SetFit with sentence-transformers/paraphrase-mpnet-base-v2

This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/paraphrase-mpnet-base-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/paraphrase-mpnet-base-v2
Classification head: a LogisticRegression instance
Maximum Sequence Length: 512 tokens
Number of Classes: 4 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
2	'Lockheed Martin Defends Polish Investment Lockheed Martin Corp. defended itself Friday against criticism it was moving too slowly in investing \$6 billion in Poland - a commitment that helped win the US company an order for 48 F-16 fighter jets over its European competitors.' 'C amp;W to sell Japanese arm to Softbank Cable amp; Wireless has agreed to sell its Japanese unit to Softbank for 72.4 million pounds. The deal, under which the Japanese Internet communications company will assume debt worth 9.5 million ' 'Martha Stewart Living to replace CEO Martha Stewart Living Omnimedia is expected to name former ABC Entertainment President Susan Lyne as its new chief executive, replacing Sharon Patrick, according to a report on New York Magazines Web site on Thursday.'
0	' #39;Resolution of J amp;K issue will be biggest CBM #39; Islamabad: Emphasising the need for quot;sincerity quot; and quot;flexibility quot; to resolve all outstanding issues with India, including the Kashmir problem, Pakistan on Sunday said the recent meeting involving the Prime Ministers of both countries has helped in ' 'Cannabis chemical pregnancy link A cannabis-like chemical may be important for normal pregnancy, researchers believe.' 'Victims buried alive in Japan quake A series of powerful earthquakes has killed at least 18 people and injured more than 800 people in northern Japan. The first quake struck on Saturday in Niigata prefecture, 200km north of Tokyo, followed by strong aftershocks.'
3	'Microsoft Gets Good Grades on SP2 Microsoft has begun sending Windows XP Service Pack 2 to home users via of its automatic update system. Despite a few flaws that already have been found in the massive patch, the update will strengthen system security for most Windows XP ...' 'Aggregator Sites: One-Stop Shopping? When Che Carsner wanted to find discount airline tickets for his parents from the Miami area to New York, he knew where to look. The Manhattan real estate agent logged onto Kayak.com, a new online travel aggregator, and typed in some dates. Within seconds, dozens of options appeared. Among them: a \$140 round-trip fare from Fort Lauderdale, Fla., to LaGuardia airport on US Airways, which he booked.' 'AOL to have desktop searching in new browser America Online will have a desktop search capability in a new browser the company is now beta-testing, an AOL spokeswoman said Friday.'
1	'Yankees hold off Blue Jays Derek Jeter, Hideki Matsui, and Bernie Williams each homered and the New York Yankees held on for an 8-7 victory over the Blue Jays last night in Toronto. Esteban Loaiza failed to make it out of the fifth inning in another poor start since being acquired from the Chicago White Sox in a July 31 trade for Jose Contreras.' 'Getting Greene #39;s GOAT a chore worthy of an Olympic medal The tattoo on Maurice Greene #39;s right shoulder succinctly sums up where he feels he ranks in the world of sprinting. quot;Greatest of all time, quot; he will shout after a victory while thumping it.' 'Klitschko retains title Vitali Klitschko has knocked out Danny Williams in the eighth round to retain his WBC heavyweight crown and become the premier champion in the division.'

Evaluation

Metrics

Label	Accuracy
all	0.8726

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("vidhi0206/setfit-paraphrase-mpnet-ag_news_v2")
# Run inference
preds = model("Mauresmo confident of LA victory Amelie Mauresmo insists she can win the Tour Championships this week and finish the year as world number one. The Frenchwoman could overtake Lindsay Davenport with a win in Los Angeles.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	15	38.1953	73

Label	Training Sample Count
0	64
1	64
2	64
3	64

Training Hyperparameters

batch_size: (8, 8)
num_epochs: (1, 1)
max_steps: -1
sampling_strategy: oversampling
num_iterations: 20
body_learning_rate: (2e-05, 2e-05)
head_learning_rate: 2e-05
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: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0008	1	0.3712	-
0.0391	50	0.2353	-
0.0781	100	0.1091	-
0.1172	150	0.0898	-
0.1562	200	0.0054	-
0.1953	250	0.0103	-
0.2344	300	0.0051	-
0.2734	350	0.0081	-
0.3125	400	0.0007	-
0.3516	450	0.0003	-
0.3906	500	0.0003	-
0.4297	550	0.0005	-
0.4688	600	0.0003	-
0.5078	650	0.0001	-
0.5469	700	0.0002	-
0.5859	750	0.0001	-
0.625	800	0.0001	-
0.6641	850	0.0001	-
0.7031	900	0.0001	-
0.7422	950	0.0001	-
0.7812	1000	0.0002	-
0.8203	1050	0.0002	-
0.8594	1100	0.0001	-
0.8984	1150	0.0002	-
0.9375	1200	0.0001	-
0.9766	1250	0.0001	-

Framework Versions

Python: 3.8.10
SetFit: 1.0.3
Sentence Transformers: 2.3.1
Transformers: 4.37.2
PyTorch: 2.2.0+cu121
Datasets: 2.17.0
Tokenizers: 0.15.1

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}
}

vidhi0206
/

setfit-paraphrase-mpnet-ag_news_v2