This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/paraphrase-multilingual-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-multilingual-mpnet-base-v2
Classification head: a LogisticRegression instance
Maximum Sequence Length: 128 tokens
Number of Classes: 11 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
Administrative	'Bookkeeper' 'Koordinerende sagsbehandler' 'Center Management'
Other	'Strategy & Investments' 'Corporate Auditor' 'Professor'
Marketing	'Marketing Manager' 'Global Marketing Director, Nonwovens' 'Brand Strategist'
Purchasing	'Program Purchasing Leader / Program / Acquisition Buyer' 'Senior Category Manager' 'Fachreferent MatWerk'
Information Technology	'Business Analyst' 'Systemadministrator Clientservices' 'IT Program Manager – Information Security'
Consulting	'Managementberater IT- und Governance' 'Berater' 'Manager Risk Advisory'
Project Management	'Senior Project Manager' 'Responsable développement projets' 'Leiter Projektmanagement / Projektleiter'
Sales	'Federal Account Manager' 'Commercial Director On Trade' 'Sales Manager'
Human Resources	'SVP People & Culture Management and Services; CEO METRO Campus Services GmbH' 'Gruppenleiter HR KPI & Reporting' 'Personalreferent'
Business Development	'Director, Digital Services Business' 'Chief Growth Officer' 'Business Development & Key Account Manager'
Customer Support	'Sr. Customer Support spec.' 'OM/Customer Success Manager' 'Customer Executive & Member of Management Northern Europe'

Evaluation

Metrics

Label	Accuracy
all	0.7343

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

python

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("Luu200/department-classifier-v2")
# Run inference
preds = model("Managing Director")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	1	2.7229	12

Label	Training Sample Count
Administrative	12
Business Development	5
Consulting	18
Customer Support	5
Human Resources	10
Information Technology	28
Marketing	13
Other	169
Project Management	18
Purchasing	10
Sales	26

Training Hyperparameters

batch_size: (16, 16)
num_epochs: (1, 1)
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
l2_weight: 0.01
seed: 42
evaluation_strategy: epoch
eval_max_steps: -1
load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0002	1	0.3802	-
0.0119	50	0.2604	-
0.0237	100	0.2071	-
0.0356	150	0.1352	-
0.0475	200	0.0718	-
0.0594	250	0.0334	-
0.0712	300	0.0377	-
0.0831	350	0.0221	-
0.0950	400	0.0162	-
0.1069	450	0.0097	-
0.1187	500	0.015	-
0.1306	550	0.0082	-
0.1425	600	0.0119	-
0.1544	650	0.0122	-
0.1662	700	0.0052	-
0.1781	750	0.0085	-
0.1900	800	0.0046	-
0.2019	850	0.0027	-
0.2137	900	0.0073	-
0.2256	950	0.0066	-
0.2375	1000	0.0044	-
0.2493	1050	0.0037	-
0.2612	1100	0.0051	-
0.2731	1150	0.0045	-
0.2850	1200	0.0056	-
0.2968	1250	0.0017	-
0.3087	1300	0.0049	-
0.3206	1350	0.0057	-
0.3325	1400	0.0046	-
0.3443	1450	0.0033	-
0.3562	1500	0.0023	-
0.3681	1550	0.002	-
0.3800	1600	0.0018	-
0.3918	1650	0.003	-
0.4037	1700	0.004	-
0.4156	1750	0.0056	-
0.4275	1800	0.0051	-
0.4393	1850	0.004	-
0.4512	1900	0.0038	-
0.4631	1950	0.0035	-
0.4749	2000	0.0026	-
0.4868	2050	0.0026	-
0.4987	2100	0.0045	-
0.5106	2150	0.0006	-
0.5224	2200	0.0036	-
0.5343	2250	0.0015	-
0.5462	2300	0.0004	-
0.5581	2350	0.0016	-
0.5699	2400	0.0025	-
0.5818	2450	0.0025	-
0.5937	2500	0.0024	-
0.6056	2550	0.0014	-
0.6174	2600	0.0035	-
0.6293	2650	0.0026	-
0.6412	2700	0.0036	-
0.6531	2750	0.0003	-
0.6649	2800	0.0042	-
0.6768	2850	0.0039	-
0.6887	2900	0.0026	-
0.7005	2950	0.0042	-
0.7124	3000	0.0056	-
0.7243	3050	0.003	-
0.7362	3100	0.0027	-
0.7480	3150	0.0025	-
0.7599	3200	0.0003	-
0.7718	3250	0.0019	-
0.7837	3300	0.0033	-
0.7955	3350	0.0004	-
0.8074	3400	0.0019	-
0.8193	3450	0.0025	-
0.8312	3500	0.0037	-
0.8430	3550	0.0037	-
0.8549	3600	0.0035	-
0.8668	3650	0.003	-
0.8787	3700	0.0032	-
0.8905	3750	0.0014	-
0.9024	3800	0.0034	-
0.9143	3850	0.0042	-
0.9261	3900	0.0014	-
0.9380	3950	0.0023	-
0.9499	4000	0.0018	-
0.9618	4050	0.0032	-
0.9736	4100	0.0061	-
0.9855	4150	0.0022	-
0.9974	4200	0.0035	-
1.0	4211	-	0.2137

Framework Versions

Python: 3.12.12
SetFit: 1.1.3
Sentence Transformers: 5.2.0
Transformers: 4.57.3
PyTorch: 2.9.0+cu126
Datasets: 4.0.0
Tokenizers: 0.22.2

Citation

BibTeX

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}
}