SetFit with Maltehb/danish-bert-botxo

This is a SetFit model that can be used for Text Classification. This SetFit model uses Maltehb/danish-bert-botxo as the Sentence Transformer embedding model. A OneVsRestClassifier 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: Maltehb/danish-bert-botxo
Classification head: a OneVsRestClassifier instance
Maximum Sequence Length: 512 tokens

Model Sources

Repository: SetFit on GitHub
Paper: Efficient Few-Shot Learning Without Prompts
Blogpost: SetFit: Efficient Few-Shot Learning Without Prompts

Evaluation

Metrics

Label	Accuracy
all	0.7317

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("OBech/IngroupOutgroup2")
# Run inference
preds = model("Jeg håber jeg igen kan få opbakning og tillid til at blive folketingsmedlem. Jeg kæmper for hjemstavnen. Jeg bor og lever i Vestjylland.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	94.5901	380

Training Hyperparameters

batch_size: (8, 8)
num_epochs: (2, 2)
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
0.0005	1	0.2605	-
0.0235	50	0.3094	-
0.0471	100	0.2222	-
0.0706	150	0.2855	-
0.0941	200	0.1699	-
0.1176	250	0.1467	-
0.1412	300	0.152	-
0.1647	350	0.2407	-
0.1882	400	0.0391	-
0.2118	450	0.0165	-
0.2353	500	0.0009	-
0.2588	550	0.0004	-
0.2824	600	0.0014	-
0.3059	650	0.0006	-
0.3294	700	0.0001	-
0.3529	750	0.0007	-
0.3765	800	0.0002	-
0.4	850	0.0004	-
0.4235	900	0.0003	-
0.4471	950	0.0001	-
0.4706	1000	0.0001	-
0.4941	1050	0.0002	-
0.5176	1100	0.0002	-
0.5412	1150	0.0005	-
0.5647	1200	0.0002	-
0.5882	1250	0.0002	-
0.6118	1300	0.062	-
0.6353	1350	0.0004	-
0.6588	1400	0.0377	-
0.6824	1450	0.0001	-
0.7059	1500	0.0001	-
0.7294	1550	0.0002	-
0.7529	1600	0.0001	-
0.7765	1650	0.0009	-
0.8	1700	0.0002	-
0.8235	1750	0.0003	-
0.8471	1800	0.0001	-
0.8706	1850	0.0068	-
0.8941	1900	0.0002	-
0.9176	1950	0.0001	-
0.9412	2000	0.0	-
0.9647	2050	0.0002	-
0.9882	2100	0.0	-
1.0	2125	-	0.205
1.0118	2150	0.0164	-
1.0353	2200	0.0002	-
1.0588	2250	0.0	-
1.0824	2300	0.0001	-
1.1059	2350	0.0	-
1.1294	2400	0.0001	-
1.1529	2450	0.0001	-
1.1765	2500	0.036	-
1.2	2550	0.0078	-
1.2235	2600	0.0002	-
1.2471	2650	0.0088	-
1.2706	2700	0.0336	-
1.2941	2750	0.0	-
1.3176	2800	0.0001	-
1.3412	2850	0.0387	-
1.3647	2900	0.0	-
1.3882	2950	0.0042	-
1.4118	3000	0.0001	-
1.4353	3050	0.0	-
1.4588	3100	0.0001	-
1.4824	3150	0.0001	-
1.5059	3200	0.0001	-
1.5294	3250	0.002	-
1.5529	3300	0.0001	-
1.5765	3350	0.0055	-
1.6	3400	0.0002	-
1.6235	3450	0.0	-
1.6471	3500	0.0	-
1.6706	3550	0.0	-
1.6941	3600	0.0	-
1.7176	3650	0.0001	-
1.7412	3700	0.0347	-
1.7647	3750	0.0	-
1.7882	3800	0.0	-
1.8118	3850	0.0	-
1.8353	3900	0.0001	-
1.8588	3950	0.0	-
1.8824	4000	0.0001	-
1.9059	4050	0.0	-
1.9294	4100	0.0001	-
1.9529	4150	0.0073	-
1.9765	4200	0.0001	-
2.0	4250	0.0	0.2099

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.13
SetFit: 1.0.3
Sentence Transformers: 2.3.0
Transformers: 4.39.0
PyTorch: 2.1.2
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}
}

OBech
/

IngroupOutgroup2