The Unreasonable Effectiveness of Model Merging for Cross-Lingual Transfer in LLMs

Open in new window