diff --git a/python/pyproject.toml b/python/pyproject.toml
index 0f5a6de6e..9be8cb833 100755
--- a/python/pyproject.toml
+++ b/python/pyproject.toml
@@ -1,5 +1,5 @@
 [build-system]
-requires = ["setuptools>=61.0", "setuptools-scm>=8.0", "setuptools-rust>=1.10", "wheel"]
+requires = ["setuptools>=61.0", "setuptools-rust>=1.10", "setuptools-scm>=8.0", "wheel"]
 build-backend = "setuptools.build_meta"
 
 [project]
@@ -14,35 +14,40 @@ classifiers = [
   "License :: OSI Approved :: Apache Software License",
 ]
 
+# Please keep dependency lists in this file sorted alphabetically by package name.
 dependencies = [
-  "IPython",
   "aiohttp",
-  "apache-tvm-ffi==0.1.9",
   "anthropic>=0.20.0",
+  "apache-tvm-ffi==0.1.9",
+  "av ; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64' or platform_machine == 'armv7l')",
   "blobfile==3.0.0",
   "build",
   "compressed-tensors",
   "cuda-python>=13.0",
-  "decord2 ; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64' or platform_machine == 'armv7l')",
   "datasets",
+  "decord2 ; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64' or platform_machine == 'armv7l')",
   "distro",
+  "easydict",  # Required by remote model code (e.g. DeepSeek-OCR) loaded via trust_remote_code; validated by transformers 5.4+ check_imports
   "einops",
   "fastapi",
-  "flashinfer_python[cu13]==0.6.12", # keep it aligned with jit-cache version in Dockerfile
+  "flash-attn-4>=4.0.0b9",
   "flashinfer_cubin==0.6.12",
+  "flashinfer_python[cu13]==0.6.12", # keep it aligned with jit-cache version in Dockerfile
   "gguf",
   "interegular",
+  "IPython",
+  "kernels>=0.14.1,<0.15",
   "llguidance>=0.7.11,<0.8.0",
+  "mistral_common>=1.11.0",
   "modelscope",
   "msgspec",
   "ninja",
-  "easydict",  # Required by remote model code (e.g. DeepSeek-OCR) loaded via trust_remote_code; validated by transformers 5.4+ check_imports
   "numpy",
   "nvidia-cutlass-dsl[cu13]==4.5.2",
-  "nvidia-ml-py",
   "nvidia-mathdx==25.6.0",
-  "openai-harmony==0.0.4",
+  "nvidia-ml-py",
   "openai==2.6.1",
+  "openai-harmony==0.0.4",
   "orjson",
   "outlines==0.1.11",
   "packaging",
@@ -60,30 +65,26 @@ dependencies = [
   "scipy",
   "sentencepiece",
   "setproctitle",
-  "flash-attn-4>=4.0.0b9",
   "sgl-deep-gemm==0.1.3",
   "sglang-kernel==0.4.4",
+  "smg-grpc-servicer>=0.5.0",
   "soundfile==0.13.1",
   "tiktoken",
   "tilelang==0.1.8",
   "timm==1.0.16",
   "tokenspeed_mla==0.1.6",
-  "torch_memory_saver>=0.0.9.post1",
   "torch==2.11.0",
+  "torch_memory_saver>=0.0.9.post1",
   "torchao==0.17.0",
   "torchaudio==2.11.0",
   "torchcodec==0.11.1 ; sys_platform != 'linux' or (sys_platform == 'linux' and platform_machine != 'aarch64' and platform_machine != 'arm64' and platform_machine != 'armv7l')", # torchcodec 0.11.1 for torch 2.11.x (0.10 is ABI-incompatible: references the pre-2.11 c10::MessageLogger ctor signature). Not available on Linux ARM.
-  "av ; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64' or platform_machine == 'armv7l')",
   "torchvision",
   "tqdm",
-  "mistral_common>=1.11.0",
   "transformers==5.8.1",
   "uvicorn",
   "uvloop",
   "watchfiles",
   "xgrammar==0.2.1",
-  "smg-grpc-servicer>=0.5.0",
-  "kernels>=0.14.1,<0.15",
 ]
 
