chore: bump vLLM and vLLM-Omni to 0.24.0

docs/CODE_STYLE.md

@@ -98,7 +98,7 @@ The worker GPU `vllm` is pinned to the `+cu129` release wheel (the PyPI
 wheel is built for CUDA 13, incompatible with the CUDA 12.9 worker). Its
 local version is not on PyPI, so pip-audit skips it — like
 `flashinfer-jit-cache` — which is why the GPU run omits `--strict`. vLLM
-CVE exposure tracks PyPI `vllm 0.22.0` regardless of the build variant.
+CVE exposure tracks PyPI `vllm 0.24.0` regardless of the build variant.
 
 | Advisory | Package | Fix version | Why ignored |
 |----------|---------|-------------|-------------|

pyproject.toml

@@ -69,9 +69,9 @@ runtime-rag = [
 ]
 runtime-inference-gpu = [
     "bitsandbytes>=0.49.2",
-    "flashinfer-python==0.6.11.post2",
-    "vllm==0.22.0",
-    "vllm-omni==0.22.0",
+    "flashinfer-python==0.6.12",
+    "vllm==0.24.0",
+    "vllm-omni==0.24.0",
 ]
 runtime-training-gpu = [
     "deepspeed>=0.19.1",
@@ -243,7 +243,7 @@ flowmesh-sdk-stack = { workspace = true }
 # vLLM's PyPI wheel is built for CUDA 13 (libcudart.so.13); the GPU worker runs
 # CUDA 12.9. Pin the cu129 release wheel for linux/x86_64 so it matches torch
 # (UV_TORCH_BACKEND=cu129) and flashinfer; other platforms fall back to PyPI.
-vllm = { url = "https://github.com/vllm-project/vllm/releases/download/v0.22.0/vllm-0.22.0+cu129-cp38-abi3-manylinux_2_28_x86_64.whl", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" }
+vllm = { url = "https://github.com/vllm-project/vllm/releases/download/v0.24.0/vllm-0.24.0+cu129-cp38-abi3-manylinux_2_28_x86_64.whl", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" }
 
 [tool.uv.workspace]
 members = ["cli", "cli/stack", "hook", "sdk", "sdk/stack"]

src/worker/requirements/requirements.gpu.txt

@@ -3,9 +3,9 @@
 --extra-index-url https://flashinfer.ai/whl/cu129
 bitsandbytes==0.49.2
 deepspeed==0.19.1
-flashinfer-cubin==0.6.11.post2
-flashinfer-jit-cache==0.6.11.post2
-flashinfer-python==0.6.11.post2
-vllm @ https://github.com/vllm-project/vllm/releases/download/v0.22.0/vllm-0.22.0+cu129-cp38-abi3-manylinux_2_28_x86_64.whl ; platform_machine == 'x86_64' and sys_platform == 'linux'
-vllm-omni==0.22.0
-vllm==0.22.0 ; platform_machine != 'x86_64' or sys_platform != 'linux'
+flashinfer-cubin==0.6.12
+flashinfer-jit-cache==0.6.12
+flashinfer-python==0.6.12
+vllm @ https://github.com/vllm-project/vllm/releases/download/v0.24.0/vllm-0.24.0+cu129-cp38-abi3-manylinux_2_28_x86_64.whl ; platform_machine == 'x86_64' and sys_platform == 'linux'
+vllm-omni==0.24.0
+vllm==0.24.0 ; platform_machine != 'x86_64' or sys_platform != 'linux'