Finetuned DeBERTa-v3-large: Scancode Required Phrase Extraction - for GSoC'26 proposal
This is a fine-tuned microsoft/deberta-v3-large model trained specifically to extract required software license phrases (using BIO tagging) for the Scancode Toolkit GSoC 2026 project
- F1 Score: 0.7931
- Architecture: DeBERTa-v3-large + Layer-wise Learning Rate Decay (LLRD)
- Task: Token Classification (O, B-REQ, I-REQ)
- Downloads last month
- 2