SetFit with FacebookAI/xlm-roberta-base

This is a SetFit model that can be used for Text Classification. This SetFit model uses FacebookAI/xlm-roberta-base 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: FacebookAI/xlm-roberta-base
Classification head: a LogisticRegression instance
Maximum Sequence Length: 512 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
0	'Uusi maakuntajohtaja haluaa tehdä määrätietoisesti avointa elinkeinopolitiikkaa.' '– Verot eivät saa nousta.' 'Enemmänkin ulkoinen apu, sikäli kun sitä tarvitaan, on luonteeltaan täydentävää, läsnäoloa.'
1	'nan' '– Seuraava laskumarkkina voi olla edessä, kun keskuspankit alkavat vähentää likviditeettiä pois markkinoilta, Rothovius tuumaa.' 'Hallituspuolueista puolestaan huomautettiin, että sotea on yritetty saadaan kuntoon 15 vuotta, josta kokoomus on ollut hallitusvastuussa valtaosan.'

Evaluation

Metrics

Label	Metric
all	0.8210

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("HY-Aalto-DIME/FinnClaim-detect-FinBERT-CF2")
# Run inference
preds = model("Toinen pulma on se, lopahtaako harrastajien into.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	11.0767	29

Label	Training Sample Count
0	727
1	329

Training Hyperparameters

batch_size: (16, 16)
num_epochs: (4, 4)
max_steps: -1
sampling_strategy: oversampling
num_iterations: 6
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: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0013	1	0.4903	-
0.0631	50	0.4847	-
0.1263	100	0.2626	-
0.1894	150	0.2597	-
0.2525	200	0.1716	-
0.3157	250	0.3103	-
0.3788	300	0.0955	-
0.4419	350	0.1019	-
0.5051	400	0.167	-
0.5682	450	0.0886	-
0.6313	500	0.0591	-
0.6944	550	0.161	-
0.7576	600	0.031	-
0.8207	650	0.0947	-
0.8838	700	0.0087	-
0.9470	750	0.0055	-
1.0	792	-	0.2484
1.0101	800	0.0034	-
1.0732	850	0.0037	-
1.1364	900	0.0043	-
1.1995	950	0.0031	-
1.2626	1000	0.0007	-
1.3258	1050	0.002	-
1.3889	1100	0.0004	-
1.4520	1150	0.0021	-
1.5152	1200	0.0005	-
1.5783	1250	0.0002	-
1.6414	1300	0.0009	-
1.7045	1350	0.0002	-
1.7677	1400	0.0392	-
1.8308	1450	0.0039	-
1.8939	1500	0.0002	-
1.9571	1550	0.0411	-
2.0	1584	-	0.2935
2.0202	1600	0.0002	-
2.0833	1650	0.0003	-
2.1465	1700	0.0548	-
2.2096	1750	0.0042	-
2.2727	1800	0.0002	-
2.3359	1850	0.0002	-
2.3990	1900	0.0001	-
2.4621	1950	0.0001	-
2.5253	2000	0.0003	-
2.5884	2050	0.0001	-
2.6515	2100	0.0001	-
2.7146	2150	0.0002	-
2.7778	2200	0.0002	-
2.8409	2250	0.0001	-
2.9040	2300	0.0002	-
2.9672	2350	0.0002	-
3.0	2376	-	0.3097
3.0303	2400	0.0001	-
3.0934	2450	0.0	-
3.1566	2500	0.0001	-
3.2197	2550	0.0001	-
3.2828	2600	0.0001	-
3.3460	2650	0.0001	-
3.4091	2700	0.0001	-
3.4722	2750	0.0001	-
3.5354	2800	0.0001	-
3.5985	2850	0.0001	-
3.6616	2900	0.0001	-
3.7247	2950	0.0001	-
3.7879	3000	0.0001	-
3.8510	3050	0.0001	-
3.9141	3100	0.0001	-
3.9773	3150	0.0001	-
4.0	3168	-	0.2759

Framework Versions

Python: 3.11.9
SetFit: 1.0.3
Sentence Transformers: 3.2.0
Transformers: 4.44.0
PyTorch: 2.4.0+cu124
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}
}

HY-Aalto-DIME
/

FinnClaim-detect-FinBERT-CF2