New Plugin - Hugging Face

[Ben8t]

Feature description

Currently the http plugins makes possible to call Hugging Face models hosted behind Inference API.

Here is an example:

  - id: llm_categorization
    type: io.kestra.plugin.core.http.Request
    uri: https://api-inference.huggingface.co/models/facebook/bart-large-mnli
    method: POST
    contentType: application/json
    headers:
      Authorization: "Bearer {{ secret('HF_API_TOKEN') }}"
    formData:
      inputs: "I want a refund"
      parameters:
        candidate_labels:
          - "legal"
          - "new feature"
          - "refund"

Most of the text models available under these API follow the same type of parameters. Here are the two main ones we can cover at first:

Simple inputs

curl https://api-inference.huggingface.co/models/meta-llama/Llama-3.2-1B \
	-X POST \
	-d '{"inputs": "Can you please let us know more details about your "}' \
	-H 'Content-Type: application/json' \
	-H 'Authorization: Bearer hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'

inputs and parameters:

curl https://api-inference.huggingface.co/models/facebook/bart-large-mnli \
	-X POST \
	-d '{"inputs": "Hi, I recently bought a device from your company but it is not working as advertised and I would like to get reimbursed!", "parameters": {"candidate_labels": ["refund", "legal", "faq"]}}' \
	-H 'Content-Type: application/json' \
	-H 'Authorization: Bearer hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'

Task Design

Task Type: io.kestra.plugin.huggingface.Inference

Properties:

• apiKey - STRING - required
• model - STRING - required - /, it's what follows the main api endpoint https://api-inference.huggingface.co/models ie. meta-llama/Llama-3.2-1B
• inputs - STRING - optional most of the time it's the prompt
• parameters - MAP/ARRAY - optional
• endpoint - STRING - optional, default to https://api-inference.huggingface.co/models

Example

So instead of using http like this:

  - id: llm_categorization
    type: io.kestra.plugin.core.http.Request
    uri: https://api-inference.huggingface.co/models/facebook/bart-large-mnli
    method: POST
    contentType: application/json
    headers:
      Authorization: "Bearer {{ secret('HF_API_TOKEN') }}"
    formData:
      inputs: "I want a refund"
      parameters:
        candidate_labels:
          - "legal"
          - "new feature"
          - "refund"

we simplify with Kestra DSL to:

  - id: llm_categorization
    type: io.kestra.plugin.huggingface.Inference
    model: facebook/bart-large-mnli
    apiKey: XXXX
    inputs: "I want a refund"
    parameters:
      candidate_labels:
         - "legal"
         - "new feature"
         - "refund"

[ksprabhatkumar]

assign this issue to me i would like to work on this issue

[MilosPaunovic]

Do you know how to approach it? Gone through the whole Plugin Developer Guide? If you're not certain about it, may I suggest that you check our curated list of good first issues to see if there is something you find interesting enough and think you can tackle independently. It can be further filtered by desired labels to narrow down the results (suggesting to filter by either area/backend or area/frontend ones, based on your preferences).