SetFit with sentence-transformers/all-MiniLM-L6-v2

This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/all-MiniLM-L6-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/all-MiniLM-L6-v2
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
Solution or Action	'The women of the community decided that they would not make the electricity bill and would send their children to school' 'Since the children did not have a father the mother decided to enroll them in school' 'Preeti Kumari age 18 does not want to appear for the matriculation examination due to failure Explain the solution to her parents and try to get her to appear for the examination and take her forward'
Challenge	'Girls have work responsibilities so they dont get a chance to study' 'Due to poverty children are unable to pursue further studies' 'Kavita didis daughter has left her studies after the fifth lesson'
Other	'The community woman shared that one childs birth certificate has not been made which could potentially affect their education' 'No community members should hold meetings with teachers at school' 'Regarding Siddhanta it was said that you must study only then your future will be bright and you will not face any problems in future'

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("setfit_model_id")
# Run inference
preds = model("Education is very important for children in poverty and can also create a future")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	19.8416	402

Label	Training Sample Count
Challenge	1634
Other	440
Solution or Action	1852

Training Hyperparameters

batch_size: (16, 16)
num_epochs: (1, 1)
max_steps: -1
sampling_strategy: oversampling
num_iterations: 20
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
eval_max_steps: -1
load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0002	1	0.2792	-
0.0102	50	0.2964	-
0.0204	100	0.2595	-
0.0306	150	0.23	-
0.0407	200	0.1881	-
0.0509	250	0.1349	-
0.0611	300	0.109	-
0.0713	350	0.1024	-
0.0815	400	0.0855	-
0.0917	450	0.0751	-
0.1019	500	0.0792	-
0.1121	550	0.074	-
0.1222	600	0.0643	-
0.1324	650	0.0668	-
0.1426	700	0.0624	-
0.1528	750	0.0604	-
0.1630	800	0.0513	-
0.1732	850	0.0455	-
0.1834	900	0.038	-
0.1936	950	0.0439	-
0.2037	1000	0.0331	-
0.2139	1050	0.0323	-
0.2241	1100	0.0269	-
0.2343	1150	0.0376	-
0.2445	1200	0.0256	-
0.2547	1250	0.0279	-
0.2649	1300	0.0276	-
0.2751	1350	0.0289	-
0.2852	1400	0.0241	-
0.2954	1450	0.0179	-
0.3056	1500	0.0177	-
0.3158	1550	0.0226	-
0.3260	1600	0.0176	-
0.3362	1650	0.0182	-
0.3464	1700	0.0171	-
0.3566	1750	0.0175	-
0.3667	1800	0.0141	-
0.3769	1850	0.0177	-
0.3871	1900	0.0136	-
0.3973	1950	0.0127	-
0.4075	2000	0.0173	-
0.4177	2050	0.0106	-
0.4279	2100	0.01	-
0.4381	2150	0.0113	-
0.4482	2200	0.0091	-
0.4584	2250	0.0071	-
0.4686	2300	0.01	-
0.4788	2350	0.0112	-
0.4890	2400	0.0104	-
0.4992	2450	0.0089	-
0.5094	2500	0.0096	-
0.5196	2550	0.0098	-
0.5297	2600	0.0113	-
0.5399	2650	0.005	-
0.5501	2700	0.0074	-
0.5603	2750	0.0078	-
0.5705	2800	0.007	-
0.5807	2850	0.0047	-
0.5909	2900	0.0111	-
0.6011	2950	0.0054	-
0.6112	3000	0.0088	-
0.6214	3050	0.0045	-
0.6316	3100	0.0031	-
0.6418	3150	0.0084	-
0.6520	3200	0.005	-
0.6622	3250	0.0058	-
0.6724	3300	0.0048	-
0.6826	3350	0.0057	-
0.6927	3400	0.0048	-
0.7029	3450	0.0059	-
0.7131	3500	0.0078	-
0.7233	3550	0.0057	-
0.7335	3600	0.0056	-
0.7437	3650	0.0039	-
0.7539	3700	0.0059	-
0.7641	3750	0.0058	-
0.7742	3800	0.0022	-
0.7844	3850	0.0064	-
0.7946	3900	0.0033	-
0.8048	3950	0.0037	-
0.8150	4000	0.003	-
0.8252	4050	0.0063	-
0.8354	4100	0.0047	-
0.8456	4150	0.0056	-
0.8557	4200	0.0026	-
0.8659	4250	0.0028	-
0.8761	4300	0.0066	-
0.8863	4350	0.0026	-
0.8965	4400	0.0045	-
0.9067	4450	0.0052	-
0.9169	4500	0.0017	-
0.9271	4550	0.0056	-
0.9372	4600	0.0043	-
0.9474	4650	0.0034	-
0.9576	4700	0.0042	-
0.9678	4750	0.0031	-
0.9780	4800	0.004	-
0.9882	4850	0.0064	-
0.9984	4900	0.0033	-

Framework Versions

Python: 3.12.13
SetFit: 1.1.0
Sentence Transformers: 3.1.1
Transformers: 4.45.2
PyTorch: 2.10.0+cu128
Datasets: 3.2.0
Tokenizers: 0.20.3

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}
}