inference-providers/pr_2713/en/tasks/token-classification.md

Token Classification

Token classification is a task in which a label is assigned to some tokens in a text. Some popular token classification subtasks are Named Entity Recognition (NER) and Part-of-Speech (PoS) tagging.

For more details about the token-classification task, check out its dedicated page! You will find examples and related materials.

Recommended models

• dslim/bert-base-NER: A robust performance model to identify people, locations, organizations and names of miscellaneous entities.
• FacebookAI/xlm-roberta-large-finetuned-conll03-english: A strong model to identify people, locations, organizations and names in multiple languages.
• blaze999/Medical-NER: A token classification model specialized on medical entity recognition.

Explore all available models and find the one that suits you best here, or from the terminal with the hf CLI:

hf models ls --warm --pipeline-tag token-classification --sort trending_score

Using the API

<InferenceSnippet pipeline=token-classification providersMapping={ {"hf-inference":{"modelId":"rizzoaiacademy/rizzo-pii-0.3B","providerModelId":"rizzoaiacademy/rizzo-pii-0.3B"}} } />

API specification

Request

Headers
authorization	string	Authentication header in the form 'Bearer: hf_****' when hf_**** is a personal user access token with "Inference Providers" permission. You can generate one from your settings page.

Payload
inputs*	string	The input text data
parameters	object
ignore_labels	string[]	A list of labels to ignore
stride	integer	The number of overlapping tokens between chunks when splitting the input text.
aggregation_strategy	string	One of the following:
(#1)	'none'	Do not aggregate tokens
(#2)	'simple'	Group consecutive tokens with the same label in a single entity.
(#3)	'first'	Similar to "simple", also preserves word integrity (use the label predicted for the first token in a word).
(#4)	'average'	Similar to "simple", also preserves word integrity (uses the label with the highest score, averaged across the word's tokens).
(#5)	'max'	Similar to "simple", also preserves word integrity (uses the label with the highest score across the word's tokens).

Response

| Body | | | :--- | :--- | :--- | | (array) | object[] | Output is an array of objects. | |         entity_group | string | The predicted label for a group of one or more tokens | |         entity | string | The predicted label for a single token | |         score | number | The associated score / probability | |         word | string | The corresponding text | |         start | integer | The character position in the input where this group begins. | |         end | integer | The character position in the input where this group ends. |