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

This is a SetFit model trained on the fancyzhx/ag_news dataset 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
Training Dataset: fancyzhx/ag_news

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
Sci/Tech	'MIT Selects Its First Female President (Reuters) Reuters - The Massachusetts Institute of\Technology on Thursday named Susan Hockfield as its first\female president, a breakthrough for the world-renowned school\that has churned out more than 50 Nobel prizewinners but which\ranks below the national average for its percentage of female\students.' "New PC Is Created Just for Teenagers (AP) AP - This isn't your typical, humdrum, slate-colored computer. Not only is the PC known as the hip-e almost all white, but its screen and keyboard are framed in fuzzy pink fur. Or a leopard skin design. Or a graffiti-themed pattern." 'Cdn X Prize team postpones launch KINDERSLEY, Sask. (CP) -- A Canadian team of engineers competing in an international race to put civilians into space has postponed its first stab at the \$10-million prize.'
Business	'Cold sends oil price above \$44 Oil futures prices have jumped 5 percent higher, climbing above \$44 a barrel in the United States Wednesday after US government data showed a slight decline in crude and heating oil supplies as colder weather in the Northeast drove up ' "2 Cable Giants Set To Bid for Adelphia Comcast Corp. and Time Warner Inc. are planning a joint bid for Adelphia Communications Corp. as part of a deal that could lead to a broad realignment of interests in the cable industry and increase Comcast's already dominant presence in the mid-Atlantic region." 'Piper Rudnick to Merge With Big British Firm Piper Rudnick Gray Cary LLP, a law firm with major operations in Washington, agreed over the weekend to merge with British firm DLA LLP, creating one of the largest combinations ever of law firms from different countries.'
World	'50 Hurt During Opposition Strike in Bangladesh Capital In Bangladesh, at least 50 people have been injured in clashes with authorities during a general strike called by the main opposition party.' "No Time Frame for Yuan Move -China WASHINGTON (Reuters) - There is no time frame for a shift away from Beijing's tight currency peg and the United States is mistaken if it thinks it will reap big benefits from a move, a top Chinese central bank official said on Sunday." 'American Forces Bomb Site in Fallujah (AP) AP - American forces bombed a site in the insurgent stronghold of Fallujah early Tuesday where several militants loyal to terror mastermind Abu Musab al-Zarqawi were believed to be holed up, the U.S. military said.'
Sports	'Blue Jays activate Halladay from DL Bronx, NY (Sports Network) - The Toronto Blue Jays activated pitcher Roy Halladay from the 15-day disabled list and he is expected to start Tuesday #39;s game against the New York Yankees.' 'Trinidad climbs off canvas to keep title options open Felix Trinidad returned to the ring after more than two years to score a thrilling eighth-round stoppage of Ricardo Mayorga at New York #39;s Madison Square Garden in a non- title bout being described as one of the fights of the year.' 'Broncos #39; rout frustrates Saints owner NEW ORLEANS -- Another blowout loss left New Orleans owner Tom Benson furious and wondering whether his team belonged in the NFL. Reuben Droughns rushed for 166 yards and a touchdown, Jake Plummer threw for '

Evaluation

Metrics

Label	Accuracy
all	0.7647

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("Google gets a bounce, ends its first day up 18 percent Shares of Google leaped \$15.34, or 18 percent, to \$100.34 on the Nasdaq exchange yesterday in an opening day of trading that harkened back to the wild run-ups of the dot-com era.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	14	39.5694	62

Label	Training Sample Count
World	28
Sports	16
Business	18
Sci/Tech	10

Training Hyperparameters

batch_size: (16, 16)
num_epochs: (5, 5)
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
0.0043	1	0.4093	-
0.2146	50	0.2095	-
0.4292	100	0.0329	-
0.6438	150	0.0008	-
0.8584	200	0.0002	-
1.0	233	-	0.1542
1.0730	250	0.0002	-
1.2876	300	0.0001	-
1.5021	350	0.0002	-
1.7167	400	0.0001	-
1.9313	450	0.0001	-
2.0	466	-	0.1631
2.1459	500	0.0001	-
2.3605	550	0.0001	-
2.5751	600	0.0001	-
2.7897	650	0.0001	-
3.0	699	-	0.1648
3.0043	700	0.0001	-
3.2189	750	0.0001	-
3.4335	800	0.0001	-
3.6481	850	0.0001	-
3.8627	900	0.0001	-
4.0	932	-	0.1663
4.0773	950	0.0001	-
4.2918	1000	0.0	-
4.5064	1050	0.0	-
4.7210	1100	0.0001	-
4.9356	1150	0.0001	-
5.0	1165	-	0.1648

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.9.19
SetFit: 1.1.0.dev0
Sentence Transformers: 3.0.1
Transformers: 4.39.0
PyTorch: 2.4.0
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}
}

kenhktsui
/

setfit_test_ag_news