Convbert Base Turkish Mc4 Cased Allnli Tr

🚀 convbert-base-turkish-mc4-cased_allnli_tr

This model is a fine - tuned version of dbmdz/convbert-base-turkish-mc4-cased on the None dataset. It is designed for zero - shot classification tasks and can be used in natural language inference scenarios.

🚀 Quick Start

This model is ready for zero - shot classification tasks. You can use it directly with the appropriate framework and input data.

✨ Features

• Zero - shot Classification: Capable of classifying text into predefined categories without explicit training on those categories.
• NLI Support: Supports natural language inference tasks, which can help in understanding the logical relationships between texts.

📚 Documentation

Model Performance

It achieves the following results on the evaluation set:

• Loss: 0.5541
• Accuracy: 0.8111

Training and Evaluation Data

The model uses the nli_tr dataset for training and evaluation.

Metrics

The accuracy metric is used to evaluate the model's performance.

Widget Examples

• Example 1:
  • Text: "Dolar yükselmeye devam ediyor."
  • Candidate Labels: "ekonomi, siyaset, spor"
• Example 2:
  • Text: "Senaryo çok saçmaydı, beğendim diyemem."
  • Candidate Labels: "olumlu, olumsuz"

🔧 Technical Details

Training Hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e - 05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
• lr_scheduler_type: linear
• num_epochs: 3

Training Results

Framework versions

• Transformers 4.12.3
• Pytorch 1.10.0+cu102
• Datasets 1.15.1
• Tokenizers 0.10.3

📄 License

This model is licensed under the Apache - 2.0 license.