Zlovoblachko/dim3_BAAI_setfit_model

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:

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
• Sentence Transformer body: BAAI/bge-small-en-v1.5
• Classification head: a LogisticRegression instance
• Maximum Sequence Length: 512 tokens
• Number of Classes: 3 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
H	"('that is a waste of time', 'acl') [SEP] Thesis [SEP] ('because most people tend', 'advcl') [SEP] Thesis4" "('and become', 'conj') [SEP] Elaboration [SEP] ('less stressed after the match', 'xcomp') [SEP] Elaboration3" "('than to waste time', 'acl') [SEP] Elaboration [SEP] ('speculating the events', 'advcl') [SEP] Elaboration3"
L	"('So as a basketballer this person might achieve some great results', 'root') [SEP] Argument [SEP] ('by looking on plays of professionals either live', 'advcl') [SEP] Argument2" "('For example my friend Andrew really like bascetball', 'root') [SEP] Elaboration [SEP] ('and NBA players has a big influense on him', 'conj') [SEP] Elaboration2" "('Some of them get inspired', 'root') [SEP] Elaboration [SEP] ('and choose sports as their career', 'conj') [SEP] Elaboration2"
0	"('that sport is an essential part of our life', 'ccomp') [SEP] Conclusion [SEP] ('watching it at home', 'csubj') [SEP] Conclusion1" "('that is a completely useless waste of time', 'ccomp') [SEP] Thesis [SEP] ('of watching such international sport activities as champions league or olympic games', 'acl') [SEP] Introductory1" "('of spending free time intellectually and socially', 'acl') [SEP] Conclusion [SEP] ('Secondly simply not watching', 'root') [SEP] Elaboration1"

Evaluation

Metrics

Label	Accuracy
all	0.8229

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("Zlovoblachko/dim3_BAAI_setfit_model")
# Run inference
preds = model("('that some people watching TV', 'acl') [SEP] Argument [SEP] ('only to relax after work', 'advcl') [SEP] Argument3")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	9	19.0967	35

Label	Training Sample Count
L	105
H	94
0	101

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
• eval_max_steps: -1
• load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0003	1	0.1536	-
0.0133	50	0.277	-
0.0267	100	0.2479	-
0.0400	150	0.2212	-
0.0534	200	0.1652	-
0.0667	250	0.1512	-
0.0801	300	0.1451	-
0.0934	350	0.1342	-
0.1068	400	0.1129	-
0.1201	450	0.1067	-
0.1334	500	0.0813	-
0.1468	550	0.0474	-
0.1601	600	0.0291	-
0.1735	650	0.0145	-
0.1868	700	0.0103	-
0.2002	750	0.0083	-
0.2135	800	0.006	-
0.2268	850	0.0049	-
0.2402	900	0.0021	-
0.2535	950	0.0037	-
0.2669	1000	0.0019	-
0.2802	1050	0.0015	-
0.2936	1100	0.0029	-
0.3069	1150	0.0013	-
0.3203	1200	0.0013	-
0.3336	1250	0.0011	-
0.3469	1300	0.0011	-
0.3603	1350	0.001	-
0.3736	1400	0.0017	-
0.3870	1450	0.0013	-
0.4003	1500	0.0023	-
0.4137	1550	0.0009	-
0.4270	1600	0.0009	-
0.4404	1650	0.0008	-
0.4537	1700	0.0008	-
0.4670	1750	0.0008	-
0.4804	1800	0.0007	-
0.4937	1850	0.0007	-
0.5071	1900	0.0007	-
0.5204	1950	0.0007	-
0.5338	2000	0.0007	-
0.5471	2050	0.0007	-
0.5604	2100	0.0007	-
0.5738	2150	0.0007	-
0.5871	2200	0.0007	-
0.6005	2250	0.0006	-
0.6138	2300	0.0007	-
0.6272	2350	0.0007	-
0.6405	2400	0.0006	-
0.6539	2450	0.0006	-
0.6672	2500	0.0013	-
0.6805	2550	0.0006	-
0.6939	2600	0.0006	-
0.7072	2650	0.0006	-
0.7206	2700	0.0006	-
0.7339	2750	0.0006	-
0.7473	2800	0.0006	-
0.7606	2850	0.0006	-
0.7740	2900	0.0005	-
0.7873	2950	0.0006	-
0.8006	3000	0.0005	-
0.8140	3050	0.0005	-
0.8273	3100	0.0005	-
0.8407	3150	0.0005	-
0.8540	3200	0.0005	-
0.8674	3250	0.0005	-
0.8807	3300	0.0005	-
0.8940	3350	0.0005	-
0.9074	3400	0.0005	-
0.9207	3450	0.0005	-
0.9341	3500	0.0005	-
0.9474	3550	0.0005	-
0.9608	3600	0.0005	-
0.9741	3650	0.0006	-
0.9875	3700	0.0005	-

Framework Versions

• Python: 3.11.13
• SetFit: 1.1.3
• Sentence Transformers: 5.1.0
• Transformers: 4.55.1
• PyTorch: 2.6.0+cu124
• Datasets: 4.0.0
• Tokenizers: 0.21.4

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}
}