This is a SetFit model that can be used for Text Classification. This SetFit model uses intfloat/multilingual-e5-base 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: intfloat/multilingual-e5-base
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
toxic	'slanteye' 'lets meet in real life cutie ;)' 'youre dead to me and soon irl'
not toxic	'wanna make a snowman?' 'word' 'boost'

Evaluation

Metrics

Label	Accuracy
all	0.9985

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

python

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("johnpaulbin/toxicity-setfit-7-norm")
# Run inference
preds = model("cooked him")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	4.6068	81

Label	Training Sample Count
not toxic	8689
toxic	4512

Training Hyperparameters

batch_size: (128, 128)
num_epochs: (1, 1)
max_steps: -1
sampling_strategy: num_iterations
num_iterations: 12
body_learning_rate: (2e-05, 1e-05)
head_learning_rate: 0.01
loss: CosineSimilarityLoss
distance_metric: cosine_distance
margin: 0.25
end_to_end: False
use_amp: True
warmup_proportion: 0.1
l2_weight: 0.01
seed: 42
eval_max_steps: -1
load_best_model_at_end: True

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0004	1	0.3528	-
0.0202	50	0.3183	-
0.0404	100	0.2297	-
0.0606	150	0.1218	-
0.0808	200	0.0752	-
0.1010	250	0.0624	-
0.1212	300	0.049	-
0.1414	350	0.0389	-
0.1616	400	0.0334	-
0.1817	450	0.0244	-
0.2019	500	0.0248	-
0.2221	550	0.0212	-
0.2423	600	0.0196	-
0.2625	650	0.0177	-
0.2827	700	0.018	-
0.3029	750	0.0137	-
0.3231	800	0.0154	-
0.3433	850	0.014	-
0.3635	900	0.0134	-
0.3837	950	0.0103	-
0.4039	1000	0.0128	-
0.4241	1050	0.0124	-
0.4443	1100	0.0115	-
0.4645	1150	0.0111	-
0.4847	1200	0.0124	-
0.5048	1250	0.0118	-
0.5250	1300	0.0109	-
0.5452	1350	0.0104	-
0.5654	1400	0.0105	-
0.5856	1450	0.0103	-
0.6058	1500	0.0097	-
0.6260	1550	0.0096	-
0.6462	1600	0.0098	-
0.6664	1650	0.008	-
0.6866	1700	0.0082	-
0.7068	1750	0.0084	-
0.7270	1800	0.0065	-
0.7472	1850	0.0069	-
0.7674	1900	0.0088	-
0.7876	1950	0.0075	-
0.8078	2000	0.0058	-
0.8279	2050	0.0077	-
0.8481	2100	0.0055	-
0.8683	2150	0.0053	-
0.8885	2200	0.0072	-
0.9087	2250	0.0079	-
0.9289	2300	0.0074	-
0.9491	2350	0.0065	-
0.9693	2400	0.0072	-
0.9895	2450	0.0061	-
1.0	2476	-	0.0028

Framework Versions

Python: 3.12.12
SetFit: 1.2.0.dev0
Sentence Transformers: 5.2.0
Transformers: 4.57.3
PyTorch: 2.9.0+cu126
Datasets: 4.0.0
Tokenizers: 0.22.2

Citation

BibTeX

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}
}