akhooli/setfit_ar_ubc_hs

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:

1. Fine-tuning a Sentence Transformer with contrastive learning.
2. 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
negative	'جبران باسيل يغرّد عن رحيل مي منسى لبنان سيفتقد روحك الجميلة' 'نشرتكم مقابلة لوزير الخارجية اللبناني جبران باسيل مع سي أن أن تثير الجدل على منصات التواصل ما السبب؟' ' حسا السنه حسا الشيعه حسا الحضران والبدوان\nتساوينا بمحبتها وساوتنا أراضيها\n🌴\nعساني يالحسا فدوة فديت الداار والسكان\nمن اول بيت لآخر بيت بقراها وحواريها\n🌴\nمكانك يالحسا عالي وقدرك فوق مهما كان\nومن حاول يقارن بك بيخسر في تواليها\n🌴وسلامتكم🌴'
positive	'جبران باسيل اكثر الساسة نفاقا' 'انا اقترح عليك طبيب بيطري' 'انتي شكلك مو سورية بس عم تزعمي'

Evaluation

Metrics

Label	Accuracy
all	0.8625

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("كول هوا و سد نيعك")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	16.5047	102

Label	Training Sample Count
negative	3709
positive	3800

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_38k_ubc_6k
• eval_max_steps: -1
• load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0003	1	0.3147	-
0.0333	100	0.2724	-
0.0667	200	0.2165	-
0.1	300	0.163	-
0.1333	400	0.1275	-
0.1667	500	0.0939	-
0.2	600	0.0725	-
0.2333	700	0.0535	-
0.2667	800	0.041	-
0.3	900	0.0369	-
0.3333	1000	0.0286	-
0.3667	1100	0.0202	-
0.4	1200	0.0186	-
0.4333	1300	0.0187	-
0.4667	1400	0.0129	-
0.5	1500	0.0124	-
0.5333	1600	0.0112	-
0.5667	1700	0.0094	-
0.6	1800	0.008	-
0.6333	1900	0.0057	-
0.6667	2000	0.0056	-
0.7	2100	0.0044	-
0.7333	2200	0.0026	-
0.7667	2300	0.0042	-
0.8	2400	0.0022	-
0.8333	2500	0.0022	-
0.8667	2600	0.0031	-
0.9	2700	0.0031	-
0.9333	2800	0.0019	-
0.9667	2900	0.0032	-
1.0	3000	0.002	-
1.0333	3100	0.0012	-
1.0667	3200	0.0022	-
1.1	3300	0.0012	-
1.1333	3400	0.0023	-
1.1667	3500	0.0032	-
1.2	3600	0.002	-
1.2333	3700	0.0021	-
1.2667	3800	0.0015	-
1.3	3900	0.0012	-
1.3333	4000	0.0013	-
1.3667	4100	0.0016	-
1.4	4200	0.0006	-
1.4333	4300	0.0013	-
1.4667	4400	0.0017	-
1.5	4500	0.0007	-
1.5333	4600	0.0014	-
1.5667	4700	0.0002	-
1.6	4800	0.0005	-
1.6333	4900	0.0005	-
1.6667	5000	0.0003	-
1.7	5100	0.0006	-
1.7333	5200	0.0003	-
1.7667	5300	0.0005	-
1.8	5400	0.0005	-
1.8333	5500	0.0008	-
1.8667	5600	0.0006	-
1.9	5700	0.0007	-
1.9333	5800	0.0002	-
1.9667	5900	0.0001	-
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}
}