侯奇 comments

Results 45 comments of


                                            侯奇

[QUESTION]flux/src/moe_ag_scatter/ths_op/gemm_grouped_v3_ag_scatter.cc里的写stream很耗时，有优化方法吗？CU_CHECK(CUStreamWriteValue(this->cp_stream_intra_node,(CUdeviceptr)(ptr_offset(barrier_block.get(), src_rank * sizeof(int))),1,CU_STREAM_WRITE_VALUE_DEFAULT));

It can be solved, but not a very easy one. have to * put the shape fully into device side * run no CUDA runtime API and use kernels for...

[BUG] pip install error:urllib.error.HTTPError: HTTP Error 404: Not Found

sees there is a bug, i will release the latest later.

[BUG]Failure to compile from `cuda::atomic_ref`

thanks for your comment. FLUX is not tested with CUDA 12.9. we will fix it later.

[BUG] Compile Error

strange that you run build with --arch 89 but got error with complain about `bytedance::flux::GemmV3Meta::check_type() const`. GemmV3Meta is for hopper only. did you build the repo once with --arch 90,...

triton 3.0.0 got FP8 GEMM precision issue with CUDA sm89

maybe it's because ada fp8 implementation has no non-fast accumulation.