davidadamczyk/setfit-model-2

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

This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/all-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:

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-mpnet-base-v2
• Classification head: a LogisticRegression instance
• Maximum Sequence Length: 384 tokens
• Number of Classes: 2 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
yes	'Ken The FBI and DOJ should open an investigation into Russian interference in the 2022 election.\n' "But you still haven't mentioned the crucial upcoming elections in Czechia, which cold alter the balance in Eastern/Central Europe.\n" 'factsonly She won the 2022 election. She beat at least one Dem primary opponent and beat her Republican opponent by a decent margin in the general election.\n'
no	"Sean Who needs a source when you have Trump's well documented relationship with Putin?\n" 'After a years-long crime spree by Donald Trump, his children, and his accomplices, we're still waiting for indictments. Why? Why is this so hard? The man who said, "Russia, if you're listening..." has openly and loudly ignored the law, the constitution, precedent, tradition, common decency and common sense for years, and yet we're still waiting for some part of his manifold misdeeds to land him in the docket. Again, why? Why?! There is so much evidence against him, it is impossible to see why he hasn't been arrested and charged for sedition, insurrection, money laundering, violating the Espionage Act, the Presidential Records Act, payoffs to hide his adulterous affairs, and other crimes up to and including attempting to mastermind a coup. There is no Witch Hunt. There's a just an inexplicably as-yet unindicted multiple felon who continues to grift dollars out of his hoodwinked followers.I am beginning to wonder if the DOJ has forgotten what upholding the law means, or if it is just the person who runs the DOJ.Donald Trump is not the only person to have questions that need to be answered: so does Merrick Garland -- and foremost amongst them is, 'What's the hold up?'\n' "Most writers just imitate what they've read. They repeat formulas and replicate familiar sentence structures. Most TV could be written by ChatGPT. So it seems like ChatGPT writes pretty much like 90 percent of writers in a creative writing class. And 90 percent of readers don't want writing that pushes creative limits—look at the success of Colleen Hoover. I'd don't see why something like ChatGPT couldn't write her books. I don't mean that to be insulting—I do doubt an AI book would touch hearts as hers apparently do because it would lack her ineffable humanity. But even if an AI novel became a popular success, it wouldn't mean that AI had bested Nabokov or Woolf or DFW or … well, it's a very large list, and I'm not even claiming these as anything more than the first three whose names came to mind.(And in answer to Elon, sure, if I had to choose, I guess I'd rather live under the rule of Marcus Aurelius than Caligula's. But in fact I wouldn't get a vote on that, and I'd rather not live under an emperor at all.)\n"

Evaluation

Metrics

Label	Accuracy
all	0.8

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("davidadamczyk/setfit-model-2")
# Run inference
preds = model("Aaron 100 percent. citizens united was a huge win for Russian citizen Vlad and Chinese citizen Xi.
")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	6	80.325	276

Label	Training Sample Count
no	18
yes	22

Training Hyperparameters

• batch_size: (16, 16)
• num_epochs: (1, 1)
• max_steps: -1
• sampling_strategy: oversampling
• num_iterations: 120
• 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
• l2_weight: 0.01
• seed: 42
• eval_max_steps: -1
• load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0017	1	0.4496	-
0.0833	50	0.1797	-
0.1667	100	0.0034	-
0.25	150	0.0003	-
0.3333	200	0.0002	-
0.4167	250	0.0002	-
0.5	300	0.0001	-
0.5833	350	0.0001	-
0.6667	400	0.0001	-
0.75	450	0.0001	-
0.8333	500	0.0001	-
0.9167	550	0.0001	-
1.0	600	0.0001	-

Framework Versions

• Python: 3.10.13
• SetFit: 1.1.0
• Sentence Transformers: 3.0.1
• Transformers: 4.45.2
• PyTorch: 2.4.0+cu124
• Datasets: 2.21.0
• Tokenizers: 0.20.0

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}
}