twright8/setfit-oversample-labels-lobbying

SetFit

This is a SetFit model that can be used for Text Classification. A SetFitHead 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
• Classification head: a SetFitHead instance
• Maximum Sequence Length: 512 tokens
• Number of Classes: 4 classes

Model Sources

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

Evaluation

Metrics

Label	F1	Accuracy
all	0.9057	0.9573

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("twright8/setfit-oversample-labels-lobbying")
# Run inference
preds = model("Electricity market")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	2	21.5644	153

Training Hyperparameters

• batch_size: (16, 2)
• num_epochs: (6, 9)
• max_steps: -1
• sampling_strategy: oversampling
• body_learning_rate: (7.928034854554858e-06, 2.7001088851580374e-05)
• head_learning_rate: 0.009321171293151879
• loss: CoSENTLoss
• distance_metric: cosine_distance
• margin: 0.25
• end_to_end: True
• 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
0.0018	1	8.669	-
0.0880	50	8.6617	-
0.1761	100	12.5549	-
0.2641	150	3.1895	-
0.3521	200	16.3181	-
0.4401	250	0.7513	-
0.5282	300	4.6653	-
0.0018	1	0.0059	-
0.0880	50	3.4564	-
0.1761	100	0.5523	-
0.2641	150	0.2372	-
0.3521	200	4.288	-
0.4401	250	0.0027	-
0.5282	300	0.0002	-
0.6162	350	0.0002	-
0.7042	400	0.0001	-
0.7923	450	0.0015	-
0.8803	500	3.5596	-
0.9683	550	0.0	-
1.0	568	-	10.2261
1.0563	600	0.0	-
1.1444	650	0.0011	-
1.2324	700	0.0013	-
1.3204	750	0.0037	-
1.4085	800	0.0013	-
1.4965	850	0.0002	-
1.5845	900	0.0	-
1.6725	950	0.0	-
1.7606	1000	0.0001	-
1.8486	1050	0.0001	-
1.9366	1100	0.0001	-
2.0	1136	-	8.4908
2.0246	1150	0.0001	-
2.1127	1200	0.0	-
2.2007	1250	0.0005	-
2.2887	1300	0.0004	-
2.3768	1350	0.0	-
2.4648	1400	0.0009	-
2.5528	1450	0.0	-
2.6408	1500	0.0	-
2.7289	1550	0.0	-
2.8169	1600	0.0	-
2.9049	1650	0.0001	-
2.9930	1700	0.0003	-
3.0	1704	-	8.5594
3.0810	1750	0.0001	-
3.1690	1800	0.0	-
3.2570	1850	0.0002	-
3.3451	1900	0.0001	-
3.4331	1950	0.0	-
3.5211	2000	0.0	-
3.6092	2050	0.0	-
3.6972	2100	0.0	-
3.7852	2150	0.0	-
3.8732	2200	0.0002	-
3.9613	2250	0.0001	-
4.0	2272	-	8.4573
4.0493	2300	0.0	-
4.1373	2350	0.0	-
4.2254	2400	0.0002	-
4.3134	2450	0.0	-
4.4014	2500	0.0003	-
4.4894	2550	0.0001	-
4.5775	2600	0.0001	-
4.6655	2650	0.0001	-
4.7535	2700	0.0001	-
4.8415	2750	0.0001	-
4.9296	2800	0.0012	-
5.0	2840	-	8.6305
5.0176	2850	0.0009	-
5.1056	2900	0.0	-
5.1937	2950	0.0001	-
5.2817	3000	0.0	-
5.3697	3050	0.0	-
5.4577	3100	0.0001	-
5.5458	3150	0.0007	-
5.6338	3200	0.0002	-
5.7218	3250	0.0	-
5.8099	3300	0.0001	-
5.8979	3350	0.0002	-
5.9859	3400	0.0	-
6.0	3408	-	8.9528

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• SetFit: 1.0.3
• Sentence Transformers: 3.0.1
• Transformers: 4.39.0
• PyTorch: 2.3.1+cu118
• 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}
}