xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-29 01:07:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

Minor

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon 2024-12-21 17:28:21 -08:00
parent 8a4180c8b6
commit 03b1e6fdbd
1 changed files with 10 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
csrc/prepare_inputs/copy_subranges.cu
View File

@ -55,7 +55,16 @@ void copy_subranges(torch::Tensor& matrix_src, torch::Tensor& matrix_diff,
// One thread block per row. // One thread block per row.
int blocks = n; int blocks = n;
int threads = 1024; int threads;
if (blocks < 128) {
threads = 1024;
} else if (blocks < 256) {
threads = 512;
} else if (blocks < 512) {
threads = 256;
} else {
threads = 128;
}
const at::cuda::OptionalCUDAGuard device_guard(device_of(matrix_tgt)); const at::cuda::OptionalCUDAGuard device_guard(device_of(matrix_tgt));
const cudaStream_t stream = at::cuda::getCurrentCUDAStream(); const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
vllm::copy_subranges_kernel<<<blocks, threads, 0, stream>>>( vllm::copy_subranges_kernel<<<blocks, threads, 0, stream>>>(