Explore TritonRL, a novel framework leveraging hybrid training and hierarchical rewards to enable LLMs to generate correct, high-performance Triton kernels w...
Level: advanced
By Jiin Woo and 5 other authors
Category: research