🚀 Try our demo on Hugging Face Spaces:
👉 DeDisco Demo
Description Instruction-tuned decoder-only model for multilingual discourse relation classification (DISRPT 2025). Predicts one of 17 discourse relation labels based on structured prompts.
Details
- Based on: Qwen3-4B
- Pruned: 1 layer removed
- Parameters: ~3.92B
- Fine-tuned on DISRPT 2025 shared task data.
- License: Apache 2.0
For more details, please refer to GitHub
- Downloads last month
- 8