biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[small][batch invariance] Rename the env and internal flags to simplify usage (#26855)

Browse Source 
Signed-off-by: Bram Wasti <bwasti@meta.com>
This commit is contained in:
Bram Wasti
2025-10-16 14:40:25 -07:00
committed by GitHub
parent 23583ee28c
commit b2f78cbad4
20 changed files with 61 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/layernorm_quant_kernels.cu
View File

@@ -254,7 +254,7 @@ void fused_add_rms_norm_static_fp8_quant(
auto wt_ptr = reinterpret_cast<std::uintptr_t>(weight.data_ptr());
bool ptrs_are_aligned =
inp_ptr % 16 == 0 && res_ptr % 16 == 0 && wt_ptr % 16 == 0;
bool batch_invariant_launch = vllm::vllm_kernel_override_batch_invariant();
bool batch_invariant_launch = vllm::vllm_is_batch_invariant();
if (ptrs_are_aligned && hidden_size % 8 == 0 && input_stride % 8 == 0 &&
!batch_invariant_launch) {
LAUNCH_FUSED_ADD_RMS_NORM(8);