Beyond Initialization Loss: A Systematic Study of Token Embedding Initialization Strategies for LLM Vocabulary Extension 文章

ArXiv CS.CL2026-08-05PAPERen作者: Raviraj Joshi, Utkarsh Vaidya, Sanjay Singh Chauhan, Niranjan Wartikar

Beyond Initialization Loss: A Systematic Study of Token Embedding Initialization Strategies for LLM Vocabulary Extension · 相关事件

暂无数据