Expand description
Common blocks for generic pipelines (e.g. token classification or sequence classification)
Provides Enums holding configuration or tokenization resources that can be used to create generic pipelines. The model component is defined in the generic pipeline itself as the pre-processing, forward pass and postprocessing differs between pipelines while basic config and tokenization objects don’t.
Structs
- Container for ONNX model resources, containing 3 optional resources (Encoder, Decoder and Decoder with past)
Enums
- Abstraction that holds a model configuration, can be of any of the supported models
- Variants to store either a Torch model resource or ONNX resources
- Identifies the type of model
- Abstraction that holds a particular tokenizer, can be of any of the supported models