chore: upgrade sgl-kernel 0.3.10 (#10500)

docker/Dockerfile

@@ -85,7 +85,7 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip setuptools wheel html5li
  && python3 -m pip install --no-cache-dir nvidia-nccl-cu12==2.27.6 --force-reinstall --no-deps \
  && python3 -m flashinfer --download-cubin \
  && if [ "$CUDA_VERSION" = "12.6.1" ]; then \
-      python3 -m pip install --no-cache-dir https://github.com/sgl-project/whl/releases/download/v0.3.9.post2/sgl_kernel-0.3.9.post2+cu124-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall --no-deps ; \
+      python3 -m pip install --no-cache-dir https://github.com/sgl-project/whl/releases/download/v0.3.10/sgl_kernel-0.3.10+cu124-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall --no-deps ; \
     fi
 
 # Download source files

docker/Dockerfile.gb200

@@ -4,7 +4,7 @@ FROM nvidia/cuda:${CUDA_VERSION}-cudnn-devel-ubuntu22.04
 ARG BUILD_TYPE=blackwell
 ARG DEEPEP_COMMIT=1b14ad661c7640137fcfe93cccb2694ede1220b0
 ARG CMAKE_BUILD_PARALLEL_LEVEL=2
-ARG SGL_KERNEL_VERSION=0.3.9.post2
+ARG SGL_KERNEL_VERSION=0.3.10
 ENV DEBIAN_FRONTEND=noninteractive \
     CUDA_HOME=/usr/local/cuda \
     GDRCOPY_HOME=/usr/src/gdrdrv-2.4.4/ \

python/pyproject.toml

@@ -57,7 +57,7 @@ dependencies = [
   "uvicorn",
   "uvloop",
   "xgrammar==0.1.24",
-  "sgl-kernel==0.3.9.post2",
+  "sgl-kernel==0.3.10",
   "torch==2.8.0",
   "torchaudio==2.8.0",
   "torchvision",
@@ -67,10 +67,10 @@ dependencies = [
   "tiktoken",
   "anthropic>=0.20.0",
   "torch_memory_saver==0.0.8",
-  "decord",
 ]
 
 [project.optional-dependencies]
+decord = ["decord"]
 test = [
   "accelerate",
   "expecttest",
@@ -88,8 +88,8 @@ tracing = [
     "opentelemetry-exporter-otlp",
     "opentelemetry-exporter-otlp-proto-grpc",
 ]
-all = ["sglang[test]"]
-blackwell = ["nvidia-cutlass-dsl==4.1.0", "sglang[test]"]
+all = ["sglang[test]", "sglang[decord]"]
+blackwell = ["nvidia-cutlass-dsl==4.2.0", "sglang[test]"]
 dev = ["sglang[test]"]
 
 [project.urls]

python/pyproject_other.toml

@@ -65,7 +65,7 @@ tracing = [
 
 srt = [
     "sglang[runtime_common]",
-    "sgl-kernel==0.3.9.post2",
+    "sgl-kernel==0.3.10",
     "torch==2.8.0",
     "torchaudio==2.8.0",
     "torchvision",
@@ -75,13 +75,13 @@ srt = [
 
 blackwell = [
     "sglang[runtime_common]",
-    "sgl-kernel==0.3.9.post2",
+    "sgl-kernel==0.3.10",
     "torch==2.8.0",
     "torchaudio==2.8.0",
     "torchvision",
     "cuda-python",
     "flashinfer_python==0.3.1",
-    "nvidia-cutlass-dsl==4.1.0",
+    "nvidia-cutlass-dsl==4.2.0",
 ]
 
 # HIP (Heterogeneous-computing Interface for Portability) for AMD

python/sglang/srt/entrypoints/engine.py

@@ -690,7 +690,7 @@ def _set_envs_and_config(server_args: ServerArgs):
     if _is_cuda and not get_bool_env_var("SGLANG_SKIP_SGL_KERNEL_VERSION_CHECK"):
         assert_pkg_version(
             "sgl-kernel",
-            "0.3.9.post2",
+            "0.3.10",
             "Please reinstall the latest version with `pip install sgl-kernel --force-reinstall`",
         )