Think in English, Answer in Korean: Efficient Adaptation of Multilingual Tool-Using Agents

2026-06-30

AI Summary

This work addresses the challenge of efficiently adapting multilingual large language models under memory and deployment constraints. The authors propose a lightweight, pretraining-free approach that extends the Cohere Command R model with a prefix-based conditional control mechanism, enabling dynamic switching between concise responses and tool-oriented reasoning. By integrating multilingual supervised fine-tuning, reinforcement learning with language-consistency rewards and verifiable multi-step tool usage, and 4-bit quantization, the method significantly enhances performance on mathematical reasoning, function calling, and natural language to SQL tasks. The resulting model maintains strong bilingual instruction-following capabilities in both Korean and English while supporting efficient deployment on a single GPU.

Abstract

We present LuckyStar 111B, a 111B-parameter hybrid reasoning model developed through a collaboration between Cohere and LG CNS for Korean-English enterprise agents under practical memory and serving constraints. The model trains from Cohere's fully post-trained Command A model rather than a new pretraining run, and uses preamble conditioning to switch between concise non-reasoning behavior and longer tool-oriented reasoning. We study four choices for scaling tool-using agents efficiently: multilingual supervised fine-tuning, reinforcement learning with verifiable rewards for multi-step tool-use tasks, language-consistency rewards for Korean user-facing responses, and 4-bit quantization for single-GPU serving. The adapted model improves mathematical reasoning, function calling, and agentic natural-language-to-SQL (NL2SQL) performance while preserving general Korean and English instruction-following quality. These results provide a practical recipe and failure-mode analysis for adapting post-trained multilingual models to verifiable agentic workflows under memory-constrained deployment.

Problem

• multilingual tool-using agents
• memory-constrained deployment
• efficient adaptation
• post-trained models
• agentic workflows

Innovation

• tool-using agents
• multilingual adaptation
• preamble conditioning
• verifiable reinforcement learning
• 4-bit quantization