EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[MOE] commit GMM custom operator (#7010)

Browse Source 
### What this PR does / why we need it?
GMM custom operator optimization in small batch scenarios

### How was this patch tested?
Submit the GMM custom operator for subsequent integration into the MOE
process.


- vLLM version: v0.16.0
- vLLM main:
15d76f74e2

---------

Signed-off-by: chenxi-hh <chen464822955@163.com>
Signed-off-by: chenxi-hh <32731611+chenxi-hh@users.noreply.github.com>
This commit is contained in:
chenxi-hh
2026-03-09 09:56:31 +08:00
committed by GitHub
parent 01d3515dcf
commit 737dfcf638
16 changed files with 1214 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
csrc/build_aclnn.sh
View File

@@ -24,7 +24,7 @@ elif [[ "$SOC_VERSION" =~ ^ascend910b ]]; then
ABSOLUTE_CATLASS_PATH=$(cd "${CATLASS_PATH}" && pwd)
export CPATH=${ABSOLUTE_CATLASS_PATH}:${CPATH}
CUSTOM_OPS="grouped_matmul_swiglu_quant_weight_nz_tensor_list;lightning_indexer_vllm;sparse_flash_attention;matmul_allreduce_add_rmsnorm;moe_init_routing_custom;moe_gating_top_k;add_rms_norm_bias;apply_top_k_top_p_custom;transpose_kv_cache_by_block;"
CUSTOM_OPS="moe_grouped_matmul;grouped_matmul_swiglu_quant_weight_nz_tensor_list;lightning_indexer_vllm;sparse_flash_attention;matmul_allreduce_add_rmsnorm;moe_init_routing_custom;moe_gating_top_k;add_rms_norm_bias;apply_top_k_top_p_custom;transpose_kv_cache_by_block;"
SOC_ARG="ascend910b"
elif [[ "$SOC_VERSION" =~ ^ascend910_93 ]]; then
# ASCEND910C (A3) series
@@ -82,6 +82,7 @@ elif [[ "$SOC_VERSION" =~ ^ascend910_93 ]]; then
"add_rms_norm_bias"
"apply_top_k_top_p_custom"
"transpose_kv_cache_by_block"
"moe_grouped_matmul"
)
CUSTOM_OPS=$(IFS=';'; echo "${CUSTOM_OPS_ARRAY[*]}")
SOC_ARG="ascend910_93"