Hugging Face Inference API

This documentation describes the integration of MindsDB with Hugging Face Inference API. The integration allows for the deployment of Hugging Face models through Inference API within MindsDB, providing the models with access to data from various data sources.

Prerequisites

Before proceeding, ensure the following prerequisites are met:

1. Install MindsDB locally via Docker or use MindsDB Cloud.
2. To use Hugging Face Inference API within MindsDB, install the required dependencies following this instruction.
3. Obtain the API key for Hugging Face Inference API required to deploy and use Hugging Face models through Inference API within MindsDB. Generate tokens in the Settings -> Access Tokens tab of the Hugging Face account.

Setup

Create an AI engine from the Hugging Face Inference API handler.

CREATE ML_ENGINE huggingface_api_engine
FROM huggingface_api
USING
      huggingface_api_api_key = 'api-key-value';

Create a model using huggingface_api_engine as an engine.

CREATE MODEL huggingface_api_model
PREDICT target_column
USING
      engine = 'huggingface_api_engine',     -- engine name as created via CREATE ML_ENGINE
      task = 'task_name',                -- choose one of 'text-classification', 'text-generation', 'question-answering', 'sentence-similarity', 'zero-shot-classification', 'summarization', 'fill-mask', 'image-classification', 'object-detection', 'automatic-speech-recognition', 'audio-classification'
      input_column = 'column_name',      -- column that stores input/question to the model
      labels = ['label 1', 'label 2'];   -- labels used to classify data (used for classification tasks)

The following parameters are supported in the USING clause of the CREATE MODEL statement:

Parameter	Required	Description
engine	Yes	It is the name of the ML engine created with the CREATE ML_ENGINE statement.
task	Only if model_name is not provided	It describes a task to be performed.
model_name	Only if task is not provided	It specifies a model to be used.
input_column	Yes	It is the name of the column that stores input to the model.
endpoint	No	It defines the endpoint to use for API calls. If not specified, the hosted Inference API from Hugging Face will be used.
options	No	It is a JSON object containing additional options to pass to the API call. More information about the available options for each task can be found here.
parameters	No	It is a JSON object containing additional parameters to pass to the API call. More information about the available parameters for each task can be found here.
context_column	Only if task is question-answering	It is used for the question-answering task to provide context to the question.
input_column2	Only if task is sentence-similarity	It is used for the sentence-similarity task to provide the second input sentence for comparison.
candidate_labels	Only if task is zero-shot-classification	It is used for the zero-shot-classification task to classify input data according to provided labels.

Usage

The following usage examples utilize huggingface_api_engine to create a model with the CREATE MODEL statement.

Create a model to classify input text as spam or ham.

CREATE MODEL spam_classifier
PREDICT is_spam
USING
      engine = 'huggingface_api_engine',
      task = 'text-classification',
      column = 'text';

Query the model to get predictions.

SELECT text, is_spam
FROM spam_classifier
WHERE text = 'Subscribe to this channel asap';

Here is the output:

+--------------------------------+---------+
| text                           | is_spam |
+--------------------------------+---------+
| Subscribe to this channel asap | spam    |
+--------------------------------+---------+

Find more quick examples below:

Text Classification

CREATE MODEL mindsdb.hf_text_classifier
PREDICT sentiment
USING
  task = 'text-classification',
  engine = 'hf_api_engine',
  input_column = 'text';

Fill Mask

CREATE MODEL mindsdb.hf_fill_mask
PREDICT sequence
USING
  task = 'fill-mask',
  engine = 'hf_api_engine',
  input_column = 'text';

Summarization

CREATE MODEL mindsdb.hf_summarizer
PREDICT summary
USING
  task = 'summarization',
  engine = 'hf_api_engine',
  input_column = 'text';

Text Generation

CREATE MODEL mindsdb.hf_text_generator
PREDICT generated_text
USING
  task = 'text-generation',
  engine = 'hf_api_engine',
  input_column = 'text';

Question Answering

CREATE MODEL mindsdb.hf_question_answerer
PREDICT answer
USING
  task = 'question-answering',
  engine = 'hf_api_engine',
  input_column = 'question',
  context_column = 'context';

Sentences Similarity

CREATE MODEL mindsdb.hf_sentence_similarity
PREDICT similarity
USING
  task = 'sentence-similarity',
  engine = 'hf_api_engine',
  input_column = 'sentence1',
  input_column2 = 'sentence2';

Zero Shot Classification

CREATE MODEL mindsdb.hf_zero_shot_classifier
PREDICT label
USING
  task = 'zero-shot-classification',
  engine = 'hf_api_engine',
  input_column = 'text',
  candidate_labels = ['label1', 'label2', 'label3'];

Image Classification

CREATE MODEL mindsdb.hf_image_classifier
PREDICT label
USING
  task = 'image-classification',
  engine = 'hf_api_engine',
  input_column = 'image_url';

Object Detection

CREATE MODEL mindsdb.hf_object_detector
PREDICT objects
USING
  task = 'object-detection',
  engine = 'hf_api_engine',
  input_column = 'image_url';

Automatic Speech Recognition

CREATE MODEL mindsdb.hf_speech_recognizer
PREDICT transcription
USING
  task = 'automatic-speech-recognition',
  engine = 'hf_api_engine',
  input_column = 'audio_url';

Audio Classification

CREATE MODEL mindsdb.hf_audio_classifier
PREDICT label
USING
  task = 'audio-classification',
  engine = 'hf_api_engine',
  input_column = 'audio_url';

Next Steps

Follow this link to see more use case examples.