SetFit

This is a SetFit model that can be used for Text Classification. 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
Classification head: a LogisticRegression instance
Maximum Sequence Length: 256 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
negative	'Biasanya cuma janji saja tapi tak pernah buat pun' 'Aku faham kalau diorang buat macam ni sebab kalau aku jadi doktor pun aku penat la kimak. Bodoh punya kerajaan !' 'Jika parlimen tidak dipanggil dalam masa 14 hari maka UMNO akan keluar menjadi pembangkang iaitu lebih kurang 4 Jul https://t.co/AvJSf1F8ux'
positive	'Kek telapuk kuda, kek lembut rasa premium! Tgh kumpoi order ni utk warga Sungai Petani. Tak terlioq ka tengok? Mai https://t.co/wgcEADvxQK' 'Kalau projek ini berjalan, peneroka Felda tak payah masuk kebun lagi, harga sawit baru RM500 1 tan, murah.' 'Justeru, kita berharap lebih banyak pusat rawatan seperti ini diwujudkan untuk membantu.'
neutral	'08.05 WIB #Jalan_Layang_MBZ Cikunir - Tambun - Cikarang - Karawang LANCAR. ; Karawang - Cikarang - Tambun - Cikunir LANCAR.' '5) Menilai Kualiti Kandungan. Selepas dah faham apa yang penulis cuba sampaikan, kita boleh nilai sama ada ia bena https://t.co/LnkjtBc3Nm' '@syafirazbd_ Siapp'

Evaluation

Metrics

Label	F1
all	0.608

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("babysharkdododo/setfit-all-minilm-l6-v2-malay_en_cn_sentiment_analysis")
# Run inference
preds = model("Kebanyakan orang Cina yang kamu temui di sini akan beritahu kamu, kebimbangan mereka tentang isu yang melanda negara.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	16.598	58

