【Hackathon 9th No.86】autogen MultiQueryDecoderAttention template_instantiation -part (#4383)

* split MultiQueryDecoderAttention template_instantiation

* update comment

* CI

custom_ops/setup_ops.py

@@ -353,6 +353,8 @@ elif paddle.is_compiled_with_cuda():
         "-Igpu_ops",
         "-Ithird_party/nlohmann_json/include",
     ]
+    worker_threads = os.cpu_count()
+    nvcc_compile_args += ["-t", str(worker_threads)]
 
     nvcc_version = get_nvcc_version()
     print(f"nvcc_version = {nvcc_version}")