🐯 Liger GRPO meets TRL

Liger GRPO has integrated with TRL (Task-Relevant Learning), enhancing its performance on various benchmarks. This integration allows for more efficient training by optimizing task-specific representations within the Liger architecture, which is designed for large-scale language processing. This development is significant for practitioners as it improves the adaptability and efficiency of LLMs in specialized applications, potentially reducing the computational resources needed for fine-tuning on specific tasks.