step

huggingface_inference

Run inference on Hugging Face models using the official huggingface_hub library.

Overview

This step provides access to thousands of pre-trained models hosted on Hugging Face for tasks
like text classification, named entity recognition, summarization, translation, question answering,
text generation, and more. It uses the official huggingface_hub library which automatically handles
API endpoint routing and authentication.

Setup:
1. Create a Hugging Face account at https://huggingface.co/
2. Generate an access token at https://huggingface.co/settings/tokens
3. Choose a model from https://huggingface.co/models (filter by task)
4. Store your token securely as an environment variable

API Token: Required. Get from https://huggingface.co/settings/tokens

Examples

Sentiment analysis (auto-detected task)

Classify text sentiment - task is automatically detected from model metadata

type: huggingface_inference
model: distilbert/distilbert-base-uncased-finetuned-sst-2-english
api_token: ${env:HUGGINGFACE_TOKEN}
input_from: review.text
output_to: review.sentiment

Sentiment analysis (explicit task)

Classify text sentiment using DistilBERT with explicit task specification

type: huggingface_inference
model: distilbert/distilbert-base-uncased-finetuned-sst-2-english
api_token: ${env:HUGGINGFACE_TOKEN}
task: text-classification
input_from: review.text
output_to: review.sentiment

Text summarization

Generate concise summaries of long documents

type: huggingface_inference
model: facebook/bart-large-cnn
api_token: ${env:HUGGINGFACE_TOKEN}
task: summarization
input_from: article.full_text
output_to: article.summary
parameters:
  max_length: 150
  min_length: 50
timeout: 60

Named entity recognition

Extract people, organizations, and locations from text

type: huggingface_inference
model: dslim/bert-base-NER
api_token: ${env:HUGGINGFACE_TOKEN}
task: token-classification
input_from: document.content
output_to: document.entities

Question answering

Answer questions based on provided context

type: huggingface_inference
model: deepset/roberta-base-squad2
api_token: ${env:HUGGINGFACE_TOKEN}
task: question-answering
input_from: qa_input
output_to: answer

Text generation

Generate text continuations using GPT-2

type: huggingface_inference
model: openai-community/gpt2
api_token: ${env:HUGGINGFACE_TOKEN}
task: text-generation
input_from: prompt
output_to: generated
parameters:
  max_new_tokens: 100
  temperature: 0.7

Zero-shot classification

Classify text into custom categories without training

type: huggingface_inference
model: facebook/bart-large-mnli
api_token: ${env:HUGGINGFACE_TOKEN}
task: zero-shot-classification
input_from: text
output_to: classification
parameters:
  candidate_labels:
    - urgent
    - not urgent
    - spam

Configuration

Parameter	Type	Required	Description
`model`	`string`	Yes	Model id from the Hugging Face Hub with namespace (e.g., 'distilbert/distilbert-base-uncased-finetuned-sst-2-english').
`api_token`	`string`	Yes	Hugging Face API token. Get one at https://huggingface.co/settings/tokens
`task`	`string`	No	The inference task to run. If omitted, automatically detected from the model metadata. Common tasks: text-classification, summarization, text-generation, question-answering, token-classification.
`input_from`	`string`	No	Dot path to extract the input text from the event. When omitted, uses the entire event.
`input_key`	`string`	No	DEPRECATED: Use 'input_from' instead.
`output_to`	`string`	No	Event key where the inference result is stored. Default: `"huggingface"`
`output_key`	`string`	No	DEPRECATED: Use 'output_to' instead.
`parameters`	`string`	No	Model-specific parameters (e.g., max_length, temperature, top_p). Passed directly to the model.
`task_params`	`string`	No	DEPRECATED: Use 'parameters' instead.
`swallow_on_error`	`boolean`	No	If True, return original event unchanged on errors instead of injecting error details. Default: `false`
`timeout`	`integer`	No	Request timeout in seconds (default 30). Default: `30`
`provider`	`string`	No	The inference provider to use. Default is 'hf-inference' (HuggingFace's native inference). Default: `"hf-inference"`

Base Configuration

These configuration options are available on all steps:

Parameter	Type	Default	Description
`name`		`null`	Optional name for this step (for documentation and debugging)
`description`		`null`	Optional description of what this step does
`retries`	`integer`	`0`	Number of retry attempts (0-10)
`backoff_seconds`	`number`	`0`	Backoff (seconds) applied between retry attempts
`retry_propagate`	`boolean`	`false`	If True, raise last exception after exhausting retries; otherwise swallow.

← Back to All Steps