SetFit with firqaaa/indo-sentence-bert-base for indonlu/smsa

Author

Kelompok 3 :

Muhammad Guntur Arfianto (20/459272/PA/19933)
Putri Iqlima Miftahuddini (23/531392/NUGM/01467)
Alan Kurniawan (23/531301/NUGM/01382)

This is a SetFit model that can be used for Text Classification. This SetFit model uses firqaaa/indo-sentence-bert-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.

The dataset that was used for fine-tuning this model is indonlu, specifically its subset, SmSa dataset.

Model Details

Model Description

Model Type: SetFit
Sentence Transformer body: firqaaa/indo-sentence-bert-base
Classification head: a LogisticRegression instance
Maximum Sequence Length: 512 tokens
Number of Classes: 3 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	'dirjen per kereta api - an kemenhub zulfikri memastikan tahun 2018 tarif kereta api kelas ekonomi tidak ada kenaikan untuk semua jurusan setelah ada subsidi dari pemerintah untuk pt kan' 'baik terima kasih banyak' 'kaitan kalung cantik bahan perak / silver 925'
0	'jokowi tidak suka sebar isu bohong' 'masih dengan hawa dingin khas lembang , d sdl menawarkan menu ayam sebagai jagoan nya . ayam ngumpet dan sate goreng adalah 2 menu khas restoran ini . selonjoran di gazebo sambil mencari ayam yang memang seolah ngumpet untuk dimakan menjadikan sensasi tersendiri . dari segi rasa , restoran ini termasuk yang rekomendasi .' 'menu utama adalah indomie dengan variasi topping . rasanya , . ya indomie . tidak terlalu istimewa . cocok untuk tempat santai dan nongkrong anak anak muda karena penyedia aneka permainan papan . kopi gayo dan latte nya oke . roti bakar green tea juga oke .'
2	'tetap tidak prabowo walau saya juga tidak suka jokowi' 'kenapa tidak rekomendasi ? 1 . pempek belum matang , tapi sudah disajikan 2 . pesan sorabi , sudah lama pakai bonus lalat 3 . pesan iga bakar coet , di menu dapat bintang 3 , realita nya tidak enak sama sekali 4 . sorabi kinca dingin , yang datang ternyata sorabi pakai sirop kopyor , nama nya kinca bukan nya air gulu merah ya ? secara keseluruhan baik , tidak puas sama pelayanan dan kualitas makanan di .' 'nabi muhammad adalah hewan gila seks .'

Evaluation

Metrics

Label	Accuracy	Precision	Recall	F1
all	0.8182	0.8182	0.8182	0.8182

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("TRUEnder/setfit-indosentencebert-indonlusmsa-32-shot")
# Run inference
preds = model("liverpool sukses di kandang tottenham")

Training Details

Training Set Metrics

Label	Training Sample Count
0	32
1	32
2	32

Training Hyperparameters

batch_size: (16, 2)
num_epochs: (6, 16)
max_steps: -1
sampling_strategy: oversampling
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
seed: 42
eval_max_steps: -1
load_best_model_at_end: True

Training Results (Epoch-to-epoch)

Epoch	Step	Training Loss	Validation Loss
1.0	384	0.0002	0.1683
2.0	768	0.0001	0.1732
3.0	1152	0.0001	0.1739
4.0	1536	0.0	0.174
5.0	1920	0.0001	0.1765
6.0	2304	0.0	0.1767

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.12
SetFit: 1.0.3
Sentence Transformers: 3.0.1
Transformers: 4.41.2
PyTorch: 2.3.0+cu121
Datasets: 2.19.2
Tokenizers: 0.19.1

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}
}

TRUEnder
/

setfit-indosentencebert-indonlusmsa-32-shot