metadata

library_name: setfit
tags:
  - setfit
  - sentence-transformers
  - text-classification
  - generated_from_setfit_trainer
metrics:
  - accuracy
widget:
  - text: >-
      is completely right on this. carnildo’s comment is just a waste of space.
      176.12.107.140
  - text: >-
      "   please do not vandalize pages, as you did with this edit to bella
      swan. if you continue to do so, you will be blocked from editing.   (talk)
      "
  - text: >-
      ipv6   mirc doesn't natively supports ipv6 protocols. it could be enabled
      by adding a external dll plugin who will enable a special protocol for dns
      and connecting to ipv6 servers.
  - text: >-
      "   link   thanks for fixing that disambiguation link on usher's album )
      flash; "
  - text: >-
      |b-class-1= yes  |b-class-2= yes  |b-class-3= yes  |b-class-4= yes 
      |b-class-5= yes
pipeline_tag: text-classification
inference: true
base_model: sentence-transformers/paraphrase-mpnet-base-v2

SetFit with sentence-transformers/paraphrase-mpnet-base-v2

This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/paraphrase-mpnet-base-v2 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: sentence-transformers/paraphrase-mpnet-base-v2
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
0	', metal unless it was 70s oldskool' "do whichever you think would be best then: if you rename the account, just let me know, and i'll go over there to recreate it; but if you'd prefer to rename the account, recreate it, and send me the password which i can then change, that's fine with me." '" no, it was a far-too-much-of-an-in-joke on the fact that principle→principal is usually one of the first things the fa regulars jump on (along with the dreaded spaced em dash)\xa0–\xa0scent "'
1	'heey fuckface you are a real bitch ass pussies whipped muthafucking faggot shitfaced dick head' 'you cocksucker fuck you ok? i hate you! stop reverting my edits. i will kill you. fuck off and go fuck urself, cockmuncher!' 'fuck you youfuckingidiot'

Label

Examples

', metal unless it was 70s oldskool'
"do whichever you think would be best then: if you rename the account, just let me know, and i'll go over there to recreate it; but if you'd prefer to rename the account, recreate it, and send me the password which i can then change, that's fine with me."
'" no, it was a far-too-much-of-an-in-joke on the fact that principle→principal is usually one of the first things the fa regulars jump on (along with the dreaded spaced em dash)\xa0–\xa0scent "'

'heey fuckface you are a real bitch ass pussies whipped muthafucking faggot shitfaced dick head'
'you cocksucker fuck you ok? i hate you! stop reverting my edits. i will kill you. fuck off and go fuck urself, cockmuncher!'
'fuck you youfuckingidiot'

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("waterabbit114/my-setfit-classifier_severe_toxic")
# Run inference
preds = model("\"   link   thanks for fixing that disambiguation link on usher's album ) flash; \"")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	3	72.05	426

