Multilingual fine-tuning via localized gradient conflict resolution

Arxiv CS.AI (top papers)

Proposes Localized Gradient Conflict Resolution to mitigate negative interference across languages during LLM fine-tuning. Aims to improve cross-lingual performance without extra data.

multilingualfine-tuning

Arxiv CS.AI (top papers)·Long P. HoangYiran ZhaoWei LuWenxuan Zhang

6 days ago