed models list --include-remote
openmed models info disease_detection_superclinical

The CLI uses the same helpers but returns terminal-friendly tables/JSON, making it easy to integrate into scripts or CI.

Metadata for UIs & validation

  • ModelInfo.size_category and .size_mb help you decide whether a model can fit on CPU-only infrastructure.
  • entity_types feed dropdowns or filter chips in your frontend.
  • recommended_confidence can drive slider defaults or guardrails on API calls (pass it to analyze_text).

Keeping the registry fresh

If you add a new Hugging Face checkpoint, update openmed/core/model_registry.py with:

  1. A unique key (e.g., radiology_detection_superclinical).
  2. The full HF model id (OpenMed/...), user-friendly name, category, specialization.
  3. Representative entity_types and a recommended confidence based on evaluation runs.

CI will enforce type safety through the unit tests, and the docs automatically pick up the new entry via the examples above.