 [[tool.uv.index]]
@@ -95,7 +96,9 @@ default = true
 checkpoint-engine = ["checkpoint-engine==0.1.2"]
 runai = ["runai-model-streamer[s3,gcs,azure]>=0.15.7"]
 diffusion = [
-  "PyYAML==6.0.1",
+  "addict==2.4.0",
+  "av==16.1.0",
+  "cache-dit==1.3.0",
   "cloudpickle==3.1.2",
   "diffusers==0.37.0",
   "imageio==2.36.0",
@@ -103,15 +106,13 @@ diffusion = [
   "moviepy>=2.0.0",
   "nvidia-modelopt",
   "opencv-python-headless==4.10.0.84",
+  "PyYAML==6.0.1",
   "remote-pdb==2.1.0",
-  "st_attn==0.0.7 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
-  "vsa==0.0.4 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "runai_model_streamer>=0.15.7",
-  "cache-dit==1.3.0",
-  "addict==2.4.0",
-  "av==16.1.0",
   "scikit-image==0.25.2",
+  "st_attn==0.0.7 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "trimesh>=4.0.0",
+  "vsa==0.0.4 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "websockets",
   "xatlas",
 ]
@@ -139,7 +140,9 @@ test = [
   "accelerate",
   "addict",
   "bitsandbytes",
+  "diff-cover",
   "expecttest",
+  "granian>=2.6.0",
   "jsonlines",
   "lm-eval[api]>=0.4.9.2",
   "matplotlib",
@@ -149,24 +152,22 @@ test = [
   "polars",
   "pytest",
   "pytest-cov",
-  "diff-cover",
   "sentence_transformers",
   "sglang[fastokens]",
   "tabulate",
-  "granian>=2.6.0",
 ]
 
 dev = ["sglang[test]"]
 
 all = [
   "sglang[diffusion]",
-  "sglang[tracing]",
   "sglang[http2]",
+  "sglang[tracing]",
 ]
 
 [tool.uv.extra-build-dependencies]
-st-attn = ["torch", "setuptools"]
-vsa = ["torch", "setuptools"]
+st-attn = ["setuptools", "torch"]
+vsa = ["setuptools", "torch"]
 
 [project.urls]
 "Homepage" = "https://github.com/sgl-project/sglang"
diff --git a/python/pyproject_cpu.toml b/python/pyproject_cpu.toml
index 58f03cb0a..f40064c55 100644
--- a/python/pyproject_cpu.toml
+++ b/python/pyproject_cpu.toml
@@ -15,27 +15,29 @@ classifiers = [
   "License :: OSI Approved :: Apache Software License",
 ]
 
+# Please keep dependency lists in this file sorted alphabetically by package name.
 dependencies = [
-  "IPython",
   "aiohttp",
   "anthropic>=0.20.0",
   "blobfile==3.0.0",
   "build",
   "compressed-tensors",
   "datasets",
+  "easydict",
   "einops",
   "fastapi",
   "gguf",
   "intel-openmp; platform_machine == 'x86_64'",
   "interegular",
+  "IPython",
   "llguidance>=0.7.11,<0.8.0",
+  "mistral_common>=1.11.0",
   "modelscope",
   "msgspec",
-  "easydict",
   "ninja",
   "numpy",
-  "openai-harmony==0.0.4",
   "openai==2.6.1",
+  "openai-harmony==0.0.4",
   "orjson",
   "outlines",
   "packaging",
@@ -52,6 +54,7 @@ dependencies = [
   "scipy",
   "sentencepiece",
   "setproctitle",
+  "smg-grpc-servicer>=0.5.0",
   "soundfile==0.13.1",
   "tabulate",
   "tiktoken",
@@ -61,41 +64,39 @@ dependencies = [
   "torchaudio==2.11.0",
   "torchvision==0.27.0",
   "tqdm",
-  "mistral_common>=1.11.0",
   "transformers==5.8.1",
   "triton==3.7.0",
   "uvicorn",
   "uvloop",
   "xgrammar==0.2.1",
-  "smg-grpc-servicer>=0.5.0",
 ]
 
