SetFit with sentence-transformers/paraphrase-mpnet-base-v2

This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/paraphrase-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:

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: sentence-transformers/paraphrase-mpnet-base-v2
Classification head: a LogisticRegression 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

Model Labels

Label	Examples
3	'academic head' 'admin director' 'admin head'
4	'account director' 'area vice president' 'assistant chief executive officer'
2	'account manager' 'admin' 'admin officer'
1	'accountant' 'administrator' 'adviser'

Evaluation

Metrics

Label	Accuracy
all	1.0

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("setfit_model_id")
# Run inference
preds = model("planner")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	2.1124	6

Label	Training Sample Count
1	380
2	107
3	67
4	193

Training Hyperparameters

batch_size: (16, 16)
num_epochs: (3, 3)
max_steps: -1
sampling_strategy: oversampling
num_iterations: 20
body_learning_rate: (2e-05, 2e-05)
head_learning_rate: 2e-05
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
0.0005	1	0.2621	-
0.0268	50	0.2631	-
0.0535	100	0.2043	-
0.0803	150	0.1561	-
0.1071	200	0.203	-
0.1338	250	0.1823	-
0.1606	300	0.1082	-
0.1874	350	0.0702	-
0.2141	400	0.1159	-
0.2409	450	0.0532	-
0.2677	500	0.0767	-
0.2944	550	0.0965	-
0.3212	600	0.0479	-
0.3480	650	0.0353	-
0.3747	700	0.0235	-
0.4015	750	0.0028	-
0.4283	800	0.004	-
0.4550	850	0.0908	-
0.4818	900	0.0078	-
0.5086	950	0.0149	-
0.5353	1000	0.0841	-
0.5621	1050	0.0141	-
0.5889	1100	0.0328	-
0.6156	1150	0.0031	-
0.6424	1200	0.0027	-
0.6692	1250	0.0205	-
0.6959	1300	0.0584	-
0.7227	1350	0.002	-
0.7495	1400	0.0009	-
0.7762	1450	0.0018	-
0.8030	1500	0.001	-
0.8298	1550	0.0004	-
0.8565	1600	0.0008	-
0.8833	1650	0.0006	-
0.9101	1700	0.0021	-
0.9368	1750	0.009	-
0.9636	1800	0.0031	-
0.9904	1850	0.0024	-
1.0171	1900	0.0327	-
1.0439	1950	0.0257	-
1.0707	2000	0.0006	-
1.0974	2050	0.0009	-
1.1242	2100	0.0006	-
1.1510	2150	0.0004	-
1.1777	2200	0.0011	-
1.2045	2250	0.0004	-
1.2313	2300	0.0012	-
1.2580	2350	0.0005	-
1.2848	2400	0.0013	-
1.3116	2450	0.0007	-
1.3383	2500	0.0002	-
1.3651	2550	0.0005	-
1.3919	2600	0.0006	-
1.4186	2650	0.0006	-
1.4454	2700	0.0004	-
1.4722	2750	0.0004	-
1.4989	2800	0.0008	-
1.5257	2850	0.0003	-
1.5525	2900	0.0012	-
1.5792	2950	0.0006	-
1.6060	3000	0.0003	-
1.6328	3050	0.0002	-
1.6595	3100	0.0026	-
1.6863	3150	0.0003	-
1.7131	3200	0.0003	-
1.7398	3250	0.0003	-
1.7666	3300	0.0003	-
1.7934	3350	0.0003	-
1.8201	3400	0.0004	-
1.8469	3450	0.0003	-
1.8737	3500	0.0005	-
1.9004	3550	0.0003	-
1.9272	3600	0.0003	-
1.9540	3650	0.0002	-
1.9807	3700	0.0003	-
2.0075	3750	0.0003	-
2.0343	3800	0.0003	-
2.0610	3850	0.0002	-
2.0878	3900	0.0004	-
2.1146	3950	0.0003	-
2.1413	4000	0.0003	-
2.1681	4050	0.0002	-
2.1949	4100	0.0541	-
2.2216	4150	0.0002	-
2.2484	4200	0.0003	-
2.2752	4250	0.0582	-
2.3019	4300	0.0003	-
2.3287	4350	0.0002	-
2.3555	4400	0.0003	-
2.3822	4450	0.0005	-
2.4090	4500	0.0004	-
2.4358	4550	0.0003	-
2.4625	4600	0.0003	-
2.4893	4650	0.0002	-
2.5161	4700	0.0002	-
2.5428	4750	0.0003	-
2.5696	4800	0.0008	-
2.5964	4850	0.0002	-
2.6231	4900	0.0002	-
2.6499	4950	0.0005	-
2.6767	5000	0.0003	-
2.7034	5050	0.0002	-
2.7302	5100	0.0004	-
2.7570	5150	0.0002	-
2.7837	5200	0.0005	-
2.8105	5250	0.0004	-
2.8373	5300	0.0394	-
2.8640	5350	0.0002	-
2.8908	5400	0.0399	-
2.9176	5450	0.0002	-
2.9443	5500	0.0002	-
2.9711	5550	0.0002	-
2.9979	5600	0.0002	-

Framework Versions

Python: 3.10.12
SetFit: 1.0.3
Sentence Transformers: 3.0.1
Transformers: 4.39.0
PyTorch: 2.3.1+cu121
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}
}

kshitijkutumbe
/

job_level_model