Module resources

Expand description

§Resource definitions for model weights, vocabularies and configuration files

This crate relies on the concept of Resources to access the data used by the models. This includes:

model weights
configuration files
vocabularies
(optional) merges files for BPE-based tokenizers

These are expected in the pipelines configurations or are used as utilities to reference to the resource location. Two types of resources are pre-defined:

LocalResource: points to a local file
RemoteResource: points to a remote file via a URL
BufferResource: refers to a buffer that contains file contents for a resource (currently only usable for weights)

For LocalResource and RemoteResource, the local location of the file can be retrieved using get_local_path, allowing to reference the resource file location regardless if it is a remote or local resource. Default implementations for a number of RemoteResources are available as pre-trained models in each model module.

Structs§

BufferResource: In-memory raw buffer resource
LocalResource: Local resource
RemoteResource: Remote resource that will be downloaded and cached locally on demand

Enums§

Resource

Traits§

ResourceProvider: Resource Trait that can provide the location or data for the model, and location of

Functions§