frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-11-11 16:50:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix numel() downcast in fused_layernorm_dynamic_per_token_quant.cu (#17316)

Browse Source
This commit is contained in:
Richard Barnes
2025-04-28 19:23:18 -07:00
committed by GitHub
parent b4ac4fa04d
commit d6da8a8ff2
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/quantization/fused_kernels/fused_layernorm_dynamic_per_token_quant.cu
View File

@ -96,7 +96,7 @@ void rms_norm_dynamic_per_token_quant_dispatch(
std::optional<at::Tensor> const& scale_ub,
std::optional<at::Tensor>& residual) {
int32_t hidden_size = input.size(-1);
int32_t num_tokens = input.numel() / hidden_size;
auto num_tokens = input.numel() / hidden_size;
dim3 grid(num_tokens);
dim3 block(std::min(hidden_size, 1024));