deberta-v3-large_boolq
| Property | Value |
|---|---|
| Parameter Count | 435M |
| License | MIT |
| Base Model | microsoft/deberta-v3-large |
| Accuracy | 88.35% |
| Training Batch Size | 32 |
What is deberta-v3-large_boolq?
deberta-v3-large_boolq is a fine-tuned version of Microsoft's DeBERTa-v3-Large model specifically optimized for boolean question answering tasks. The model demonstrates impressive performance with an accuracy of 88.35% on the BoolQ dataset validation split, making it particularly effective for yes/no question answering scenarios.
Implementation Details
The model utilizes the DeBERTa-v3 architecture with 435M parameters and implements sequence classification for boolean questions. It was trained using the Adam optimizer with a learning rate of 1e-05 over 5 epochs, employing a linear learning rate scheduler and gradient accumulation steps of 2.
- Training performed with PyTorch 2.0.1
- Implements Transformers 4.32.1
- Uses F32 tensor type for computations
- Achieves 0.4601 validation loss
Core Capabilities
- Boolean question answering with high accuracy
- Processes question-context pairs efficiently
- Returns probability distributions for yes/no answers
- Handles various text lengths through proper tokenization
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in boolean question answering with state-of-the-art accuracy of 88.35%, leveraging the powerful DeBERTa-v3 architecture while maintaining efficient inference capabilities.
Q: What are the recommended use cases?
The model is ideal for applications requiring yes/no answers based on provided context, such as fact verification systems, automated Q&A platforms, and document analysis tools where binary decisions are needed.