 [project.optional-dependencies]
 diffusion = [
-  "PyYAML==6.0.1",
+  "addict==2.4.0",
+  "av==16.1.0",
+  "cache-dit==1.3.0",
   "cloudpickle==3.1.2",
   "diffusers==0.37.0",
   "imageio==2.36.0",
   "imageio-ffmpeg==0.5.1",
   "moviepy>=2.0.0",
   "opencv-python-headless==4.10.0.84",
+  "PyYAML==6.0.1",
   "remote-pdb==2.1.0",
-  "st_attn==0.0.7 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
-  "vsa==0.0.4 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "runai_model_streamer>=0.15.5",
-  "cache-dit==1.3.0",
-  "addict==2.4.0",
-  "av==16.1.0",
   "scikit-image==0.25.2",
+  "st_attn==0.0.7 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "trimesh>=4.0.0",
+  "vsa==0.0.4 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "xatlas",
 ]
 
 tracing = [
-  "opentelemetry-sdk",
   "opentelemetry-api",
   "opentelemetry-exporter-otlp",
   "opentelemetry-exporter-otlp-proto-grpc",
+  "opentelemetry-sdk",
 ]
 test = [
   "accelerate",
diff --git a/python/pyproject_npu.toml b/python/pyproject_npu.toml
index 79c846b08..1fdda9a3a 100644
--- a/python/pyproject_npu.toml
+++ b/python/pyproject_npu.toml
@@ -14,30 +14,32 @@ classifiers = [
   "License :: OSI Approved :: Apache Software License",
 ]
 
+# Please keep dependency lists in this file sorted alphabetically by package name.
 dependencies = [
-  "IPython",
   "aiohttp",
   "anthropic>=0.20.0",
-  "blobfile==3.0.0",
   "av",
+  "blobfile==3.0.0",
   "build",
   "compressed-tensors",
   "datasets",
   "decord2 ; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64' or platform_machine == 'armv7l')",
+  "easydict",
   "einops",
   "fastapi",
-  "easydict",
   "gguf",
   "hf_transfer",
   "huggingface_hub",
   "interegular",
+  "IPython",
   "llguidance>=0.7.11,<0.8.0",
+  "mistral_common>=1.11.0",
   "modelscope",
   "msgspec",
   "ninja",
   "numpy",
-  "openai-harmony==0.0.4",
   "openai==2.6.1",
+  "openai-harmony==0.0.4",
   "orjson",
   "outlines==0.1.11",
   "packaging",
@@ -54,35 +56,34 @@ dependencies = [
   "scipy",
   "sentencepiece",
   "setproctitle",
+  "smg-grpc-servicer>=0.5.0",
   "soundfile==0.13.1",
   "tiktoken",
   "timm==1.0.16",
   "torchao==0.9.0",
   "tqdm",
-  "mistral_common>=1.11.0",
   "transformers==5.8.1",
   "uvicorn",
   "uvloop",
   "xgrammar==0.2.1",
-  "smg-grpc-servicer>=0.5.0",
 ]
 
 [project.optional-dependencies]
 checkpoint-engine = ["checkpoint-engine==0.1.2"]
 diffusion = [
-    "PyYAML==6.0.1",
-    "cloudpickle",
-    "diffusers==0.37.0",
-    "imageio==2.36.0",
-    "imageio-ffmpeg==0.5.1",
-    "moviepy>=2.0.0",
-    "opencv-python==4.10.0.84",
-    "remote-pdb",
-    "cache-dit==1.3.5",
-    "addict",
-    "scikit-image==0.25.2",
-    "trimesh>=4.0.0",
-    "xatlas",
+  "addict",
+  "cache-dit==1.3.5",
+  "cloudpickle",
+  "diffusers==0.37.0",
+  "imageio==2.36.0",
+  "imageio-ffmpeg==0.5.1",
+  "moviepy>=2.0.0",
+  "opencv-python==4.10.0.84",
+  "PyYAML==6.0.1",
+  "remote-pdb",
+  "scikit-image==0.25.2",
+  "trimesh>=4.0.0",
+  "xatlas",
 ]
 
