obscura/vllm
2
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2026-06-06 00:16:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

Use hidden_pad and intermediate_pad from vLLM #34301 (#42098)

Browse Source 
Signed-off-by: Rebecca Lee <Rebecca.Lee@amd.com>
This commit is contained in:
Rebecca Lee
2026-05-13 23:21:04 -07:00
committed by GitHub
parent b26558d4a3
commit fd7d858c8a
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/model_executor/layers/fused_moe/experts/rocm_aiter_moe.py
+2 -2
View File
@@ -356,8 +356,8 @@ def rocm_aiter_fused_experts(
doweight_stage1=apply_router_weight_on_input,
num_local_tokens=num_local_tokens,
output_dtype=output_dtype,
hidden_pad=hidden_pad,
intermediate_pad=intermediate_pad,
hidden_pad=hidden_pad // 128 * 128,
intermediate_pad=intermediate_pad // 64 * 64 * 2,
bias1=quant_config.w1_bias if quant_config.use_mxfp4_w4a16 else None,
bias2=quant_config.w2_bias if quant_config.use_mxfp4_w4a16 else None,
)