tomaarsen/setfit-all-MiniLM-L6-v2-sst2-32-shot

SetFit with sentence-transformers/all-MiniLM-L6-v2 on sst2

This is a SetFit model trained on the sst2 dataset 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:

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: sentence-transformers/all-MiniLM-L6-v2
• Classification head: a LogisticRegression instance
• Maximum Sequence Length: 256 tokens
• Number of Classes: 2 classes
• Training Dataset: sst2
• Language: en
• License: apache-2.0

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	'a tough pill to swallow and ' 'indignation ' 'that the typical hollywood disregard for historical truth and realism is at work here '
positive	"a moving experience for people who have n't read the book " 'in the best possible senses of both those words ' 'to serve the work especially well '

Evaluation

Metrics

Label	Accuracy
all	0.7513

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 🤗 Hub
model = SetFitModel.from_pretrained("tomaarsen/setfit-all-MiniLM-L6-v2-sst2-8-shot")
# Run inference
preds = model("a fast , funny , highly enjoyable movie . ")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	2	10.2812	36

Label	Training Sample Count
negative	32
positive	32

Training Hyperparameters

• batch_size: (16, 16)
• num_epochs: (3, 3)
• 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
• load_best_model_at_end: True

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0076	1	0.3787	-
0.0758	10	0.2855	-
0.1515	20	0.3458	0.29
0.2273	30	0.2496	-
0.3030	40	0.2398	0.2482
0.3788	50	0.2068	-
0.4545	60	0.2471	0.244
0.5303	70	0.2053	-
0.6061	80	0.1802	0.2361
0.6818	90	0.0767	-
0.7576	100	0.0279	0.2365
0.8333	110	0.0192	-
0.9091	120	0.0095	0.2527
0.9848	130	0.0076	-
1.0606	140	0.0082	0.2651
1.1364	150	0.0068	-
1.2121	160	0.0052	0.2722
1.2879	170	0.0029	-
1.3636	180	0.0042	0.273
1.4394	190	0.0026	-
1.5152	200	0.0036	0.2761
1.5909	210	0.0044	-
1.6667	220	0.0027	0.2796
1.7424	230	0.0025	-
1.8182	240	0.0025	0.2817
1.8939	250	0.003	-
1.9697	260	0.0026	0.2817
2.0455	270	0.0035	-
2.1212	280	0.002	0.2816
2.1970	290	0.0023	-
2.2727	300	0.0016	0.2821
2.3485	310	0.0023	-
2.4242	320	0.0015	0.2838
2.5	330	0.0014	-
2.5758	340	0.002	0.2842
2.6515	350	0.002	-
2.7273	360	0.0013	0.2847
2.8030	370	0.0009	-
2.8788	380	0.0018	0.2857
2.9545	390	0.0016	-

• The bold row denotes the saved checkpoint.

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Carbon Emitted: 0.003 kg of CO2
• Hours Used: 0.072 hours

Training Hardware

• On Cloud: No
• GPU Model: 1 x NVIDIA GeForce RTX 3090
• CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
• RAM Size: 31.78 GB

Framework Versions

• Python: 3.9.16
• SetFit: 1.0.0.dev0
• Sentence Transformers: 2.2.2
• Transformers: 4.29.0
• PyTorch: 1.13.1+cu117
• Datasets: 2.15.0
• Tokenizers: 0.13.3

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}
}