Aggregates the ML models produced by codet5-models-builder and minilm-builder into a single consumable package, so downstream tools only need one workspace dep to pull in both the CodeT5 code embedder and the MiniLM sentence embedder.
The model definitions live in package.json under the moduleSources field (not sources — these are HuggingFace models, not git upstreams).
pnpm --filter models run build # build both (minilm + codet5)
pnpm --filter models run build:minilm # minilm only
pnpm --filter models run build:codet5 # codet5 onlymodels delegates to the individual builder packages; their prereqs apply (Python venv with the pinned pip packages — see their READMEs).
Output: build/<mode>/<platform-arch>/out/Final/{minilm-l6,codet5}/model.onnx (each model directory also contains tokenizer files).