Tag
This paper proposes ST-Merge, a steerable model merging framework that uses a gated cross-attention mechanism to adaptively modulate contributions of a multilingual model and a reasoning model, outperforming fixed merging approaches on multilingual reasoning benchmarks across 21 languages.
This paper revisits the multilingual reasoning gap in LLMs, finding it smaller than previously reported under comparable supervision. It introduces Layer Swap, which transfers mid-layer weights from an English reasoning specialist to native language specialists, nearly closing the gap while preserving native-language chain-of-thought.
Researchers introduce x1, a family of reasoning models that adaptively select optimal languages for reasoning on a per-instance basis, demonstrating that language choice impacts reasoning quality in multilingual and cultural tasks.
This paper investigates multilingual latent reasoning in large reasoning models across 11 languages, revealing that while latent reasoning capabilities exist, they are unevenly distributed—stronger in resource-rich languages and weaker in low-resource ones. The study finds that despite surface-level differences, the internal reasoning mechanisms are largely aligned with an English-centered pathway.