tclin914 wrote: > Can we have some evaluations about impacts on code size and performance when using large code model? Our internal evaluation shows that the impact on both code size and performance is about 2.2% compared to the medany. https://github.com/llvm/llvm-project/pull/70308