Label	Training Sample Count
0	10
1	10

Training Hyperparameters

batch_size: (1, 1)
num_epochs: (10, 10)
max_steps: -1
sampling_strategy: oversampling
num_iterations: 20
body_learning_rate: (2e-05, 2e-05)
head_learning_rate: 2e-05
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: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0013	1	0.1472	-
0.0625	50	0.0173	-
0.125	100	0.0288	-
0.1875	150	0.0004	-
0.25	200	0.0002	-
0.3125	250	0.0012	-
0.375	300	0.0011	-
0.4375	350	0.0012	-
0.5	400	0.0018	-
0.5625	450	0.0009	-
0.625	500	0.0001	-
0.6875	550	0.0	-
0.75	600	0.0005	-
0.8125	650	0.0	-
0.875	700	0.0	-
0.9375	750	0.0	-
1.0	800	0.0	-
1.0625	850	0.0002	-
1.125	900	0.0	-
1.1875	950	0.0003	-
1.25	1000	0.0003	-
1.3125	1050	0.0	-
1.375	1100	0.0	-
1.4375	1150	0.0001	-
1.5	1200	0.0	-
1.5625	1250	0.0001	-
1.625	1300	0.0001	-
1.6875	1350	0.0	-
1.75	1400	0.0	-
1.8125	1450	0.0	-
1.875	1500	0.0004	-
1.9375	1550	0.0	-
2.0	1600	0.0003	-
2.0625	1650	0.0001	-
2.125	1700	0.0001	-
2.1875	1750	0.0001	-
2.25	1800	0.0	-
2.3125	1850	0.0	-
2.375	1900	0.0001	-
2.4375	1950	0.0001	-
2.5	2000	0.0	-
2.5625	2050	0.0	-
2.625	2100	0.0	-
2.6875	2150	0.0	-
2.75	2200	0.0002	-
2.8125	2250	0.0001	-
2.875	2300	0.0	-
2.9375	2350	0.0001	-
3.0	2400	0.0001	-
3.0625	2450	0.0	-
3.125	2500	0.0	-
3.1875	2550	0.0001	-
3.25	2600	0.0	-
3.3125	2650	0.0	-
3.375	2700	0.0	-
3.4375	2750	0.0	-
3.5	2800	0.0	-
3.5625	2850	0.0	-
3.625	2900	0.0	-
3.6875	2950	0.0001	-
3.75	3000	0.0	-
3.8125	3050	0.0	-
3.875	3100	0.0	-
3.9375	3150	0.0	-
4.0	3200	0.0	-
4.0625	3250	0.0	-
4.125	3300	0.0	-
4.1875	3350	0.0	-
4.25	3400	0.0	-
4.3125	3450	0.0	-
4.375	3500	0.0	-
4.4375	3550	0.0001	-
4.5	3600	0.0	-
4.5625	3650	0.0	-
4.625	3700	0.0	-
4.6875	3750	0.0	-
4.75	3800	0.0	-
4.8125	3850	0.0	-
4.875	3900	0.0	-
4.9375	3950	0.0	-
5.0	4000	0.0	-
5.0625	4050	0.0001	-
5.125	4100	0.0	-
5.1875	4150	0.0	-
5.25	4200	0.0	-
5.3125	4250	0.0	-
5.375	4300	0.0	-
5.4375	4350	0.0	-
5.5	4400	0.0	-
5.5625	4450	0.0	-
5.625	4500	0.0	-
5.6875	4550	0.0	-
5.75	4600	0.0	-
5.8125	4650	0.0	-
5.875	4700	0.0	-
5.9375	4750	0.0	-
6.0	4800	0.0	-
6.0625	4850	0.0	-
6.125	4900	0.0001	-
6.1875	4950	0.0001	-
6.25	5000	0.0	-
6.3125	5050	0.0	-
6.375	5100	0.0	-
6.4375	5150	0.0	-
6.5	5200	0.0	-
6.5625	5250	0.0	-
6.625	5300	0.0	-
6.6875	5350	0.0	-
6.75	5400	0.0	-
6.8125	5450	0.0	-
6.875	5500	0.0	-
6.9375	5550	0.0	-
7.0	5600	0.0	-
7.0625	5650	0.0	-
7.125	5700	0.0	-
7.1875	5750	0.0	-
7.25	5800	0.0	-
7.3125	5850	0.0	-
7.375	5900	0.0	-
7.4375	5950	0.0	-
7.5	6000	0.0	-
7.5625	6050	0.0	-
7.625	6100	0.0	-
7.6875	6150	0.0	-
7.75	6200	0.0	-
7.8125	6250	0.0	-
7.875	6300	0.0	-
7.9375	6350	0.0	-
8.0	6400	0.0	-
8.0625	6450	0.0	-
8.125	6500	0.0	-
8.1875	6550	0.0	-
8.25	6600	0.0	-
8.3125	6650	0.0	-
8.375	6700	0.0	-
8.4375	6750	0.0	-
8.5	6800	0.0	-
8.5625	6850	0.0	-
8.625	6900	0.0	-
8.6875	6950	0.0001	-
8.75	7000	0.0	-
8.8125	7050	0.0	-
8.875	7100	0.0	-
8.9375	7150	0.0	-
9.0	7200	0.0	-
9.0625	7250	0.0	-
9.125	7300	0.0	-
9.1875	7350	0.0	-
9.25	7400	0.0	-
9.3125	7450	0.0	-
9.375	7500	0.0	-
9.4375	7550	0.0	-
9.5	7600	0.0	-
9.5625	7650	0.0	-
9.625	7700	0.0	-
9.6875	7750	0.0	-
9.75	7800	0.0	-
9.8125	7850	0.0	-
9.875	7900	0.0	-
9.9375	7950	0.0	-
10.0	8000	0.0	-

Framework Versions

Python: 3.11.7
SetFit: 1.0.3
Sentence Transformers: 2.2.2
Transformers: 4.35.2
PyTorch: 2.1.1+cu121
Datasets: 2.14.5
Tokenizers: 0.15.1

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}
}