[chore]Upgrade flashinfer to 0.5.3 (#13751)

docker/Dockerfile

@@ -18,7 +18,7 @@ ARG PIP_DEFAULT_INDEX
 ARG UBUNTU_MIRROR
 ARG GITHUB_ARTIFACTORY=github.com
 ARG INSTALL_FLASHINFER_JIT_CACHE=0
-ARG FLASHINFER_VERSION=0.5.2
+ARG FLASHINFER_VERSION=0.5.3
 
 ENV DEBIAN_FRONTEND=noninteractive \
     CUDA_HOME=/usr/local/cuda \

python/pyproject.toml

@@ -26,8 +26,8 @@ dependencies = [
   "datasets",
   "einops",
   "fastapi",
-  "flashinfer_python==0.5.2", # keep it aligned with jit-cache version in Dockerfile
-  "flashinfer_cubin==0.5.2",
+  "flashinfer_python==0.5.3", # keep it aligned with jit-cache version in Dockerfile
+  # "flashinfer_cubin==0.5.3", # TODO: Add it back when the cubin is ready: https://github.com/flashinfer-ai/flashinfer/issues/2133
   "gguf",
   "hf_transfer",
   "huggingface_hub",

python/sglang/srt/entrypoints/engine.py

@@ -727,7 +727,7 @@ def _set_envs_and_config(server_args: ServerArgs):
     if server_args.attention_backend == "flashinfer":
         assert_pkg_version(
             "flashinfer_python",
-            "0.5.2",
+            "0.5.3",
             "Please uninstall the old version and "
             "reinstall the latest version by following the instructions "
             "at https://docs.flashinfer.ai/installation.html.",