xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 02:35:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Make moe_align_block_size AMD-compatible (#3470)

Browse Source
This commit is contained in:
Woosuk Kwon 2024-03-18 11:26:24 -07:00 committed by GitHub
parent 93348d9458
commit 9101d832e6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
csrc/moe_align_block_size_kernels.cu
View File

@ -111,7 +111,8 @@ void moe_align_block_size(
// set dynamic shared mem // set dynamic shared mem
auto kernel = vllm::moe_align_block_size_kernel<scalar_t>; auto kernel = vllm::moe_align_block_size_kernel<scalar_t>;
AT_CUDA_CHECK(cudaFuncSetAttribute(kernel, cudaFuncAttributeMaxDynamicSharedMemorySize, shared_mem)); AT_CUDA_CHECK(
VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize((void *)kernel, shared_mem));
kernel<<<1, num_experts, shared_mem, stream>>>( kernel<<<1, num_experts, shared_mem, stream>>>(
topk_ids.data_ptr<scalar_t>(), topk_ids.data_ptr<scalar_t>(),
sorted_token_ids.data_ptr<int32_t>(), sorted_token_ids.data_ptr<int32_t>(),