SetFit with akhooli/sbert_ar_nli_500k_ubc_norm

This is a SetFit model that can be used for Text Classification. This SetFit model uses akhooli/sbert_ar_nli_500k_ubc_norm 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: akhooli/sbert_ar_nli_500k_ubc_norm
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
positive	' سبحان الله الفلسطينيين شعب خاين في كل مكان \nلاحول ولا قوة إلا بالله' 'يا بيك عّم تخبرنا عن شي ما فينا تعملو نحن ماًعندنا نواب ولا وزراء بمثلونا بالدولة الا اذا زهقان وعبالك ليك' 'جوز كذابين منافقين...'
negative	'ربي لا تجعلني أسيء الظن بأحد ولا تجعل في قلبي شيئا على أحد ، اللهم أسألك قلباً نقياً صافيا' 'هشام حداد عامل فيها جون ستيوارت' ' بحياة اختك من وين بتجيبي اخبارك؟؟ من صغري وانا عبالي كون... LINK'

Evaluation

Metrics

Label	Accuracy
all	0.8398

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("akhooli/setfit_ar_ubc_hs")
# Run inference
preds = model("شيوعي 
علماني 
مسيحي
انصار سنه 
صوفي 
يمثلك التجمع 
لا يمثلك التجمع 
اهلا بكم جميعا فنحن نريد بناء وطن ❤")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	18.8448	185

Label	Training Sample Count
negative	5200
positive	4943

Training Hyperparameters

batch_size: (32, 32)
num_epochs: (1, 1)
max_steps: 6000
sampling_strategy: undersampling
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: False
warmup_proportion: 0.1
l2_weight: 0.01
seed: 42
run_name: setfit_hate_52k_ubc_6k
eval_max_steps: -1
load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0003	1	0.297	-
0.0333	100	0.2741	-
0.0667	200	0.2178	-
0.1	300	0.1724	-
0.1333	400	0.1449	-
0.1667	500	0.1137	-
0.2	600	0.0902	-
0.2333	700	0.0708	-
0.2667	800	0.0535	-
0.3	900	0.0483	-
0.3333	1000	0.0386	-
0.3667	1100	0.0319	-
0.4	1200	0.0279	-
0.4333	1300	0.0201	-
0.4667	1400	0.0234	-
0.5	1500	0.0151	-
0.5333	1600	0.0151	-
0.5667	1700	0.0137	-
0.6	1800	0.0117	-
0.6333	1900	0.011	-
0.6667	2000	0.0097	-
0.7	2100	0.0077	-
0.7333	2200	0.0089	-
0.7667	2300	0.0069	-
0.8	2400	0.0064	-
0.8333	2500	0.0083	-
0.8667	2600	0.0061	-
0.9	2700	0.0063	-
0.9333	2800	0.0051	-
0.9667	2900	0.0047	-
1.0	3000	0.0044	-
1.0333	3100	0.0035	-
1.0667	3200	0.0034	-
1.1	3300	0.0035	-
1.1333	3400	0.0043	-
1.1667	3500	0.0035	-
1.2	3600	0.0024	-
1.2333	3700	0.003	-
1.2667	3800	0.002	-
1.3	3900	0.0029	-
1.3333	4000	0.003	-
1.3667	4100	0.002	-
1.4	4200	0.0022	-
1.4333	4300	0.0027	-
1.4667	4400	0.004	-
1.5	4500	0.001	-
1.5333	4600	0.0027	-
1.5667	4700	0.0027	-
1.6	4800	0.0014	-
1.6333	4900	0.0022	-
1.6667	5000	0.0027	-
1.7	5100	0.0018	-
1.7333	5200	0.0018	-
1.7667	5300	0.0012	-
1.8	5400	0.0014	-
1.8333	5500	0.0015	-
1.8667	5600	0.0009	-
1.9	5700	0.0012	-
1.9333	5800	0.0009	-
1.9667	5900	0.001	-
2.0	6000	0.0007	-

Framework Versions

Python: 3.10.14
SetFit: 1.2.0.dev0
Sentence Transformers: 3.3.0
Transformers: 4.45.1
PyTorch: 2.4.0
Datasets: 3.0.1
Tokenizers: 0.20.0

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}
}

akhooli
/

setfit_ar_ubc_hs