ruRoberta-large-rucola
This model is a fine-tuned version of ai-forever/ruRoberta-large on the RuCoLa (Russian Corpus of Linguistic Acceptability) dataset. It predicts whether a given Russian sentence is linguistically correct or contains errors.
Key Features
- Task: Binary classification (acceptable vs. unacceptable)
- Training data: RuCoLa (~10k labeled sentences)
- Max sequence length: 512 tokens
- Fine-tuning framework: PyTorch + Hugging Face transformers
Hyperparameters
| Parameter | Value |
|---|---|
| Batch size | 32 |
| Learning rate | 1e-5 |
| Epochs | 64 |
| Warmup steps | 100 |
| Optimizer | adamw_bnb_8bit |
- Downloads last month
- 2