 tracing = [
diff --git a/python/pyproject_other.toml b/python/pyproject_other.toml
index 5334cb404..aa73f89ab 100755
--- a/python/pyproject_other.toml
+++ b/python/pyproject_other.toml
@@ -13,16 +13,16 @@ classifiers = [
   "Programming Language :: Python :: 3",
   "License :: OSI Approved :: Apache Software License",
 ]
-dependencies = ["aiohttp", "requests", "tqdm", "numpy", "IPython", "setproctitle"]
+# Please keep dependency lists in this file sorted alphabetically by package name.
+dependencies = ["aiohttp", "IPython", "numpy", "requests", "setproctitle", "tqdm"]
 
 [project.optional-dependencies]
 runtime_common = [
-  "IPython",
   "aiohttp",
-  "apache-tvm-ffi",
   "anthropic>=0.20.0",
-  "blobfile==3.0.0",
+  "apache-tvm-ffi",
   "av",
+  "blobfile==3.0.0",
   "build",
   "compressed-tensors",
   "datasets",
@@ -31,13 +31,15 @@ runtime_common = [
   "fastapi",
   "gguf",
   "interegular",
+  "IPython",
   "llguidance>=0.7.11,<0.8.0",
+  "mistral_common>=1.11.0",
   "modelscope",
   "msgspec",
   "ninja",
   "numpy",
-  "openai-harmony==0.0.4",
   "openai==2.6.1",
+  "openai-harmony==0.0.4",
   "orjson",
   "outlines==0.1.11",
   "packaging",
@@ -54,61 +56,60 @@ runtime_common = [
   "scipy",
   "sentencepiece",
   "setproctitle",
+  "smg-grpc-servicer>=0.5.0",
   "soundfile==0.13.1",
   "tiktoken",
   "timm==1.0.16",
   "torchao==0.9.0",
   "tqdm",
-  "mistral_common>=1.11.0",
   "transformers==5.8.1",
   "uvicorn",
   "uvloop",
   "xgrammar==0.2.1",
-  "smg-grpc-servicer>=0.5.0",
 ]
 
 diffusion_common = [
-  "PyYAML==6.0.1",
+  "addict",
   "cloudpickle",
   "diffusers==0.37.0",
   "imageio==2.36.0",
   "imageio-ffmpeg==0.5.1",
   "moviepy>=2.0.0",
   "opencv-python-headless==4.10.0.84",
+  "PyYAML==6.0.1",
   "remote-pdb",
-  "addict",
   "scikit-image==0.25.2",
   "trimesh>=4.0.0",
   "xatlas",
 ]
 
 tracing = [
-  "opentelemetry-sdk",
   "opentelemetry-api",
   "opentelemetry-exporter-otlp",
   "opentelemetry-exporter-otlp-proto-grpc",
+  "opentelemetry-sdk",
 ]
 
 # HIP (Heterogeneous-computing Interface for Portability) for AMD
 # => base docker rocm/vllm-dev:20250114, not from public vllm whl
 srt_hip = [
-  "sglang[runtime_common]",
-  "torch",
-  "petit_kernel==0.0.2",
-  "wave-lang==3.8.2",
   # Pin to 0.15.0: 0.16.0 needs torch>=2.10 (incompatible with ROCm torch
   # 2.9.1). An open-ended `<0.16.0` made pip backtrack into an unbuildable
   # ancient setuptools sdist; an exact pin keeps the resolver converging.
   "compressed-tensors==0.15.0",
+  "petit_kernel==0.0.2",
+  "sglang[runtime_common]",
+  "torch",
+  "wave-lang==3.8.2",
 ]
 
