Module registry 

The static table of known deep-learning QSM models.

Weights are hosted externally and fetched on use (see super); nothing is vendored here. Entries marked WeightStatus::Pending are recognized targets whose ONNX weights are not yet converted and hosted — their url and sha256 are filled in once the exported .onnx is uploaded (Hugging Face). Until then a host can still run them via bring-your-own-weights ($QSM_MODEL_DIR, or super::provide_weights).

Editing checklist when a model goes from Pending → Available:

1. export/convert to ONNX, upload the file,
2. set url, sha256 (lowercase hex), bytes,
3. flip status to WeightStatus::Available,
4. confirm inputs/outputs/size_divisor match the exported graph.

Constants

MODELS 🔒

Functions

all_models

All models known to QSM-Core, in a stable order.

find_model

Look up a model by its ModelSpec::id (case-sensitive), e.g. "qsmnet".

mem 🔒

A NetworkMemory, in the order its fields are listed: resident, load peak, inference fixed bytes and bytes per input voxel (native), WASM fixed bytes and bytes per input voxel. The figures come from examples/inference_memory.rs and the browser_memory::net_* cases in tests/wasm_pipeline.rs (see the module docs).

pending_onnx 🔒