go huggingface

go-huggingface, download, tokenize and convert models from HuggingFace.

Overview

Simple APIs for downloading (hub), tokenizing (tokenizers) and (future work) model conversion (models) of HuggingFace? models using GoMLX.

EXPERIMENTAL and IN DEVELOPMENT: While the hub package has been stable, the tokenizers and the future models are still under intense development.

Examples

Preamble: Imports And Variables

import ("github.com/gomlx/go-huggingface/hub""github.com/gomlx/go-huggingface/tokenizers")var (	// Model ids for testing.hfModelIDs = []string{		"google/gemma-2-2b-it",		"sentence-transformers/all-MiniLM-L6-v2",		"protectai/deberta-v3-base-zeroshot-v1-onnx",		"KnightsAnalytics/distilbert-base-uncased-finetuned-sst-2-english",		"KnightsAnalytics/distilbert-NER",		"SamLowe/roberta-base-go_emotions-onnx",
	}	hfAuthToken = os.Getenv("HF_TOKEN")  // Create your HuggingFace authentication token in huggingface.co, to allow download of models.)

List files for each model

for _, modelID := range hfModelIDs {	fmt.Printf("n%s:n", modelID)	repo := hub.New(modelID).WithAuth(hfAuthToken)	for fileName, err := range repo.IterFileNames() {		if err != nil { panic(err) }		fmt.Printf("t%sn", fileName)
	}
}

List tokenizer classes for each model

for _, modelID := range hfModelIDs {	fmt.Printf("n%s:n", modelID)	repo := hub.New(modelID).WithAuth(hfAuthToken)	config, err := tokenizers.GetConfig(repo)	if err != nil { panic(err) }	fmt.Printf("ttokenizer_class=%sn", config.TokenizerClass)
}

Tokenize for google/gemma-2-2b-it

• The output "Downloaded" message happens only the tokenizer file is not yet cached, so only the first time:

repo := hub.New("google/gemma-2-2b-it").WithAuth(hfAuthToken)tokenizer, err := tokenizers.New(repo)if err != nil { panic(err) }sentence := "The book is on the table."tokens := tokenizer.Encode(sentence)fmt.Printf("Sentence:t%sn", sentence)fmt.Printf("Tokens:  t%vn", tokens)

Downloaded 1/1 files, 4.2 MB downloaded         
Sentence:	The book is on the table.
Tokens:  	[651 2870 603 611 573 3037 235265]

Download and execute ONNX model for sentence-transformers/all-MiniLM-L6-v2

Only the first 3 lines are actually demoing go-huggingface. The remainder lines uses github.com/gomlx/onnx-gomlxto parse and convert the ONNX model to GoMLX, and thengithub.com/gomlx/gomlx to execute the converted model for a couple of sentences.

// Get ONNX model.repo := hub.New("sentence-transformers/all-MiniLM-L6-v2").WithAuth(hfAuthToken)onnxFilePath, err := repo.DownloadFile("onnx/model.onnx")if err != nil { panic(err) }onnxModel, err := onnx.ReadFile(onnxFilePath)if err != nil { panic(err) }// Convert ONNX variables to GoMLX context (which stores variables):ctx := context.New()err = onnxModel.VariablesToContext(ctx)if err != nil { panic(err) }// Test input.sentences := []string{	"This is an example sentence",	"Each sentence is converted"}inputIDs := [][]int64{
	{101, 2023, 2003, 2019, 2742, 6251,  102},
	{ 101, 2169, 6251, 2003, 4991,  102,    0}}tokenTypeIDs := [][]int64{
	{0, 0, 0, 0, 0, 0, 0},
	{0, 0, 0, 0, 0, 0, 0}}attentionMask := [][]int64{
	{1, 1, 1, 1, 1, 1, 1},
	{1, 1, 1, 1, 1, 1, 0}}// Execute GoMLX graph with model.embeddings := context.ExecOnce(	backends.New(), ctx,	func (ctx *context.Context, inputs []*graph.Node) *graph.Node {		modelOutputs := onnxModel.CallGraph(ctx, inputs[0].Graph(), map[string]*graph.Node{			"input_ids": inputs[0],			"attention_mask": inputs[1],			"token_type_ids": inputs[2]})		return modelOutputs[0]
	}, 
	inputIDs, attentionMask, tokenTypeIDs)fmt.Printf("Sentences: t%qn", sentences)fmt.Printf("Embeddings:t%sn", embeddings)

Sentences: 	["This is an example sentence" "Each sentence is converted"]
Embeddings:	[2][7][384]float32{
 {{0.0366, -0.0162, 0.1682, ..., 0.0554, -0.1644, -0.2967},
  {0.7239, 0.6399, 0.1888, ..., 0.5946, 0.6206, 0.4897},
  {0.0064, 0.0203, 0.0448, ..., 0.3464, 1.3170, -0.1670},
  ...,
  {0.1479, -0.0643, 0.1457, ..., 0.8837, -0.3316, 0.2975},
  {0.5212, 0.6563, 0.5607, ..., -0.0399, 0.0412, -1.4036},
  {1.0824, 0.7140, 0.3986, ..., -0.2301, 0.3243, -1.0313}},
 {{0.2802, 0.1165, -0.0418, ..., 0.2711, -0.1685, -0.2961},
  {0.8729, 0.4545, -0.1091, ..., 0.1365, 0.4580, -0.2042},
  {0.4752, 0.5731, 0.6304, ..., 0.6526, 0.5612, -1.3268},
  ...,
  {0.6113, 0.7920, -0.4685, ..., 0.0854, 1.0592, -0.2983},
  {0.4115, 1.0946, 0.2385, ..., 0.8984, 0.3684, -0.7333},
  {0.1374, 0.5555, 0.2678, ..., 0.5426, 0.4665, -0.5284}}}