Module model_files 

HuggingFace Hub download + cache-layout helpers for the BGE-M3 model files.

Structs

ModelFiles 🔒

Paths to the ONNX model and tokenizer files resolved from the hf-hub cache.

Constants

REPO_ID 🔒

REPO_REVISION 🔒

Pinned HF commit — prevents silent model updates and provides supply-chain integrity for the ONNX weights and tokenizer. Update this hash intentionally after verifying a new revision produces equivalent embeddings.

XENOVA_REPO_ID 🔒

XENOVA_REPO_REVISION 🔒

Pinned HF commit for the Xenova/bge-m3 FP16 (~1.08 GB) and INT8 (~568 MB) models. Update intentionally after verifying equivalent embedding quality vs FP32.

Functions

download_model_files 🔒

Downloads (or retrieves from the local hf-hub snapshot cache) the ONNX model and tokenizer files for the given model variant.

is_model_cached 🔒

Returns true when the primary ONNX model file already exists in the hf-hub snapshot cache, meaning repo.get() will return immediately without fetching from the network.