Implementation:Neuml Txtai Text Classification

Knowledge Sources	Neuml_Txtai
Domains	Machine Learning, NLP, Text Classification, Transformers
Last Updated	2026-02-10 01:00 GMT

Overview

Concrete tool for applying text classification models (both zero-shot and standard) to text provided by txtai.

Description

Labels extends HFPipeline and supports two modes of text classification: zero-shot classification (dynamic labels at inference time) and standard text classification (fixed labels from model training). In dynamic mode, it uses the Hugging Face zero-shot-classification pipeline with user-supplied candidate labels. In standard mode, it uses the text-classification pipeline with model-trained labels. The class supports multilabel classification, score normalization (sigmoid, softmax, or raw), label filtering, score thresholding via the flatten parameter, and batch processing.

Usage

Use Labels when you need to classify text against a set of labels. Choose dynamic mode (default) for zero-shot classification where labels can change at inference time, or standard mode for traditional text classification with fixed model-trained labels.

Code Reference

Source Location

• Repository: Neuml_Txtai
• File: src/python/txtai/pipeline/text/labels.py

Signature

class Labels(HFPipeline):
    def __init__(self, path=None, quantize=False, gpu=True, model=None, dynamic=True, **kwargs)
    def __call__(self, text, labels=None, multilabel=False, flatten=None, workers=0, **kwargs)
    def labels(self)
    def outputs(self, results, labels, flatten)
    def limit(self, result, labels)

Import

from txtai.pipeline.text.labels import Labels

I/O Contract

Inputs

Name	Type	Required	Description
text	str or list	Yes	Input text or list of texts to classify.
labels	list	Conditional	List of candidate labels. Required for zero-shot (dynamic) mode; optional for standard mode to filter results.
multilabel	bool or None	No	If True, labels are independent (sigmoid). If False, normalized to sum to 1 (softmax). If None, raw scores. Defaults to False.
flatten	bool or float	No	If set, returns label names instead of (id, score) tuples. If a float, only labels with scores >= that value are returned. If True, returns top-1 label.
workers	int	No	Number of concurrent workers for data processing. Defaults to 0.
dynamic	bool	No (init)	If True (default), uses zero-shot classification. If False, uses standard text classification.
kwargs	dict	No	Additional keyword arguments passed to the pipeline.

Outputs

Name	Type	Description
result	list	When flatten is not set: list of (label_id, score) tuples sorted by descending score. When flatten is set: list of label name strings. For string input, returns a 1D list; for list input, returns a 2D list.

Usage Examples

from txtai.pipeline.text.labels import Labels

# Zero-shot classification (dynamic labels)
labels = Labels()
result = labels("This is a great movie", labels=["positive", "negative"])
# Returns: [(0, 0.95), (1, 0.05)]

# Flatten to get just the top label
result = labels("This is a great movie", labels=["positive", "negative"], flatten=True)
# Returns: ["positive"]

# Standard text classification (fixed labels)
labels = Labels("distilbert-base-uncased-finetuned-sst-2-english", dynamic=False)
result = labels("This is a great movie")
# Returns: [(1, 0.99), (0, 0.01)]  where 1=POSITIVE, 0=NEGATIVE

# Batch classification
results = labels(["Great movie!", "Terrible film."])
# Returns: [[(1, 0.99), (0, 0.01)], [(0, 0.98), (1, 0.02)]]

# Multilabel classification
result = labels("This is a fun action movie", labels=["action", "comedy", "drama"], multilabel=True)

Related Pages

• Environment:Neuml_Txtai_Python_Core_Dependencies