[Bugfix] Fix triton placeholder patch period (#704)

Fix triton placeholder patch period

Signed-off-by: MengqingCao <cmq0113@163.com>

vllm_ascend/__init__.py

@@ -23,5 +23,9 @@ def register():
 
 
 def register_model():
+    # TODO: fixme when TritonPlaceholder fixed
+    from vllm_ascend.utils import vllm_version_is
+    if vllm_version_is("0.8.4"):
+        import vllm_ascend.patch.worker.patch_0_8_4.patch_tritonplaceholder  # noqa
     from .models import register_model
     register_model()

vllm_ascend/patch/worker/patch_0_8_4/patch_tritonplaceholder.py

@@ -17,6 +17,7 @@
 # Adapted from vllm/triton_utils/importing.py
 #
 
+import importlib
 import sys
 import types
 from importlib.util import find_spec
@@ -40,6 +41,8 @@ if not HAS_TRITON:
             self.autotune = self._dummy_decorator("autotune")
             self.heuristics = self._dummy_decorator("heuristics")
             self.language = TritonLanguagePlaceholder()
+            self.__spec__ = importlib.machinery.ModuleSpec(
+                name="triton", loader=None, origin="placeholder")
             logger.warning_once(
                 "Triton is not installed. Using dummy decorators. "
                 "Install it via `pip install triton` to enable kernel"