Mending the Holes: Mitigating Reward Hacking in Reinforcement Learning for Multilingual Translation
-
lyf07/LLaMAX3-8B-Alpaca-WALAR
Translation • 8B • Updated • 14 -
lyf07/Qwen3-8B-WALAR
Translation • 8B • Updated • 16 -
lyf07/Translategemma-4B-it-WALAR
Translation • 769k • Updated • 26 • 4 -
Mending the Holes: Mitigating Reward Hacking in Reinforcement Learning for Multilingual Translation
Paper • 2603.13045 • Published • 2