 diffusion_hip = [
-  "sglang[diffusion_common]",
+  "cache-dit==1.3.0",
   "peft>=0.18.0,<0.19.0", # Pin to <0.19.0 due to torchao incompatibility
+  "runai_model_streamer>=0.15.5",
+  "sglang[diffusion_common]",
   "st_attn==0.0.7",
   "vsa==0.0.4",
-  "runai_model_streamer>=0.15.5",
-  "cache-dit==1.3.0",
 ]
 
 # For Intel Gaudi(device : hpu) follow the installation guide
@@ -117,44 +118,44 @@ srt_hpu = ["sglang[runtime_common]"]
 
 # https://docs.sglang.io/platforms/mthreads_gpu.md
 srt_musa = [
+  "deep-gemm>=0.1.3",
+  "flash_attn_3>=0.1.4",
+  "mate>=0.2.0",
+  "mthreads-ml-py",
+  "numpy<2.0",
   "sglang[runtime_common]",
   "torch",
   "torch_musa",
   "torchada>=0.1.59",
-  "mthreads-ml-py",
-  "mate>=0.2.0",
-  "deep-gemm>=0.1.3",
-  "flash_attn_3>=0.1.4",
-  "numpy<2.0",
 ]
 
 diffusion_musa = [
+  "cache-dit==1.1.8",
+  "runai_model_streamer>=0.15.5",
   "sglang[diffusion_common]",
   "st_attn==0.0.7",
   "vsa==0.0.4",
-  "runai_model_streamer>=0.15.5",
-  "cache-dit==1.1.8",
 ]
 
 # https://docs.sglang.io/platforms/mps.md
 srt_mps = [
+    "mlx",
+    "mlx-lm",
     "sglang[runtime_common]",
     "torch==2.11.0",
     "torchao==0.9.0",
     "torchaudio==2.11.0",
     "torchvision",
-    "mlx",
-    "mlx-lm",
 ]
 
 diffusion_mps = [
-  "sglang[diffusion_common]",
-  "cloudpickle==3.1.2",
-  "remote-pdb==2.1.0",
-  "cache-dit==1.2.3",
   "addict==2.4.0",
   "av==16.1.0",
+  "cache-dit==1.2.3",
+  "cloudpickle==3.1.2",
+  "remote-pdb==2.1.0",
   "scikit-image==0.25.2",
+  "sglang[diffusion_common]",
   "trimesh>=4.0.0",
   "xatlas",
 ]
@@ -172,10 +173,10 @@ test = [
   "tabulate",
 ]
 
-all_hip = ["sglang[srt_hip]", "sglang[diffusion_hip]", "sglang[tracing]"]
+all_hip = ["sglang[diffusion_hip]", "sglang[srt_hip]", "sglang[tracing]"]
 all_hpu = ["sglang[srt_hpu]"]
-all_musa = ["sglang[srt_musa]", "sglang[diffusion_musa]"]
-all_mps = ["sglang[srt_mps]", "sglang[diffusion_mps]"]
+all_musa = ["sglang[diffusion_musa]", "sglang[srt_musa]"]
+all_mps = ["sglang[diffusion_mps]", "sglang[srt_mps]"]
 
 dev_hip = ["sglang[all_hip]", "sglang[test]"]
 dev_hpu = ["sglang[all_hpu]", "sglang[test]"]
diff --git a/python/pyproject_xpu.toml b/python/pyproject_xpu.toml
index b47aa707f..989fec648 100644
--- a/python/pyproject_xpu.toml
+++ b/python/pyproject_xpu.toml
@@ -14,33 +14,30 @@ classifiers = [
   "License :: OSI Approved :: Apache Software License",
 ]
 
