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 Docker Desktop.
  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:

ParameterRequiredDescription
engineYesIt is the name of the ML engine created with the CREATE ML_ENGINE statement.
taskOnly if model_name is not providedIt describes a task to be performed.
model_nameOnly if task is not providedIt specifies a model to be used.
input_columnYesIt is the name of the column that stores input to the model.
endpointNoIt defines the endpoint to use for API calls. If not specified, the hosted Inference API from Hugging Face will be used.
optionsNoIt 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.
parametersNoIt 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_columnOnly if task is question-answeringIt is used for the question-answering task to provide context to the question.
input_column2Only if task is sentence-similarityIt is used for the sentence-similarity task to provide the second input sentence for comparison.
candidate_labelsOnly if task is zero-shot-classificationIt 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:

Next Steps

Follow this link to see more use case examples.

Was this page helpful?

Hugging FaceLangChain