lucienbaumgartner/moralizedMP2

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:

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/paraphrase-mpnet-base-v2
• Classification head: a LogisticRegression instance
• Maximum Sequence Length: 512 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
Linguistic (in)felicity	'because the second statement negates what was stated in the first part of the sentence' 'there is a logic conflict in the statement that renders it bizarre and nonsensical.' 'there was a contradiction of statements if read at face value, however, it could be read that being homeless is not right in which case the statement would make sense. it is unclear.'
Enrichment / reinterpretation	'the statement recognised the objective compassion but the opinion contradicted it' "because while it is compassionate to help the homeless people don't always do it out of compassion." 'it could be the way how homeless are helped. there could be better ways to handle that'
Lack of understanding / clear misunderstanding	'it simply sounded stupid. i doubt it makes any sense' 'it statement didnt make any sense, for us to better understand, tom needs to further explain his reason for stating why its not cruel after first saying it is' 'it sounds very contradictory'

Evaluation

Metrics

Label	Accuracy	Precision	Recall	F1
all	0.8684	0.5643	0.5629	0.5626

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("it contradicted itself")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	2	16.6447	92

Label	Training Sample Count
Enrichment / reinterpretation	31
Lack of understanding / clear misunderstanding	10
Linguistic (in)felicity	111

Training Hyperparameters

• batch_size: (16, 16)
• num_epochs: (10, 10)
• 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
• l2_weight: 0.01
• seed: 3786
• eval_max_steps: -1
• load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0026	1	0.2539	-
0.1316	50	0.2248	-
0.2632	100	0.1681	-
0.3947	150	0.0854	-
0.5263	200	0.0128	-
0.6579	250	0.0074	-
0.7895	300	0.0017	-
0.9211	350	0.0021	-
1.0526	400	0.0024	-
1.1842	450	0.0004	-
1.3158	500	0.0011	-
1.4474	550	0.0016	-
1.5789	600	0.0003	-
1.7105	650	0.0002	-
1.8421	700	0.0002	-
1.9737	750	0.0002	-
2.1053	800	0.0002	-
2.2368	850	0.0002	-
2.3684	900	0.0002	-
2.5	950	0.0001	-
2.6316	1000	0.0001	-
2.7632	1050	0.0001	-
2.8947	1100	0.0001	-
3.0263	1150	0.0001	-
3.1579	1200	0.0001	-
3.2895	1250	0.0001	-
3.4211	1300	0.0001	-
3.5526	1350	0.0001	-
3.6842	1400	0.0001	-
3.8158	1450	0.0001	-
3.9474	1500	0.0001	-
4.0789	1550	0.0001	-
4.2105	1600	0.0001	-
4.3421	1650	0.0001	-
4.4737	1700	0.0001	-
4.6053	1750	0.0001	-
4.7368	1800	0.0001	-
4.8684	1850	0.0001	-
5.0	1900	0.0001	-
5.1316	1950	0.0001	-
5.2632	2000	0.0001	-
5.3947	2050	0.0001	-
5.5263	2100	0.0001	-
5.6579	2150	0.0001	-
5.7895	2200	0.0001	-
5.9211	2250	0.0001	-
6.0526	2300	0.0001	-
6.1842	2350	0.0001	-
6.3158	2400	0.0001	-
6.4474	2450	0.0001	-
6.5789	2500	0.0001	-
6.7105	2550	0.0001	-
6.8421	2600	0.0001	-
6.9737	2650	0.0001	-
7.1053	2700	0.0001	-
7.2368	2750	0.0001	-
7.3684	2800	0.0001	-
7.5	2850	0.0001	-
7.6316	2900	0.0001	-
7.7632	2950	0.0001	-
7.8947	3000	0.0001	-
8.0263	3050	0.0001	-
8.1579	3100	0.0001	-
8.2895	3150	0.0001	-
8.4211	3200	0.0001	-
8.5526	3250	0.0001	-
8.6842	3300	0.0001	-
8.8158	3350	0.0001	-
8.9474	3400	0.0012	-
9.0789	3450	0.0003	-
9.2105	3500	0.0001	-
9.3421	3550	0.0001	-
9.4737	3600	0.0001	-
9.6053	3650	0.0001	-
9.7368	3700	0.0001	-
9.8684	3750	0.0001	-
10.0	3800	0.0	-

Framework Versions

• Python: 3.11.9
• SetFit: 1.1.3
• Sentence Transformers: 5.1.0
• Transformers: 4.55.2
• PyTorch: 2.8.0
• Datasets: 4.0.0
• Tokenizers: 0.21.4

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}
}