Label	Training Sample Count
positive	235
neutral	77
negative	188

Training Hyperparameters

batch_size: (128, 128)
num_epochs: (10, 10)
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	Step	Training Loss	Validation Loss
0.0001	1	0.189	-
0.0052	50	0.3053	-
0.0104	100	0.2779	-
0.0156	150	0.4526	-
0.0208	200	0.3073	-
0.0261	250	0.4156	-
0.0313	300	0.3912	-
0.0365	350	0.2259	-
0.0417	400	0.2445	-
0.0004	1	0.3311	-
0.0208	50	0.3569	-
0.0417	100	0.2946	-
0.0625	150	0.3397	-
0.0008	1	0.3177	-
0.0417	50	0.2764	-
0.0833	100	0.226	-
0.125	150	0.2649	-
0.1667	200	0.282	-
0.2083	250	0.2438	-
0.25	300	0.2425	-
0.2917	350	0.2555	-
0.3333	400	0.2266	-
0.375	450	0.14	-
0.4167	500	0.1446	-
0.4583	550	0.152	-
0.5	600	0.184	-
0.5417	650	0.095	-
0.5833	700	0.1358	-
0.625	750	0.0859	-
0.6667	800	0.0756	-
0.7083	850	0.0622	-
0.75	900	0.0719	-
0.7917	950	0.0681	-
0.8333	1000	0.0684	-
0.875	1050	0.0356	-
0.9167	1100	0.0233	-
0.9583	1150	0.0126	-
1.0	1200	0.0022	0.3748
1.0417	1250	0.0095	-
1.0833	1300	0.0095	-
1.125	1350	0.0376	-
1.1667	1400	0.0075	-
1.2083	1450	0.0075	-
1.25	1500	0.0142	-
1.2917	1550	0.0113	-
1.3333	1600	0.0022	-
1.375	1650	0.0006	-
1.4167	1700	0.0005	-
1.4583	1750	0.0005	-
1.5	1800	0.0003	-
1.5417	1850	0.0021	-
1.5833	1900	0.0004	-
1.625	1950	0.0006	-
1.6667	2000	0.001	-
1.7083	2050	0.0002	-
1.75	2100	0.0002	-
1.7917	2150	0.0002	-
1.8333	2200	0.0002	-
1.875	2250	0.0059	-
1.9167	2300	0.0002	-
1.9583	2350	0.0005	-
2.0	2400	0.0001	0.3806
2.0417	2450	0.0001	-
2.0833	2500	0.0012	-
2.125	2550	0.0001	-
2.1667	2600	0.0002	-
2.2083	2650	0.0002	-
2.25	2700	0.0001	-
2.2917	2750	0.0011	-
2.3333	2800	0.0002	-
2.375	2850	0.0001	-
2.4167	2900	0.0003	-
2.4583	2950	0.0007	-
2.5	3000	0.0001	-
2.5417	3050	0.0001	-
2.5833	3100	0.0001	-
2.625	3150	0.0001	-
2.6667	3200	0.0001	-
2.7083	3250	0.0001	-
2.75	3300	0.0001	-
2.7917	3350	0.0002	-
2.8333	3400	0.0001	-
2.875	3450	0.0001	-
2.9167	3500	0.0001	-
2.9583	3550	0.0001	-
3.0	3600	0.0001	0.4004
3.0417	3650	0.0001	-
3.0833	3700	0.0001	-
3.125	3750	0.0001	-
3.1667	3800	0.0001	-
3.2083	3850	0.0002	-
3.25	3900	0.0001	-
3.2917	3950	0.0001	-
3.3333	4000	0.0005	-
3.375	4050	0.0001	-
3.4167	4100	0.0001	-
3.4583	4150	0.0001	-
3.5	4200	0.0004	-
3.5417	4250	0.0	-
3.5833	4300	0.0001	-
3.625	4350	0.0001	-
3.6667	4400	0.0001	-
3.7083	4450	0.0001	-
3.75	4500	0.0	-
3.7917	4550	0.0	-
3.8333	4600	0.0	-
3.875	4650	0.0001	-
3.9167	4700	0.0001	-
3.9583	4750	0.0001	-
4.0	4800	0.0	0.4004
4.0417	4850	0.0001	-
4.0833	4900	0.0003	-
4.125	4950	0.0	-
4.1667	5000	0.0001	-
4.2083	5050	0.0001	-
4.25	5100	0.0	-
4.2917	5150	0.0003	-
4.3333	5200	0.0001	-
4.375	5250	0.0	-
4.4167	5300	0.0	-
4.4583	5350	0.0002	-
4.5	5400	0.0	-
4.5417	5450	0.0001	-
4.5833	5500	0.0001	-
4.625	5550	0.0	-
4.6667	5600	0.0006	-
4.7083	5650	0.0	-
4.75	5700	0.0	-
4.7917	5750	0.0	-
4.8333	5800	0.0	-
4.875	5850	0.0	-
4.9167	5900	0.0	-
4.9583	5950	0.0	-
5.0	6000	0.0001	0.391
5.0417	6050	0.0	-
5.0833	6100	0.0001	-
5.125	6150	0.0	-
5.1667	6200	0.0	-
5.2083	6250	0.0	-
5.25	6300	0.0	-
5.2917	6350	0.0	-
5.3333	6400	0.0	-
5.375	6450	0.0	-
5.4167	6500	0.0	-
5.4583	6550	0.0	-
5.5	6600	0.0001	-
5.5417	6650	0.0	-
5.5833	6700	0.0	-
5.625	6750	0.0	-
5.6667	6800	0.0	-
5.7083	6850	0.0	-
5.75	6900	0.0001	-
5.7917	6950	0.0	-
5.8333	7000	0.0001	-
5.875	7050	0.0	-
5.9167	7100	0.0	-
5.9583	7150	0.0	-
6.0	7200	0.0001	0.4026
6.0417	7250	0.0	-
6.0833	7300	0.0	-
6.125	7350	0.0	-
6.1667	7400	0.0	-
6.2083	7450	0.0	-
6.25	7500	0.0	-
6.2917	7550	0.0	-
6.3333	7600	0.0	-
6.375	7650	0.0	-
6.4167	7700	0.0	-
6.4583	7750	0.0	-
6.5	7800	0.0	-
6.5417	7850	0.0	-
6.5833	7900	0.0	-
6.625	7950	0.0	-
6.6667	8000	0.0001	-
6.7083	8050	0.0005	-
6.75	8100	0.0063	-
6.7917	8150	0.0	-
6.8333	8200	0.0	-
6.875	8250	0.0	-
6.9167	8300	0.0	-
6.9583	8350	0.0	-
7.0	8400	0.0	0.4018
7.0417	8450	0.0	-
7.0833	8500	0.0	-
7.125	8550	0.0	-
7.1667	8600	0.0	-
7.2083	8650	0.0	-
7.25	8700	0.0	-
7.2917	8750	0.0	-
7.3333	8800	0.0	-
7.375	8850	0.0	-
7.4167	8900	0.0	-
7.4583	8950	0.0	-
7.5	9000	0.0	-
7.5417	9050	0.0	-
7.5833	9100	0.0	-
7.625	9150	0.0	-
7.6667	9200	0.0	-
7.7083	9250	0.0	-
7.75	9300	0.0	-
7.7917	9350	0.0	-
7.8333	9400	0.0	-
7.875	9450	0.0	-
7.9167	9500	0.0	-
7.9583	9550	0.0	-
8.0	9600	0.0	0.4001
8.0417	9650	0.0	-
8.0833	9700	0.0	-
8.125	9750	0.0	-

Framework Versions

Python: 3.10.14
SetFit: 1.0.3
Sentence Transformers: 3.0.1
Transformers: 4.39.0
PyTorch: 2.2.2
Datasets: 2.20.0
Tokenizers: 0.15.2

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}
}

babysharkdododo
/

setfit-all-minilm-l6-v2-malay_en_cn_sentiment_analysis