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

[ROCm] Fix MoE kernel test failures on gfx950 (#37833)

Browse Source 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com>
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com>
This commit is contained in:
Andreas Karatzas
2026-03-25 13:46:40 -05:00
committed by GitHub
parent e38817fadb
commit 7d6917bef5
12 changed files with 478 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/kernels/utils.py
View File

@@ -941,7 +941,7 @@ def torch_experts(
if b_bias1 is not None:
tmp1 = tmp1 + b_bias1[i].view(1, -1).to(out.dtype)
tmp2 = SiluAndMul()(tmp1).to(out.dtype)
tmp2 = act()(tmp1).to(out.dtype)
tmp2, b_scale = moe_kernel_quantize_input(
tmp2, a2_scale, quant_dtype, per_act_token_quant, block_shape