SetFit with deutsche-telekom/gbert-large-paraphrase-cosine

This is a SetFit model that can be used for Text Classification. This SetFit model uses deutsche-telekom/gbert-large-paraphrase-cosine as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification.

The model has been trained using an efficient few-shot learning technique that involves:

Fine-tuning a Sentence Transformer with contrastive learning.
Training a classification head with features from the fine-tuned Sentence Transformer.

Model Details

Model Description

Model Type: SetFit
Sentence Transformer body: deutsche-telekom/gbert-large-paraphrase-cosine
Classification head: a LogisticRegression instance
Maximum Sequence Length: 512 tokens
Number of Classes: 2 classes

Model Sources

Repository: SetFit on GitHub
Paper: Efficient Few-Shot Learning Without Prompts
Blogpost: SetFit: Efficient Few-Shot Learning Without Prompts

Model Labels

Label	Examples
1	'Deutschland ist ein gastfreundliches und weltoffenes Land.' 'Aber auch in der Polizei und Justiz muss sich einiges ändern.' 'Die FDP sucht das Gespräch mit der evangelischen und katholischen Kirche ebenso wie mit dem Judentum, dem Islam und allen anderen Religionsgemeinschaften.'
0	'Wir wollen eine Integrationsoffensive.' 'Kenntnisse der deutschen Sprache sind eine unverzichtbare Voraussetzung zur Beseitigung sozialer Benachteiligungen und zum Erreichen schulischer, beruflicher und gesellschaftlicher Erfolge.' 'Wir erwarten von Zuwandernden, dass sie die deutsche Sprache erlernen.'

Evaluation

Metrics

Label	F1	Precision	Recall
all	0.8564	0.8585	0.8548

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("gehaustein/gbert-large-stance-multiculturalism")
# Run inference
preds = model("für Integration")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	14.6336	42

Label	Training Sample Count
0	128
1	366

Training Hyperparameters

batch_size: (128, 128)
num_epochs: (1, 1)
max_steps: -1
sampling_strategy: oversampling
body_learning_rate: (1e-05, 1e-05)
head_learning_rate: 0.01
loss: CosineSimilarityLoss
distance_metric: cosine_distance
margin: 0.25
end_to_end: False
use_amp: False
warmup_proportion: 0.1
seed: 42
eval_max_steps: -1
load_best_model_at_end: True

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0008	1	0.3283	-
0.0424	50	0.2401	0.234
0.0848	100	0.0852	0.202
0.1272	150	0.0054	0.2493
0.1696	200	0.001	0.2502
0.2120	250	0.0002	0.2513
0.2545	300	0.0012	0.2496
0.2969	350	0.0046	0.2485
0.3393	400	0.0056	0.2538
0.3817	450	0.0001	0.2543
0.4241	500	0.0001	0.2443
0.4665	550	0.0001	0.2472
0.5089	600	0.0051	0.2655
0.5513	650	0.0002	0.2646

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.11
SetFit: 1.1.0.dev0
Sentence Transformers: 3.3.1
Transformers: 4.48.1
PyTorch: 2.5.1+cu121
Datasets: 2.14.4
Tokenizers: 0.21.0

Citation

BibTeX

@article{https://doi.org/10.48550/arxiv.2209.11055,
    doi = {10.48550/ARXIV.2209.11055},
    url = {https://arxiv.org/abs/2209.11055},
    author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren},
    keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
    title = {Efficient Few-Shot Learning Without Prompts},
    publisher = {arXiv},
    year = {2022},
    copyright = {Creative Commons Attribution 4.0 International}
}

gehaustein
/

gbert-large-stance-multiculturalism