huggingface

README

Yet another Go package to download from HuggingFace Hub

Features

• Uses the same cache as the Hugging Face official python client for both authentication token and model files.
• Parallel download.

See whole documentation at

Documentation

Overview

Package huggingface is the best library to fetch files from an huggingface repository.

Index

• func AuthRequest(ctx context.Context, h *http.Client, method, url, token string, ...) (*http.Response, error)
• type Client
  • func New(token string) (*Client, error)
  • func (c *Client) EnsureFile(ctx context.Context, ref ModelRef, revision, file string) (string, error)
  • func (c *Client) EnsureSnapshot(ctx context.Context, ref ModelRef, revision string, glob []string) ([]string, error)
  • func (c *Client) GetFileInfo(ctx context.Context, ref ModelRef, revision, file string) (string, string, int64, error)
  • func (c *Client) GetModelInfo(ctx context.Context, m *Model, ref string) error
• type Model
• type ModelRef
  • func (m *ModelRef) RepoID() string
  • func (m *ModelRef) URL() string

Functions

func AuthRequest

func AuthRequest(ctx context.Context, h *http.Client, method, url, token string, hdr map[string]string) (*http.Response, error)

AuthRequest does an authenticated HTTP request with a Bearer token, which retries automatically 429 and 5xx.

Method must be HEAD or GET.

Types

type Client

type Client struct {
	// contains filtered or unexported fields
}

Client is the client for https://huggingface.co/.

func New

func New(token string) (*Client, error)

New returns a new *Client client to download files and list repositories.

It uses the endpoints as described at https://huggingface.co/docs/hub/api.

Respects the following environment variables described at https://huggingface.co/docs/huggingface_hub/package_reference/environment_variables: HF_HOME, HF_HUB_CACHE, HF_TOKEN_PATh and HF_TOKEN.

func (*Client) EnsureFile

func (c *Client) EnsureFile(ctx context.Context, ref ModelRef, revision, file string) (string, error)

EnsureFile ensures the file is available, downloads it otherwise.

Similar to https://huggingface.co/docs/huggingface_hub/package_reference/file_download

func (*Client) EnsureSnapshot

func (c *Client) EnsureSnapshot(ctx context.Context, ref ModelRef, revision string, glob []string) ([]string, error)

EnsureSnapshot ensures files available from the snapshot, downloads them otherwise.

Downloads files concurrently.

Similar to https://huggingface.co/docs/huggingface_hub/package_reference/file_download#huggingface_hub.snapshot_download

func (*Client) GetFileInfo

func (c *Client) GetFileInfo(ctx context.Context, ref ModelRef, revision, file string) (string, string, int64, error)

GetFileInfo retrieves the information about the file.

Returns the commitish, etag, size.

func (*Client) GetModelInfo

func (c *Client) GetModelInfo(ctx context.Context, m *Model, ref string) error

GetModelInfo fills the supplied Model with information from the HuggingFace Hub.

Use "main" as ref unless you need a specific commit.

type Model

type Model struct {
	ModelRef
	// Upstream is the upstream repo when the model is based on another one.
	Upstream ModelRef

	// Tensor is the native quantization of the weight. This is found in
	// config.json in Upstream.
	TensorType safetensors.DType
	// Number of weights. Has direct impact on performance and memory usage.
	NumWeights int64
	// ContentLength is the number of tokens that the LLM can take as context
	// when relevant. Has impact on performance and memory usage. Not relevant
	// for image generators.
	ContextLength int
	// License is the license of the weights, for whatever that means. Use the
	// name for well known licences (e.g. "Apache v2.0" or "MIT") or an URL for
	// custom licenses.
	License string
	// LicenseURL is the URL to the license file.
	LicenseURL string
	// Files is the list of files in the repository.
	Files []string
	// Created is the time the repository was created. It can be at the earliest
	// 2022-03-02 as documented at
	// https://huggingface.co/docs/hub/api#repo-listing-api.
	Created time.Time
	// Modified is the last time the repository was modified.
	Modified time.Time
	// SHA of the reference requested.
	SHA string
	// contains filtered or unexported fields
}

Model is a model stored on https://huggingface.co

type ModelRef

type ModelRef struct {
	// Author is the owner, either a person or an organization.
	Author string
	// Repo is the name of the repository owned by the Author.
	Repo string
	// contains filtered or unexported fields
}

ModelRef is a reference to a model stored on https://huggingface.co

func (*ModelRef) RepoID

func (m *ModelRef) RepoID() string

RepoID is a shorthand to return .m.Author + "/" + m.Repo

func (*ModelRef) URL

func (m *ModelRef) URL() string

URL returns the Model's canonical URL.