EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Fix bug of deepseek-v3 under DP+EP mode with large batchsize/seqlen (#6449)

Browse Source
This commit is contained in:
likesen-alibaba
2025-07-10 16:19:56 +08:00
committed by GitHub
parent 5748241549
commit 4a0d19198b
2 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
sgl-kernel/csrc/gemm/per_token_group_quant_8bit.cu
View File

@@ -35,12 +35,12 @@ __global__ void per_token_group_quant_8bit_kernel(
const int scale_num_rows = 0,
const int scale_stride = 0) {
const int threads_per_group = 16;
const int local_group_id = threadIdx.x / threads_per_group;
const int64_t local_group_id = threadIdx.x / threads_per_group;
const int lane_id = threadIdx.x % threads_per_group;
const int block_group_id = blockIdx.x * groups_per_block;
const int global_group_id = block_group_id + local_group_id;
const int block_group_offset = global_group_id * group_size;
const int64_t block_group_id = blockIdx.x * groups_per_block;
const int64_t global_group_id = block_group_id + local_group_id;
const int64_t block_group_offset = global_group_id * group_size;
float local_absmax = eps;