File size: 7,490 Bytes
0651dda 1e07c00 50b4207 0651dda 1e07c00 68ef56f 1e07c00 68ef56f 8f5754c ab5114e 4599a43 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 | ---
license: apache-2.0
datasets:
- knowledgator/gliclass-v2.0
pipeline_tag: text-classification
---
# ⭐ GLiClass: Generalist and Lightweight Model for Sequence Classification
This is an efficient zero-shot classifier inspired by [GLiNER](https://github.com/urchade/GLiNER/tree/main) work. It demonstrates the same performance as a cross-encoder while being more compute-efficient because classification is done at a single forward path.
It can be used for `topic classification`, `sentiment analysis` and as a reranker in `RAG` pipelines.
The model was trained on synthetic and licensed data that allow commercial use and can be used in commercial applications.
The backbone model is [mdeberta-v3-base](huggingface.co/microsoft/mdeberta-v3-base). It supports multilingual understanding, making it well-suited for tasks involving texts in different languages.
### How to use:
First of all, you need to install GLiClass library:
```bash
pip install gliclass
pip install -U transformers>=4.48.0
```
Than you need to initialize a model and a pipeline:
<details>
<summary>English</summary>
```python
from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer
model = GLiClassModel.from_pretrained("knowledgator/gliclass-x-base")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-x-base", add_prefix_space=True)
pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device='cuda:0')
text = "One day I will see the world!"
labels = ["travel", "dreams", "sport", "science", "politics"]
results = pipeline(text, labels, threshold=0.5)[0] #because we have one text
for result in results:
print(result["label"], "=>", result["score"])
```
</details>
<details>
<summary>Spanish</summary>
```python
from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer
model = GLiClassModel.from_pretrained("knowledgator/gliclass-x-base")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-x-base", add_prefix_space=True)
pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device='cuda:0')
text = "¡Un día veré el mundo!"
labels = ["viajes", "sueños", "deportes", "ciencia", "política"]
results = pipeline(text, labels, threshold=0.5)[0]
for result in results:
print(result["label"], "=>", result["score"])
```
</details>
<details>
<summary>Italitan</summary>
```python
from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer
model = GLiClassModel.from_pretrained("knowledgator/gliclass-x-base")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-x-base", add_prefix_space=True)
pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device='cuda:0')
text = "Un giorno vedrò il mondo!"
labels = ["viaggi", "sogni", "sport", "scienza", "politica"]
results = pipeline(text, labels, threshold=0.5)[0]
for result in results:
print(result["label"], "=>", result["score"])
```
</details>
<details>
<summary>French</summary>
```python
from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer
model = GLiClassModel.from_pretrained("knowledgator/gliclass-x-base")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-x-base", add_prefix_space=True)
pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device='cuda:0')
text = "Un jour, je verrai le monde!"
labels = ["voyage", "rêves", "sport", "science", "politique"]
results = pipeline(text, labels, threshold=0.5)[0]
for result in results:
print(result["label"], "=>", result["score"])
```
</details>
<details>
<summary>German</summary>
```python
from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer
model = GLiClassModel.from_pretrained("knowledgator/gliclass-x-base")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-x-base", add_prefix_space=True)
pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device='cuda:0')
text = "Eines Tages werde ich die Welt sehen!"
labels = ["Reisen", "Träume", "Sport", "Wissenschaft", "Politik"]
results = pipeline(text, labels, threshold=0.5)[0]
for result in results:
print(result["label"], "=>", result["score"])
```
</details>
### Benchmarks:
Below, you can see the F1 score on several text classification datasets. All tested models were not fine-tuned on those datasets and were tested in a zero-shot setting.
#### Multilingual benchmarks
| Dataset | gliclass-x-base | gliclass-base-v3.0 | gliclass-large-v3.0 |
| ------------------------ | --------------- | ------------------ | ------------------- |
| FredZhang7/toxi-text-3M | 0.5972 | 0.5072 | 0.6118 |
| SetFit/xglue\_nc | 0.5014 | 0.5348 | 0.5378 |
| Davlan/sib200\_14classes | 0.4663 | 0.2867 | 0.3173 |
| uhhlt/GermEval2017 | 0.3999 | 0.4010 | 0.4299 |
| dolfsai/toxic\_es | 0.1250 | 0.1399 | 0.1412 |
| **Average** | **0.41796** | **0.37392** | **0.4076** |
#### General benchmarks
| Dataset | gliclass-x-base | gliclass-base-v3.0 | gliclass-large-v3.0 |
| ---------------------------- | --------------- | ------------------ | ------------------- |
| SetFit/CR | 0.8630 | 0.9127 | 0.9398 |
| SetFit/sst2 | 0.8554 | 0.8959 | 0.9192 |
| SetFit/sst5 | 0.3287 | 0.3376 | 0.4606 |
| AmazonScience/massive | 0.2611 | 0.5040 | 0.5649 |
| stanfordnlp/imdb | 0.8840 | 0.9251 | 0.9366 |
| SetFit/20\_newsgroups | 0.4116 | 0.4759 | 0.5958 |
| SetFit/enron\_spam | 0.5929 | 0.6760 | 0.7584 |
| PolyAI/banking77 | 0.3098 | 0.4698 | 0.5574 |
| takala/financial\_phrasebank | 0.7851 | 0.8971 | 0.9000 |
| ag\_news | 0.6815 | 0.7279 | 0.7181 |
| dair-ai/emotion | 0.3667 | 0.4447 | 0.4506 |
| MoritzLaurer/cap\_sotu | 0.3935 | 0.4614 | 0.4589 |
| cornell/rotten\_tomatoes | 0.7252 | 0.7943 | 0.8411 |
| snips | 0.6307 | 0.9474 | 0.9692 |
| **Average** | **0.5778** | **0.6764** | **0.7193** |
## Citation
```bibtex
@misc{stepanov2025gliclassgeneralistlightweightmodel,
title={GLiClass: Generalist Lightweight Model for Sequence Classification Tasks},
author={Ihor Stepanov and Mykhailo Shtopko and Dmytro Vodianytskyi and Oleksandr Lukashov and Alexander Yavorskyi and Mykyta Yaroshenko},
year={2025},
eprint={2508.07662},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2508.07662},
}
``` |