SetFit with BAAI/bge-small-en-v1.5

This is a SetFit model that can be used for Text Classification. This SetFit model uses BAAI/bge-small-en-v1.5 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: BAAI/bge-small-en-v1.5
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
1	'board diversity affects ceo pay' 'perceptions of formal learning affects entrepreneurship intention' 'proactiveness affects entrepreneurship intention'
0	'sales and takeovers affects entrepreneurship intention' 'uk affects entrepreneurship intention' 'economics affects entrepreneurship intention'

Evaluation

Metrics

Label	Accuracy
all	0.9059

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("abehandlerorg/setfit")
# Run inference
preds = model("sales affects ceo pay")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	4	5.4307	12

Label	Training Sample Count
0	168
1	171

Training Hyperparameters

batch_size: (32, 32)
num_epochs: (4, 4)
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: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0006	1	0.3133	-
0.0277	50	0.289	-
0.0553	100	0.2506	-
0.0830	150	0.2243	-
0.1107	200	0.2388	-
0.1384	250	0.2084	-
0.1660	300	0.1316	-
0.1937	350	0.0142	-
0.2214	400	0.0065	-
0.2490	450	0.0037	-
0.2767	500	0.003	-
0.3044	550	0.002	-
0.3320	600	0.0018	-
0.3597	650	0.0026	-
0.3874	700	0.0013	-
0.4151	750	0.0012	-
0.4427	800	0.0284	-
0.4704	850	0.0145	-
0.4981	900	0.0053	-
0.5257	950	0.0075	-
0.5534	1000	0.005	-
0.5811	1050	0.0008	-
0.6087	1100	0.0008	-
0.6364	1150	0.0008	-
0.6641	1200	0.0007	-
0.6918	1250	0.0008	-
0.7194	1300	0.0009	-
0.7471	1350	0.0007	-
0.7748	1400	0.0008	-
0.8024	1450	0.0006	-
0.8301	1500	0.0006	-
0.8578	1550	0.0192	-
0.8854	1600	0.0005	-
0.9131	1650	0.002	-
0.9408	1700	0.0204	-
0.9685	1750	0.0039	-
0.9961	1800	0.0007	-
1.0238	1850	0.0005	-
1.0515	1900	0.0004	-
1.0791	1950	0.0005	-
1.1068	2000	0.0006	-
1.1345	2050	0.0004	-
1.1621	2100	0.0006	-
1.1898	2150	0.0004	-
1.2175	2200	0.0004	-
1.2452	2250	0.0018	-
1.2728	2300	0.0041	-
1.3005	2350	0.0004	-
1.3282	2400	0.0107	-
1.3558	2450	0.0005	-
1.3835	2500	0.0004	-
1.4112	2550	0.0004	-
1.4388	2600	0.0167	-
1.4665	2650	0.0068	-
1.4942	2700	0.0004	-
1.5219	2750	0.0064	-
1.5495	2800	0.0041	-
1.5772	2850	0.0004	-
1.6049	2900	0.0003	-
1.6325	2950	0.0004	-
1.6602	3000	0.0004	-
1.6879	3050	0.0003	-
1.7156	3100	0.0057	-
1.7432	3150	0.0044	-
1.7709	3200	0.0004	-
1.7986	3250	0.0166	-
1.8262	3300	0.0004	-
1.8539	3350	0.0032	-
1.8816	3400	0.0133	-
1.9092	3450	0.0003	-
1.9369	3500	0.0003	-
1.9646	3550	0.0052	-
1.9923	3600	0.0004	-
2.0199	3650	0.004	-
2.0476	3700	0.0003	-
2.0753	3750	0.0054	-
2.1029	3800	0.0057	-
2.1306	3850	0.0004	-
2.1583	3900	0.0272	-
2.1859	3950	0.0003	-
2.2136	4000	0.006	-
2.2413	4050	0.0044	-
2.2690	4100	0.0003	-
2.2966	4150	0.0167	-
2.3243	4200	0.0048	-
2.3520	4250	0.0086	-
2.3796	4300	0.0051	-
2.4073	4350	0.0003	-
2.4350	4400	0.0037	-
2.4626	4450	0.0003	-
2.4903	4500	0.0021	-
2.5180	4550	0.0003	-
2.5457	4600	0.004	-
2.5733	4650	0.0025	-
2.6010	4700	0.0003	-
2.6287	4750	0.0003	-
2.6563	4800	0.0003	-
2.6840	4850	0.0031	-
2.7117	4900	0.0168	-
2.7393	4950	0.0019	-
2.7670	5000	0.004	-
2.7947	5050	0.0003	-
2.8224	5100	0.0003	-
2.8500	5150	0.003	-
2.8777	5200	0.0003	-
2.9054	5250	0.0003	-
2.9330	5300	0.0171	-
2.9607	5350	0.0003	-
2.9884	5400	0.0162	-
3.0160	5450	0.0143	-
3.0437	5500	0.0134	-
3.0714	5550	0.0133	-
3.0991	5600	0.0003	-
3.1267	5650	0.0003	-
3.1544	5700	0.0093	-
3.1821	5750	0.0003	-
3.2097	5800	0.0003	-
3.2374	5850	0.0003	-
3.2651	5900	0.0003	-
3.2928	5950	0.0003	-
3.3204	6000	0.0029	-
3.3481	6050	0.0126	-
3.3758	6100	0.0003	-
3.4034	6150	0.0002	-
3.4311	6200	0.0003	-
3.4588	6250	0.0062	-
3.4864	6300	0.0002	-
3.5141	6350	0.0002	-
3.5418	6400	0.0003	-
3.5695	6450	0.0002	-
3.5971	6500	0.0041	-
3.6248	6550	0.0465	-
3.6525	6600	0.0148	-
3.6801	6650	0.0181	-
3.7078	6700	0.0037	-
3.7355	6750	0.0002	-
3.7631	6800	0.0003	-
3.7908	6850	0.0003	-
3.8185	6900	0.0034	-
3.8462	6950	0.0002	-
3.8738	7000	0.0148	-
3.9015	7050	0.0002	-
3.9292	7100	0.0003	-
3.9568	7150	0.0002	-
3.9845	7200	0.0003	-

Framework Versions

Python: 3.10.12
SetFit: 1.0.3
Sentence Transformers: 2.7.0
Transformers: 4.40.1
PyTorch: 2.2.1+cu121
Datasets: 2.19.1
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}
}

abehandlerorg
/

setfit