PhD student, University of the Chinese Academy of Sciences
1 paper at NeurIPS 2025
We present IR-OptSet, a public LLVM IR dataset tailored for optimization-sensitive LLM training, significantly improving compiler code generation performance.