w11wo's picture
Update README.md
b39e385
metadata
library_name: setfit
tags:
  - setfit
  - sentence-transformers
  - text-classification
  - generated_from_setfit_trainer
metrics:
  - accuracy
  - f1
  - precision
  - recall
widget:
  - text: so i am currently stuck in an automatic revolving door .
  - text: ah my favorite pastime , watching logan and crying
  - text: i have a new instagram account ! go give theollyjackson a follow
  - text: guess they are not rich enough to get their precious cars in a garage .
  - text: last day in my twenties
pipeline_tag: text-classification
inference: true
base_model: BAAI/bge-small-en-v1.5
model-index:
  - name: SetFit with BAAI/bge-small-en-v1.5
    results:
      - task:
          type: text-classification
          name: Text Classification
        dataset:
          name: iSarcasm
          type: isarcasm
          split: test
        metrics:
          - type: accuracy
            value: 0.6617812852311161
            name: Accuracy
          - type: f1
            value: 0.3951612903225807
            name: F1
          - type: precision
            value: 0.2890855457227139
            name: Precision
          - type: recall
            value: 0.6242038216560509
            name: Recall

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 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
  • Sentence Transformer body: BAAI/bge-small-en-v1.5
  • Classification head: a SetFitHead instance
  • Maximum Sequence Length: 512 tokens
  • Number of Classes: 2 classes

Model Sources

Model Labels

Label Examples
NON_SARCASTIC
  • 'so the newer devices have the ios screenshot i m still on ios but my ipad mini 1 st gen shows the ios screenshot . odd .'
  • 'why do amazon need a test authorisation when i add a new payment card , as well as the authorisation they get when i actually use it ?'
  • 'waterboarding sounds like a lot of fun until you find out what it is'
SARCASTIC
  • "have you been reading long ? you are not very good at it . it has nothing to do with who i like , especially since i am not a fan of corbyn anyway . it ' s that in one case someone was literally slapped in the face , and in the other someone wore a milkshake . battery > being annoying"
  • 'wish one of the many people dressed as killers were actually one n killed me'
  • 'is it even christmas if there isn t a fight with neighbours and a broken wrist ?'

Evaluation

Metrics

Label Accuracy F1 Precision Recall
all 0.6618 0.3952 0.2891 0.6242

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("w11wo/bge-small-en-v1.5-isarcasm")
# Run inference
preds = model("last day in my twenties")

Training Details

Training Set Metrics

Training set Min Median Max
Word count 2 19.8489 63
Label Training Sample Count
NON_SARCASTIC 609
SARCASTIC 609

Training Hyperparameters

  • batch_size: (256, 16)
  • num_epochs: (3, 8)
  • max_steps: -1
  • sampling_strategy: oversampling
  • body_learning_rate: (2e-05, 5e-06)
  • head_learning_rate: 0.002
  • loss: CosineSimilarityLoss
  • distance_metric: cosine_distance
  • margin: 0.25
  • end_to_end: True
  • use_amp: False
  • warmup_proportion: 0.1
  • l2_weight: 0.01
  • seed: 42
  • eval_max_steps: -1
  • load_best_model_at_end: True

