[10c/n] Migrate MoE kernels to torch stable ABI (#44565)

Signed-off-by: Chris Leonard <chleonar@redhat.com>
Co-authored-by: Shengqi Chen <harry-chen@outlook.com>
This commit is contained in:
Chris Leonard
2026-06-10 23:02:26 -07:00
committed by GitHub
co-authored by Shengqi Chen
parent 2f2c5cf4f1
commit 6e64c1bab1
34 changed files with 1296 additions and 1098 deletions
+2 -2
View File
@@ -755,7 +755,7 @@ class precompiled_wheel_utils:
{
"vllm/_C.abi3.so",
"vllm/_C_stable_libtorch.abi3.so",
"vllm/_moe_C.abi3.so",
"vllm/_moe_C_stable_libtorch.abi3.so",
"vllm/_flashmla_C.abi3.so",
"vllm/_flashmla_extension_C.abi3.so",
"vllm/_sparse_flashmla_C.abi3.so",
@@ -1081,7 +1081,6 @@ def get_requirements() -> list[str]:
ext_modules = []
if _is_cuda() or _is_hip():
ext_modules.append(CMakeExtension(name="vllm._moe_C"))
ext_modules.append(CMakeExtension(name="vllm.cumem_allocator"))
# Optional since this doesn't get built (produce an .so file). This is just
# copying the relevant .py files from the source repository.
@@ -1135,6 +1134,7 @@ if _build_custom_ops():
ext_modules.append(CMakeExtension(name="vllm._C"))
if _is_cuda() or _is_hip():
ext_modules.append(CMakeExtension(name="vllm._C_stable_libtorch"))
ext_modules.append(CMakeExtension(name="vllm._moe_C_stable_libtorch"))
package_data = {
"vllm": [