Crate dspk_synapse_ml 

dspk-synapse-ml: pretrained neuro models from the literature, each traceable to its authors. The spike sorters live in their own crates, one per upstream project, each under its upstream’s licence (dspk-sorter-kilosort, dspk-sorter-ms5, dspk-sorter-si).

• models: pretrained networks run as released (parked until their artifacts are verified).
• manifest: how to run each published model (format, tensors, provenance); with feature hub its file is downloaded through the dspk-hub catalog.
• provenance: paper, code, license and download source of every sorter and model (from dspk-synapse).
• runtime: tensor execution for the pretrained models.

Re-exports

pub use manifest::ArraySpec;

pub use manifest::ModelCatalog;

pub use manifest::ModelFormat;

pub use manifest::ModelManifest;

pub use manifest::SafetensorEntryHeader;

pub use manifest::SafetensorsMap;

pub use manifest::TensorIoSpec;

pub use manifest::TensorPortSpec;

pub use models::DARTSORT_DENOISER_MODEL_ID;

pub use models::DARTSORT_VAE_MODEL_ID;

pub use models::DartsortVaeEmbedder;

pub use models::DartsortWaveformDenoiser;

pub use models::SPIKENET2_IED_MODEL_ID;

pub use models::SpikeNet2Detector;

pub use models::UNITREFINE_CURATION_MODEL_ID;

pub use models::UnitCurationResult;

pub use models::UnitRefineClassifier;

pub use runtime::OnnxRuntimeSession;

pub use runtime::RuntimeTensor;

pub use runtime::burn_conv1d;

pub use runtime::burn_linear_2d;

pub use runtime::validate_tensor_port;

Modules

manifest

Model manifests of dspk-synapse-ml: the catalog (catalog/models.json: what each published model is, its tensors, whom to credit), tensor specifications and the safetensors reader. Downloading a model’s file (with feature hub) goes through the dspk-hub catalog entry of the same id ([crate::runtime::pull_model]).

models

Pretrained networks run as their authors released them (artifacts through the catalog, crate::manifest). None has a validated artifact yet: their catalog entries are parked until verified (see .refactoring/dspk-synapse-ml/). The spike sorters live in their own crates (synapse/sorters/).

provenance

Where every sorter and model comes from: the paper to cite, the upstream code and its license, and the files downloaded to run it. Exposed by every sorter and model through Attributed, and stored in the catalog, so a result can always be traced to its authors.

runtime

Tensor execution for the pretrained models: Burn operations and an ONNX (onnx-ir) graph evaluator, on a dspk_core::ComputeTarget the caller chooses. To be replaced by burn-onnx with device-resident weights once a model artifact is validated (review RT1–RT3, ONNX1–4).

Structs

ArtifactSource

A file downloaded to run the sorter or model.

Paper

The publication to cite.

Provenance

Provenance of one sorter or model. See the module docs.

UpstreamCode

The upstream source code.

Enums

ProvenanceKind

How a sorter or model relates to its upstream.

Traits

Attributed

Anything whose origin can be cited.