[Bugfix] Fix int32 token offset overflow in fused SiLU block quant (#53409)
Signed-off-by: Anni <dranox2222@gmail.com> Co-authored-by: OpenAI Codex <noreply@openai.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Co-authored-by: Yongye Zhu <zyy1102000@gmail.com>
This commit is contained in:
co-authored by
OpenAI Codex
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Yongye Zhu
parent
67e86d1e6f
commit
06cccf8730
@@ -26,7 +26,7 @@ __global__ void silu_and_mul_per_block_quant_kernel(
|
||||
"group_size must be a power of 2 for correct reduction");
|
||||
|
||||
// Grid: (num_tokens, num_groups)
|
||||
int const token_idx = blockIdx.x;
|
||||
int64_t const token_idx = blockIdx.x;
|
||||
int const group_idx = blockIdx.y;
|
||||
int const tid = threadIdx.x; // tid in [0, group_size)
|
||||
int const num_tokens = gridDim.x;
|
||||
|
||||
Reference in New Issue
Block a user