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

[Feature]: Support 310P device run qwen2.5/3 dense and qwen2.5vl models (#5776)

Browse Source 
### What this PR does / why we need it?
Add basic 310p support. Only dense models work with eager mode now.

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

---------

Signed-off-by: Tflowers-0129 <2906339855@qq.com>
Signed-off-by: Shaoxu Cheng <2906339855@qq.com>
This commit is contained in:
Shaoxu Cheng
2026-01-17 11:49:18 +08:00
committed by GitHub
parent 7feb74590b
commit 1ffca8673f
17 changed files with 682 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
CMakeLists.txt
View File

@@ -62,14 +62,17 @@ set(VLLM_ASCEND_CUSTOM_OP
)
set(VLLM_ASCEND_CUSTOM_OP_EXCLUDE
${KERNEL_FILES}/bgmv_expand.cpp
${KERNEL_FILES}/bgmv_shrink.cpp
${KERNEL_FILES}/sgmv_expand.cpp
${KERNEL_FILES}/sgmv_shrink.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/kernels/bgmv_expand.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/kernels/bgmv_shrink.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/kernels/sgmv_expand.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/kernels/sgmv_shrink.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/mla_preprocess/op_kernel/mla_preprocess_kernel.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/batch_matmul_transpose/op_kernel/batch_matmul_transpose_kernel.cpp
)
if(SOC_VERSION STREQUAL "ASCEND310P3")
if(SOC_VERSION STREQUAL "ascend310p3")
message(STATUS "310P hardware detected: disabling MLAPO operators")
message(STATUS "310P hardware detected: excluding batch_matmul_transpose operators")
list(REMOVE_ITEM VLLM_ASCEND_CUSTOM_OP ${VLLM_ASCEND_CUSTOM_OP_EXCLUDE})
endif()
@@ -79,7 +82,7 @@ ascendc_library(vllm_ascend_kernels SHARED
message("TORCH_NPU_PATH is ${TORCH_NPU_PATH}")
if(SOC_VERSION STREQUAL "ASCEND310P3")
if(SOC_VERSION STREQUAL "ascend310p3")
file(GLOB VLLM_ASCEND_SRC
${CMAKE_CURRENT_SOURCE_DIR}/csrc/*.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/aclnn_torch_adapter/*.cpp)