metadata
base_model: akhooli/sbert_ar_nli_500k_norm
library_name: setfit
metrics:
- accuracy
pipeline_tag: text-classification
tags:
- setfit
- sentence-transformers
- text-classification
- generated_from_setfit_trainer
widget:
- text: >-
جيد. كل شي جيد فندق عائلي به مواقف مجانيه وامام الفندق العاب للاطفال.
الانترنت سيئ
- text: مقبول. . الاثاث قديم جدا
- text: المرأة الخارقة
- text: >-
الاسئلة اللي بتطرحا الرواية مقبلة لكن الصيغة غير مقبولة لأننا حتى لما
نتسائل عن الله و حكمه في خلق أو موت الانسان يجب أن نسأل بإحترام أكثر من
ذلك . تعبيرات الكاتب في وصف الصحراء و الحياة بها جميل و مفصل. أعجبني
التنوع في التسلسل الزمني و كأني بقرأ أكثر من قصة لكن يربطها رابط يمنع تفكك
تفاصيل الرواية. الغرض من الرواية جيد لكن بضع جزئيات كانت مبتزلة بسبب إسلوب
الطرح الوقح من رأيي
- text: >-
ضعيف. المول. المويه لونها اصفر المويه الحارة تنقطع الغرفة وسخة الانترنت
بفلوس وسعره غالي منيب مكرر زيارتي للفندق مرة ثانية
inference: true
model-index:
- name: SetFit with akhooli/sbert_ar_nli_500k_norm
results:
- task:
type: text-classification
name: Text Classification
dataset:
name: Unknown
type: unknown
split: test
metrics:
- type: accuracy
value: 0.6874
name: Accuracy
SetFit with akhooli/sbert_ar_nli_500k_norm
This is a SetFit model that can be used for Text Classification. This SetFit model uses akhooli/sbert_ar_nli_500k_norm as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification. It was trained on akhooli/ar_reviews_100k_3 dataset (4500 samples, as few shot) with 68.7% accuracy. There are 3 labels in the dataset: 0: negative, 1:positive, 2:mixed/neutral. Normalize the text before classifying as the model uses normalized text. Here's how to use the model:
pip install setfit
from setfit import SetFitModel
from unicodedata import normalize
# Download model from Hub
model = SetFitModel.from_pretrained("akhooli/setfit_ar_100k_reviews")
# Run inference
queries = [
"يغلي الماء عند 100 درجة مئوية",
"فعلا لقد أحببت ذلك الفيلم",
"🤮 اﻷناناس مع البيتزا؟ إنه غير محبذ",
"رأيت أناسا بائسين في الطريق",
"لم يعجبني المطعم رغم أن السعر مقبول",
"من باب جبر الخاطر هذه 3 نجوم لتقييم الخدمة",
"من باب جبر الخواطر، هذه نجمة واحدة لخدمة ﻻ تستحق"
]
queries_n = [normalize('NFKC', query) for query in queries]
preds = model.predict(queries_n)
print(preds)
# if you want to see the probabilities for each label
probas = model.predict_proba(queries_n)
print(probas)
The rest of this model card is auto generated.
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: akhooli/sbert_ar_nli_500k_norm
- 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 |
---|---|
Negative |
|
Mixed |
|
Positive |
|
Evaluation
Metrics
Label | Accuracy |
---|---|
all | 0.6874 |
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("akhooli/setfit_ar_100k_reviews")
# Run inference
preds = model("المرأة الخارقة")
Training Details
Training Set Metrics
Training set | Min | Median | Max |
---|---|---|---|
Word count | 1 | 53.0251 | 1598 |
Label | Training Sample Count |
---|---|
Negative | 1500 |
Positive | 1500 |
Mixed | 1500 |
Training Hyperparameters
- batch_size: (16, 16)
- num_epochs: (1, 1)
- max_steps: 5000
- sampling_strategy: undersampling
- 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
- run_name: setfit_reviews_7.5k
- eval_max_steps: -1
- load_best_model_at_end: False
Training Results
Epoch | Step | Training Loss | Validation Loss |
---|---|---|---|
0.0004 | 1 | 0.3397 | - |
0.04 | 100 | 0.2846 | - |
0.08 | 200 | 0.2523 | - |
0.12 | 300 | 0.2248 | - |
0.16 | 400 | 0.2089 | - |
0.2 | 500 | 0.1947 | - |
0.24 | 600 | 0.182 | - |
0.28 | 700 | 0.1614 | - |
0.32 | 800 | 0.1493 | - |
0.36 | 900 | 0.139 | - |
0.4 | 1000 | 0.1128 | - |
0.44 | 1100 | 0.1056 | - |
0.48 | 1200 | 0.0896 | - |
0.52 | 1300 | 0.0748 | - |
0.56 | 1400 | 0.0616 | - |
0.6 | 1500 | 0.0585 | - |
0.64 | 1600 | 0.048 | - |
0.68 | 1700 | 0.0422 | - |
0.72 | 1800 | 0.0371 | - |
0.76 | 1900 | 0.0306 | - |
0.8 | 2000 | 0.028 | - |
0.84 | 2100 | 0.0236 | - |
0.88 | 2200 | 0.0211 | - |
0.92 | 2300 | 0.0173 | - |
0.96 | 2400 | 0.0175 | - |
1.0 | 2500 | 0.0158 | - |
1.04 | 2600 | 0.0153 | - |
1.08 | 2700 | 0.0195 | - |
1.12 | 2800 | 0.0141 | - |
1.16 | 2900 | 0.0113 | - |
1.2 | 3000 | 0.0084 | - |
1.24 | 3100 | 0.0073 | - |
1.28 | 3200 | 0.0073 | - |
1.32 | 3300 | 0.007 | - |
1.3600 | 3400 | 0.0075 | - |
1.4 | 3500 | 0.0068 | - |
1.44 | 3600 | 0.0038 | - |
1.48 | 3700 | 0.0028 | - |
1.52 | 3800 | 0.0031 | - |
1.56 | 3900 | 0.0056 | - |
1.6 | 4000 | 0.0059 | - |
1.6400 | 4100 | 0.0022 | - |
1.6800 | 4200 | 0.0052 | - |
1.72 | 4300 | 0.004 | - |
1.76 | 4400 | 0.004 | - |
1.8 | 4500 | 0.0047 | - |
1.8400 | 4600 | 0.0027 | - |
1.88 | 4700 | 0.0036 | - |
1.92 | 4800 | 0.0039 | - |
1.96 | 4900 | 0.004 | - |
2.0 | 5000 | 0.0048 | - |
Framework Versions
- Python: 3.10.14
- SetFit: 1.2.0.dev0
- Sentence Transformers: 3.1.1
- Transformers: 4.45.1
- PyTorch: 2.4.0
- Datasets: 3.0.1
- Tokenizers: 0.20.0
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}
}