knowledgator/gliclass-modern-base-v2.0

⭐ GLiClass: Generalist and Lightweight Model for Sequence Classification

This is an efficient zero-shot classifier inspired by GLiNER 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.

This version of the model uses a layer-wise selection of features that enables a better understanding of different levels of language. The backbone model is ModernBERT-base, which effectively processes long sequences.

The model was fine-tuned using a new RL-based approach to classification, with F1 and recall rewards.

How to use:

First of all, you need to install GLiClass library:

pip install gliclass
pip install -U transformers>=4.48.0

Than you need to initialize a model and a pipeline:

from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer

model = GLiClassModel.from_pretrained("knowledgator/gliclass-modern-base-v2.0")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-modern-base-v2.0", 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"])

If you want to use it for NLI type of tasks, we recommend representing your premise as a text and hypothesis as a label, you can put several hypotheses, but the model works best with a single input hypothesis.

# Initialize model and multi-label pipeline
text = "The cat slept on the windowsill all afternoon"
labels = ["The cat was awake and playing outside."]
results = pipeline(text, labels, threshold=0.0)[0]
print(results)

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.

Model	IMDB	AG_NEWS	Emotions
gliclass-modern-large-v2.0-init (399 M)	0.9137	0.7357	0.4140
gliclass-modern-base-v2.0-init (151 M)	0.8264	0.6637	0.2985
gliclass-modern-large-v2.0 (399 M)	0.9448	0.736	0.4970
gliclass-modern-base-v2.0 (151 M)	0.9188	0.7089	0.4250
gliclass-large-v1.0 (438 M)	0.9404	0.7516	0.4874
gliclass-base-v1.0 (186 M)	0.8650	0.6837	0.4749
gliclass-small-v1.0 (144 M)	0.8650	0.6805	0.4664
Bart-large-mnli (407 M)	0.89	0.6887	0.3765
Deberta-base-v3 (184 M)	0.85	0.6455	0.5095
Comprehendo (184M)	0.90	0.7982	0.5660
SetFit BAAI/bge-small-en-v1.5 (33.4M)	0.86	0.5636	0.5754

Below you can find a comparison with other GLiClass models:

Dataset	gliclass-modern-base-v2.0	gliclass-modern-large-v2.0	gliclass-modern-base-v2.0-init	gliclass-modern-large-v2.0-init
CR	0.8976	0.9198	0.9041	0.8980
sst2	0.8525	0.9318	0.9011	0.9434
sst5	0.2348	0.2147	0.1972	0.1123
20_news_groups	0.351	0.3755	0.2448	0.2792
spam	0.483	0.6608	0.5074	0.6364
financial_phrasebank	0.3475	0.3157	0.2537	0.2562
imdb	0.9188	0.9448	0.8255	0.9137
ag_news	0.6835	0.7025	0.6050	0.6933
emotion	0.3925	0.4325	0.2474	0.3746
cap_sotu	0.3725	0.4157	0.2929	0.2919
rotten_tomatoes	0.6955	0.7357	0.6630	0.5928
AVERAGE:	0.5563	0.6045	0.5129	0.5447