chore: upgrade flashinfer 0.4.0 (#11364)

python/pyproject.toml

@@ -24,7 +24,7 @@ dependencies = [
   "datasets",
   "einops",
   "fastapi",
-  "flashinfer_python==0.4.0rc3",
+  "flashinfer_python==0.4.0",
   "hf_transfer",
   "huggingface_hub",
   "interegular",

python/pyproject_other.toml

@@ -70,7 +70,7 @@ srt = [
     "torchaudio==2.8.0",
     "torchvision",
     "cuda-python",
-    "flashinfer_python==0.4.0rc3",
+    "flashinfer_python==0.4.0",
 ]
 
 # HIP (Heterogeneous-computing Interface for Portability) for AMD

python/sglang/srt/entrypoints/engine.py

@@ -703,7 +703,7 @@ def _set_envs_and_config(server_args: ServerArgs):
     if server_args.attention_backend == "flashinfer":
         assert_pkg_version(
             "flashinfer_python",
-            "0.4.0rc3",
+            "0.4.0",
             "Please uninstall the old version and "
             "reinstall the latest version by following the instructions "
             "at https://docs.flashinfer.ai/installation.html.",

python/sglang/srt/layers/attention/flashinfer_mla_backend.py

@@ -1060,7 +1060,7 @@ def fast_mla_decode_plan(
 
     try:
         # Standard version with just the required arguments (no use_profiler)
-        self._cached_module.plan.default(
+        self._cached_module.plan(
             self._float_workspace_buffer,
             self._int_workspace_buffer,
             self._pin_memory_int_workspace_buffer,