[feature] add_rms_norm support bias (#5790)

### What this PR does / why we need it?
This PR is to replace addRmsNorm and Add With addRmsNormBias. This way
can lead to a more effecient result.

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
Full Test Pass

- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef

Signed-off-by: Chen_HaoWen <chenhaowen12@huawei.com>
Co-authored-by: Chen_HaoWen <chenhaowen12@huawei.com>
This commit is contained in:
yjmyl
2026-01-23 21:09:54 +08:00
committed by GitHub
parent 6c73b88dd6
commit e90b14140b
24 changed files with 3537 additions and 13 deletions

View File

@@ -24,7 +24,7 @@ elif [[ "$SOC_VERSION" =~ ^ascend910b ]]; then
ABSOLUTE_CATLASS_PATH=$(cd "${CATLASS_PATH}" && pwd)
export CPATH=${ABSOLUTE_CATLASS_PATH}:${CPATH}
CUSTOM_OPS="grouped_matmul_swiglu_quant_weight_nz_tensor_list;lightning_indexer;sparse_flash_attention;matmul_allreduce_add_rmsnorm;moe_init_routing_custom;moe_gating_top_k;"
CUSTOM_OPS="grouped_matmul_swiglu_quant_weight_nz_tensor_list;lightning_indexer;sparse_flash_attention;matmul_allreduce_add_rmsnorm;moe_init_routing_custom;moe_gating_top_k;add_rms_norm_bias;"
SOC_ARG="ascend910b"
elif [[ "$SOC_VERSION" =~ ^ascend910_93 ]]; then
# ASCEND910C (A3) series
@@ -79,6 +79,7 @@ elif [[ "$SOC_VERSION" =~ ^ascend910_93 ]]; then
"notify_dispatch"
"moe_init_routing_custom"
"moe_gating_top_k"
"add_rms_norm_bias"
)
CUSTOM_OPS=$(IFS=';'; echo "${CUSTOM_OPS_ARRAY[*]}")
SOC_ARG="ascend910_93"