| Property | Value |
|---|---|
| Developer | deepset |
| Training Data | GermanQuAD (~12MB) |
| Language | German |
| Release Date | April 21st, 2021 |
| Framework Support | Haystack, Transformers |
What is gelectra-base-germanquad?
gelectra-base-germanquad is a specialized German language model designed for extractive question answering tasks. Built on the ELECTRA architecture and trained on the GermanQuAD dataset, it represents a significant advancement in German language understanding and information extraction capabilities.
Implementation Details
The model was trained using specific hyperparameters including a batch size of 24, 2 epochs, maximum sequence length of 384, and a learning rate of 3e-5. It utilizes linear warmup scheduling and implements embeddings dropout probability of 0.1. The training was conducted on a V100 GPU infrastructure.
- Trained on 11,518 questions and answers in the training set
- Evaluated on a three-way annotated test set with 2,204 questions
- Warm-started on German translation of SQuAD v1.1
- Fine-tuned on GermanQuAD dataset
Core Capabilities
- Extractive question answering in German
- Integration with both Haystack and Transformers frameworks
- Optimized for production deployment
- Capable of processing complex German text queries
Frequently Asked Questions
Q: What makes this model unique?
This model is specifically optimized for German language question answering, utilizing the ELECTRA architecture and trained on the hand-annotated GermanQuAD dataset. It represents one of the few specialized German QA models available.
Q: What are the recommended use cases?
The model is ideal for extractive question answering tasks in German text, particularly in applications requiring precise information extraction from documents. It's well-suited for integration into production systems through frameworks like Haystack or Transformers.