Training Results

Epoch Step Training Loss Validation Loss
0.0003 1 0.2571 -
0.0172 50 0.251 -
0.0344 100 0.2556 -
0.0517 150 0.2513 -
0.0689 200 0.2531 -
0.0861 250 0.2518 -
0.1033 300 0.2553 -
0.1206 350 0.2501 -
0.1378 400 0.2546 -
0.1550 450 0.2506 -
0.1722 500 0.2317 -
0.1895 550 0.093 -
0.2067 600 0.0139 -
0.2239 650 0.0166 -
0.2411 700 0.0053 -
0.2584 750 0.0013 -
0.2756 800 0.0121 -
0.2928 850 0.0096 -
0.3100 900 0.0043 -
0.3272 950 0.0014 -
0.3445 1000 0.0009 -
0.3617 1050 0.0117 -
0.3789 1100 0.0144 -
0.3961 1150 0.0084 -
0.4134 1200 0.0006 -
0.4306 1250 0.0005 -
0.4478 1300 0.0081 -
0.4650 1350 0.0144 -
0.4823 1400 0.0045 -
0.4995 1450 0.0042 -
0.5167 1500 0.0005 -
0.5339 1550 0.003 -
0.5512 1600 0.0004 -
0.5684 1650 0.0005 -
0.5856 1700 0.0004 -
0.6028 1750 0.0004 -
0.6200 1800 0.0026 -
0.6373 1850 0.0004 -
0.6545 1900 0.0004 -
0.6717 1950 0.0003 -
0.6889 2000 0.0014 -
0.7062 2050 0.0004 -
0.7234 2100 0.0003 -
0.7406 2150 0.0003 -
0.7578 2200 0.0004 -
0.7751 2250 0.0003 -
0.7923 2300 0.0003 -
0.8095 2350 0.0003 -
0.8267 2400 0.0003 -
0.8440 2450 0.0003 -
0.8612 2500 0.0003 -
0.8784 2550 0.0003 -
0.8956 2600 0.0003 -
0.9128 2650 0.0003 -
0.9301 2700 0.0003 -
0.9473 2750 0.0004 -
0.9645 2800 0.0003 -
0.9817 2850 0.0003 -
0.9990 2900 0.0036 -
1.0162 2950 0.0003 -
1.0334 3000 0.0003 -
1.0506 3050 0.0002 -
1.0679 3100 0.0002 -
1.0851 3150 0.0002 -
1.1023 3200 0.0002 -
1.1195 3250 0.0002 -
1.1368 3300 0.0003 -
1.1540 3350 0.0004 -
1.1712 3400 0.0002 -
1.1884 3450 0.0002 -
1.2056 3500 0.0002 -
1.2229 3550 0.0002 -
1.2401 3600 0.0002 -
1.2573 3650 0.0009 -
1.2745 3700 0.0002 -
1.2918 3750 0.0002 -
1.3090 3800 0.0002 -
1.3262 3850 0.0002 -
1.3434 3900 0.0002 -
1.3607 3950 0.0002 -
1.3779 4000 0.0002 -
1.3951 4050 0.0002 -
1.4123 4100 0.0002 -
1.4296 4150 0.0002 -
1.4468 4200 0.0003 -
1.4640 4250 0.0002 -
1.4812 4300 0.0002 -
1.4984 4350 0.0002 -
1.5157 4400 0.0002 -
1.5329 4450 0.0002 -
1.5501 4500 0.0002 -
1.5673 4550 0.0002 -
1.5846 4600 0.0002 -
1.6018 4650 0.0002 -
1.6190 4700 0.0002 -
1.6362 4750 0.0002 -
1.6535 4800 0.0002 -
1.6707 4850 0.0002 -
1.6879 4900 0.0002 -
1.7051 4950 0.0002 -
1.7224 5000 0.0003 -
1.7396 5050 0.0002 -
1.7568 5100 0.0002 -
1.7740 5150 0.0002 -
1.7913 5200 0.0002 -
1.8085 5250 0.0002 -
1.8257 5300 0.0038 -
1.8429 5350 0.0002 -
1.8601 5400 0.0002 -
1.8774 5450 0.0002 -
1.8946 5500 0.0002 -
1.9118 5550 0.0002 -
1.9290 5600 0.0005 -
1.9463 5650 0.0002 -
1.9635 5700 0.0002 -
1.9807 5750 0.0002 -
1.9979 5800 0.0002 -
2.0152 5850 0.0001 -
2.0324 5900 0.0002 -
2.0496 5950 0.0002 -
2.0668 6000 0.0002 -
2.0841 6050 0.0002 -
2.1013 6100 0.0002 -
2.1185 6150 0.0002 -
2.1357 6200 0.0001 -
2.1529 6250 0.0002 -
2.1702 6300 0.0002 -
2.1874 6350 0.0001 -
2.2046 6400 0.0001 -
2.2218 6450 0.0001 -
2.2391 6500 0.0001 -
2.2563 6550 0.0001 -
2.2735 6600 0.0001 -
2.2907 6650 0.0001 -
2.3080 6700 0.0001 -
2.3252 6750 0.0001 -
2.3424 6800 0.0001 -
2.3596 6850 0.0001 -
2.3769 6900 0.0001 -
2.3941 6950 0.0001 -
2.4113 7000 0.0001 -
2.4285 7050 0.0001 -
2.4457 7100 0.0001 -
2.4630 7150 0.0001 -
2.4802 7200 0.0001 -
2.4974 7250 0.0001 -
2.5146 7300 0.0001 -
2.5319 7350 0.0001 -
2.5491 7400 0.0001 -
2.5663 7450 0.0001 -
2.5835 7500 0.0001 -
2.6008 7550 0.0001 -
2.6180 7600 0.0001 -
2.6352 7650 0.0001 -
2.6524 7700 0.0001 -
2.6697 7750 0.0001 -
2.6869 7800 0.0001 -
2.7041 7850 0.0001 -
2.7213 7900 0.0001 -
2.7385 7950 0.0001 -
2.7558 8000 0.0001 -
2.7730 8050 0.0001 -
2.7902 8100 0.0001 -
2.8074 8150 0.0001 -
2.8247 8200 0.0001 -
2.8419 8250 0.0001 -
2.8591 8300 0.0001 -
2.8763 8350 0.0001 -
2.8936 8400 0.0001 -
2.9108 8450 0.0001 -
2.9280 8500 0.0001 -
2.9452 8550 0.0001 -
2.9625 8600 0.0001 -
2.9797 8650 0.0001 -
2.9969 8700 0.0001 -

Framework Versions

  • Python: 3.10.12
  • SetFit: 1.0.1
  • Sentence Transformers: 2.2.2
  • Transformers: 4.32.0
  • PyTorch: 2.1.1+cu121
  • Datasets: 2.14.5
  • Tokenizers: 0.13.3

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}
}