Francesco-A/setfit-all-mpnet-base-v2-non-augmented_dataset-133-shot-just_transition-v1.4.1

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
1	'Does that mean that a fair transition must be ensured through taxation, including with a capital tax for the most wealthy?' 'In fact, there are alternatives, there is a need for motivation to create reasonable parallel opportunities for job creation during a gradual transition.' 'We show that it is possible to combine ecological sustainability with welfare, justice and development.'
0	'As a representative of the Center Party, I am convinced that a transition to a fossil-independent transport sector and the fleet of vehicles is both necessary and possible.' 'Natural solutions supporting the green digital transition aim to mitigate and adapt to climate change.' 'Such a project is at the heart of the ecological transition: the Government, as well as parliamentarians and all the actors involved in this concession, have shown their commitment to this model, which is very innovative, and their ambition to accompany projects at the crossroads of these various issues.'

Evaluation

Metrics

Label	Accuracy
all	0.9375

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("Francesco-A/setfit-all-mpnet-base-v2-non-augmented_dataset-133-shot-just_transition-v1.4.1")
# Run inference
preds = model("The protection of protected areas and nature reserves is essential to conserve biodiversity and preserve wild habitats.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	5	31.4436	120

Label	Training Sample Count
0	133
1	133

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
• l2_weight: 0.01
• seed: 1234
• eval_max_steps: -1
• load_best_model_at_end: True

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0009	1	0.2933	-
0.0449	50	0.2605	-
0.0898	100	0.2551	-
0.1346	150	0.2467	-
0.1795	200	0.233	-
0.2244	250	0.1117	-
0.2693	300	0.0049	-
0.3142	350	0.0007	-
0.3591	400	0.0004	-
0.4039	450	0.0003	-
0.4488	500	0.0002	-
0.4937	550	0.0002	-
0.5386	600	0.0002	-
0.5835	650	0.0002	-
0.6284	700	0.0001	-
0.6732	750	0.0001	-
0.7181	800	0.0001	-
0.7630	850	0.0001	-
0.8079	900	0.0001	-
0.8528	950	0.0001	-
0.8977	1000	0.0001	-
0.9425	1050	0.0001	-
0.9874	1100	0.0001	-
1.0	1114	-	0.0938
1.0323	1150	0.0001	-
1.0772	1200	0.0001	-
1.1221	1250	0.0001	-
1.1670	1300	0.0001	-
1.2118	1350	0.0001	-
1.2567	1400	0.0001	-
1.3016	1450	0.0001	-
1.3465	1500	0.0001	-
1.3914	1550	0.0001	-
1.4363	1600	0.0	-
1.4811	1650	0.0	-
1.5260	1700	0.0	-
1.5709	1750	0.0	-
1.6158	1800	0.0	-
1.6607	1850	0.0	-
1.7056	1900	0.0	-
1.7504	1950	0.0	-
1.7953	2000	0.0	-
1.8402	2050	0.0	-
1.8851	2100	0.0	-
1.9300	2150	0.0	-
1.9749	2200	0.0	-
2.0	2228	-	0.0951
2.0197	2250	0.0003	-
2.0646	2300	0.0012	-
2.1095	2350	0.0005	-
2.1544	2400	0.001	-
2.1993	2450	0.0001	-
2.2442	2500	0.0001	-
2.2890	2550	0.0001	-
2.3339	2600	0.0001	-
2.3788	2650	0.0001	-
2.4237	2700	0.0001	-
2.4686	2750	0.0001	-
2.5135	2800	0.0	-
2.5583	2850	0.0001	-
2.6032	2900	0.0	-
2.6481	2950	0.0	-
2.6930	3000	0.0	-
2.7379	3050	0.0	-
2.7828	3100	0.0	-
2.8276	3150	0.0	-
2.8725	3200	0.0	-
2.9174	3250	0.0	-
2.9623	3300	0.0	-
3.0	3342	-	0.0964

Framework Versions

• Python: 3.10.14
• SetFit: 1.1.0
• Sentence Transformers: 3.3.1
• Transformers: 4.44.0
• PyTorch: 2.4.0
• Datasets: 2.21.0
• Tokenizers: 0.19.1

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}
}