+# Please keep dependency lists in this file sorted alphabetically by package name.
 dependencies = [
-  "torch==2.12.0+xpu",
-  "torchcodec==0.12.0 ; sys_platform != 'linux' or (sys_platform == 'linux' and platform_machine != 'aarch64' and platform_machine != 'arm64' and platform_machine != 'armv7l')", # torch==2.12.0 on XPU uses torchcodec 0.12.0
-  "av ; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64' or platform_machine == 'armv7l')",
-  "torchaudio==2.11.0+xpu",
-  "torchvision==0.27.0+xpu",
-  "sgl-kernel @ git+https://github.com/sgl-project/sgl-kernel-xpu.git",
-  "IPython",
+  "addict",
   "aiohttp",
   "anthropic>=0.20.0",
+  "av ; sys_platform == 'linux' and (platform_machine == 'aarch64' or platform_machine == 'arm64' or platform_machine == 'armv7l')",
   "blobfile==3.0.0",
   "build",
   "compressed-tensors",
-  "addict",
   "datasets",
   "easydict",
   "einops",
   "fastapi",
   "gguf",
   "interegular",
+  "IPython",
   "llguidance>=0.7.11,<0.8.0",
+  "mistral_common>=1.11.0",
   "modelscope",
   "msgspec",
   "ninja",
   "numpy",
-  "openai-harmony==0.0.4",
   "openai==2.6.1",
+  "openai-harmony==0.0.4",
   "orjson",
   "outlines==0.1.11",
   "packaging",
@@ -57,35 +54,39 @@ dependencies = [
   "scipy",
   "sentencepiece",
   "setproctitle",
+  "sgl-kernel @ git+https://github.com/sgl-project/sgl-kernel-xpu.git",
+  "smg-grpc-servicer>=0.5.0",
   "soundfile==0.13.1",
   "tiktoken",
   "timm==1.0.16",
+  "torch==2.12.0+xpu",
   "torchao==0.17.0+xpu",
+  "torchaudio==2.11.0+xpu",
+  "torchcodec==0.12.0 ; sys_platform != 'linux' or (sys_platform == 'linux' and platform_machine != 'aarch64' and platform_machine != 'arm64' and platform_machine != 'armv7l')", # torch==2.12.0 on XPU uses torchcodec 0.12.0
+  "torchvision==0.27.0+xpu",
   "tqdm",
-  "mistral_common>=1.11.0",
   "transformers==5.8.1",
   "uvicorn",
   "uvloop",
   # "xgrammar==0.2.1", xgrammar depends on CUDA PyTorch and Triton only
-  "smg-grpc-servicer>=0.5.0",
 ]
 
 [project.optional-dependencies]
 diffusion = [
-  "PyYAML==6.0.1",
+  "addict==2.4.0",
+  "av==16.1.0",
+  "cache-dit==1.3.0",
   "cloudpickle==3.1.2",
   "diffusers==0.38.0",
   "imageio==2.36.0",
   "imageio-ffmpeg==0.5.1",
   "moviepy>=2.0.0",
   "opencv-python==4.10.0.84",
+  "PyYAML==6.0.1",
   "remote-pdb==2.1.0",
-  "st_attn==0.0.7 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "runai_model_streamer>=0.15.5",
-  "cache-dit==1.3.0",
-  "addict==2.4.0",
-  "av==16.1.0",
   "scikit-image==0.25.2",
+  "st_attn==0.0.7 ; platform_machine != 'aarch64' and platform_machine != 'arm64'",
   "trimesh>=4.0.0",
   "xatlas",
 ]
diff --git a/python/sglang/launch_server.py b/python/sglang/launch_server.py
index 6b80e1330..04a5ef919 100644
--- a/python/sglang/launch_server.py
+++ b/python/sglang/launch_server.py
@@ -34,6 +34,7 @@ def run_server(server_args):
 
         asyncio.run(serve_grpc(server_args))
     elif server_args.use_ray:
+        # Ray mode: HTTP mode with Ray backend.
         try:
             from sglang.srt.ray.http_server import launch_server
         except ImportError: