knowledgator/gliclass-large-v1.0-lw

⭐ 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 data 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.

How to use:

First of all, you need to install GLiClass library:

pip install gliclass

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-large-v1.0-lw")
tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-large-v1.0-lw")

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"])

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-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-small-v1.0-lw	gliclass-base-v1.0-lw	gliclass-large-v1.0-lw	gliclass-small-v1.0	gliclass-base-v1.0	gliclass-large-v1.0
CR	0.8886	0.9097	0.9226	0.8824	0.8942	0.9219
sst2	0.8392	0.8987	0.9247	0.8518	0.8979	0.9269
sst5	0.2865	0.3779	0.2891	0.2424	0.2789	0.3900
20_news_groups	0.4572	0.3953	0.4083	0.3366	0.3576	0.3863
spam	0.5118	0.5126	0.3642	0.4089	0.4938	0.3661
rotten_tomatoes	0.8015	0.8429	0.8807	0.7987	0.8508	0.8808
massive	0.3180	0.4635	0.5606	0.2546	0.1893	0.4376
banking	0.1768	0.4396	0.3317	0.1374	0.2077	0.2847
yahoo_topics	0.4686	0.4784	0.4760	0.4477	0.4516	0.4921
financial_phrasebank	0.8665	0.8880	0.9044	0.8901	0.8955	0.8735
imdb	0.9048	0.9351	0.9429	0.8982	0.9238	0.9333
ag_news	0.7252	0.6985	0.7559	0.7242	0.6848	0.7503
dair_emotion	0.4012	0.3516	0.3951	0.3450	0.2357	0.4013
capsotu	0.3794	0.4643	0.4749	0.3432	0.4375	0.4644
Average:	0.5732	0.6183	0.6165	0.5401	0.5571	0.6078

Here you can see how the performance of the model grows providing more examples:

Model	Num Examples	sst5	spam	massive	banking	ag news	dair emotion	capsotu	Average
gliclass-small-v1.0-lw	0	0.2865	0.5118	0.318	0.1768	0.7252	0.4012	0.3794	0.3998428571
gliclass-base-v1.0-lw	0	0.3779	0.5126	0.4635	0.4396	0.6985	0.3516	0.4643	0.4725714286
gliclass-large-v1.0-lw	0	0.2891	0.3642	0.5606	0.3317	0.7559	0.3951	0.4749	0.4530714286
gliclass-small-v1.0	0	0.2424	0.4089	0.2546	0.1374	0.7242	0.345	0.3432	0.3508142857
gliclass-base-v1.0	0	0.2789	0.4938	0.1893	0.2077	0.6848	0.2357	0.4375	0.3611
gliclass-large-v1.0	0	0.39	0.3661	0.4376	0.2847	0.7503	0.4013	0.4644	0.4420571429
gliclass-small-v1.0-lw	8	0.2709	0.84026	0.62	0.6883	0.7786	0.449	0.4918	0.5912657143
gliclass-base-v1.0-lw	8	0.4275	0.8836	0.729	0.7667	0.7968	0.3866	0.4858	0.6394285714
gliclass-large-v1.0-lw	8	0.3345	0.8997	0.7658	0.848	0.84843	0.5219	0.508	0.67519
gliclass-small-v1.0	8	0.3042	0.5683	0.6332	0.7072	0.759	0.4509	0.4434	0.5523142857
gliclass-base-v1.0	8	0.3387	0.7361	0.7059	0.7456	0.7896	0.4323	0.4802	0.6040571429
gliclass-large-v1.0	8	0.4365	0.9018	0.77	0.8533	0.8509	0.5061	0.4935	0.6874428571