furiosa.registry package

Subpackages

furiosa.registry.client package
- Subpackages
  - furiosa.registry.client.transport package
- Module contents

Submodules

furiosa.registry.errors module

exception furiosa.registry.errors.RegistryError(msg): Bases: Exception

exception furiosa.registry.errors.TransportNotFound(uri: str, transports: List[Transport]): Bases: furiosa.registry.errors.RegistryError

furiosa.registry.model module

class furiosa.registry.model.Config

Bases: pydantic.config.BaseConfig

extra: pydantic.config.Extra = 'forbid'

class furiosa.registry.model.Format(value)

Bases: str, enum.Enum

Model binary format to represent the binary specified.

ONNX = 'onnx'

TFLite = 'tflite'

class furiosa.registry.model.Metadata(*, description: Optional[str] = None, publication: Optional[furiosa.registry.model.Publication] = None)

Bases: pydantic.main.BaseModel

Model metadata to understand a model.

description: Optional[str]

publication: Optional[furiosa.registry.model.Publication]

class furiosa.registry.model.Model(*, name: str, source: bytes, format: furiosa.registry.model.Format, dfg: Optional[bytes] = None, enf: Optional[bytes] = None, family: Optional[str] = None, version: Optional[str] = None, metadata: Optional[furiosa.registry.model.Metadata] = None, inputs: Optional[List[furiosa.registry.model.ModelTensor]] = [], outputs: Optional[List[furiosa.registry.model.ModelTensor]] = [], compiler_config: Optional[Dict] = None)

Bases: pydantic.main.BaseModel

Represent the artifacts and metadata of a neural network model

name: a name of this model

format: the binary format type of model source; e.g., ONNX, tflite

source: a source binary in ONNX or tflite. It can be used for compiling this model with a custom compiler configuration.

dfg: an intermediate representation of furiosa-compiler. Native post processor implementation uses dfg binary. Users don’t need to use dfg directly.

enf: the executable binary for furiosa runtime and NPU

version: model version

inputs: data type and shape of input tensors

outputs: data type and shape of output tensors

compiler_config: a pre-defined compiler option

compiler_config: Optional[Dict]

dfg: Optional[bytes]

enf: Optional[bytes]

family: Optional[str]

format: furiosa.registry.model.Format

inputs: Optional[List[furiosa.registry.model.ModelTensor]]

metadata: Optional[furiosa.registry.model.Metadata]

name: str

outputs: Optional[List[furiosa.registry.model.ModelTensor]]

source: bytes

version: Optional[str]

class furiosa.registry.model.ModelTensor(*, name: str, datatype: str, shape: List[int], tags: Optional[furiosa.registry.model.Tags] = None)

Bases: pydantic.main.BaseModel

datatype: str

name: str

shape: List[int]

tags: Optional[furiosa.registry.model.Tags]

class furiosa.registry.model.Publication(*, authors: Optional[List[str]] = None, title: Optional[str] = None, publisher: Optional[str] = None, date: Optional[datetime.date] = None, url: Optional[str] = None)

Bases: pydantic.main.BaseModel

Model publication information.

authors: Optional[List[str]]

date: Optional[datetime.date]

publisher: Optional[str]

title: Optional[str]

url: Optional[str]

class furiosa.registry.model.Tags(*, content_type: Optional[str] = None, **extra_data: Any)

Bases: pydantic.main.BaseModel

class Config

Bases: object

extra = 'allow'

content_type: Optional[str]

furiosa.registry.utils module

furiosa.registry.utils.import_module(directory: str, path: str) → module: Import module via specified path from the local directory.

furiosa.registry.utils.python_path(directory: str) → Iterator[None]: Context adding the directory into PYTHONPATH.

furiosa.registry.utils.removeprefix(word: str, prefix: str) → str

Python 3.9 removeprefix().

See https://docs.python.org/3/library/stdtypes.html#str.removeprefix

furiosa.registry.utils.working_directory(directory: str) → Iterator[None]: Context replacing current working directory.

Module contents

FuriosaAI model registry