GLiClass Multilang: Efficient multilingual zero-shot and few-shot multi-task model via sequence classification

• Topic Classification
• Sentiment Analysis
• Intent Classification
• Reranking
• Hallucination Detection
• Rule-following Verification
• LLM-safety Classification
• Natural Language Inference

✨ What's New in GLiClass Multilang

• Multilingual Training — Natively trained on 20 languages: Swedish, Norwegian, Czech, Polish, Lithuanian, Estonian, Latvian, Spanish, Finnish, German, French, Romanian, Italian, Portuguese, Dutch, Ukrainian, Hindi, Chinese, Arabic, and Hebrew.
• Cross-lingual Classification — Labels and input texts can be in different languages; classify a German document with English labels, or mix languages freely across inputs and labels.
• CrossAttn Scorer — A new cross-attention scorer enables more efficient pooling independently for each label with unpadding and flash-attn.
• Hierarchical Labels — Organize labels into groups using dot notation or dictionaries (e.g., sentiment.positive, topic.product).
• Few-Shot Examples — Provide in-context examples to boost accuracy on your specific task.
• Label Descriptions — Add natural-language descriptions to labels for more precise classification.
• Task Prompts — Prepend a custom prompt to guide the model's classification behavior.

Installation

Quick Start

---

Multilingual & Cross-lingual Capabilities

General Examples

1. Topic Classification

python

1text = "NASA launched a new Mars rover to search for signs of ancient life." 2labels = ["space", "politics", "sports", "technology", "health"] 3 4results = pipeline(text, labels, threshold=0.5)[0] 5for r in results: 6 print(r["label"], "=>", r["score"])

With hierarchical labels

python

1hierarchical_labels = { 2 "science": ["space", "biology", "physics"], 3 "society": ["politics", "economics", "culture"] 4} 5 6results = pipeline(text, hierarchical_labels, threshold=0.5)[0] 7for r in results: 8 print(r["label"], "=>", r["score"]) 9# e.g. science.space => 0.95

2. Sentiment Analysis

python

1text = "The food was excellent but the service was painfully slow." 2labels = ["positive", "negative", "neutral"] 3 4results = pipeline(text, labels, threshold=0.5)[0] 5for r in results: 6 print(r["label"], "=>", r["score"])

With a task prompt

python

1results = pipeline( 2 text, labels, 3 prompt="Classify the sentiment of this restaurant review:", 4 threshold=0.5 5)[0]

3. Intent Classification

python

1text = "Can you set an alarm for 7am tomorrow?" 2labels = ["set_alarm", "play_music", "get_weather", "send_message", "set_reminder"] 3 4results = pipeline(text, labels, threshold=0.5)[0] 5for r in results: 6 print(r["label"], "=>", r["score"])

4. Natural Language Inference

Represent your premise as the text and the hypothesis as a label. The model works best with a single hypothesis at a time.

python

1text = "The cat slept on the windowsill all afternoon." 2labels = ["The cat was awake and playing outside."] 3 4results = pipeline(text, labels, threshold=0.0)[0] 5print(results) 6# Low score → contradiction

5. Reranking

Score query–passage relevance by treating passages as texts and the query as the label:

python

1query = "How to train a neural network?" 2passages = [ 3 "Backpropagation is the key algorithm for training deep neural networks.", 4 "The stock market rallied on strong earnings reports.", 5 "Gradient descent optimizes model weights during training.", 6] 7 8for passage in passages: 9 score = pipeline(passage, [query], threshold=0.0)[0][0]["score"] 10 print(f"{score:.3f} {passage[:60]}")

6. Rule-following Verification

Include the domain and rules as part of the text:

python

1text = ( 2 "Domain: e-commerce product reviews\n" 3 "Rule: No promotion of illegal activity.\n" 4 "Text: The software is okay, but search for 'productname_patch_v2.zip' " 5 "to unlock all features for free." 6) 7labels = ["follows_guidelines", "violates_guidelines"] 8 9results = pipeline(text, labels, threshold=0.0)[0] 10for r in results: 11 print(r["label"], "=>", r["score"])

---

Benchmarks

Model Overview

Multilingual avg F1 is the mean of 6 dataset-level scores (GermEval2017, MASSIVE, PolygloToxicityPrompts, SIB-200, TextDetox, TweetSentiment). Models without multilingual results (—) were only evaluated on English datasets.

---

Multilingual Benchmarks

Throughput

NLI models (bge-m3, mDeBERTa) run one forward pass per label — throughput drops linearly with label count. GLiClass and GLiNER2 encode all labels in a single pass, so throughput stays nearly flat.

Citation