Jiajun Chen

A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$Δ$ Integration into Upcycled MoE

The First Impression Problem: Internal Bias Triggers Overthinking in Reasoning Models