chore: bump vLLM to 0.12.0 (#4736)

Signed-off-by: alec-flowers <aflowers@nvidia.com>
Signed-off-by: Karen Chung <karenc@nvidia.com>
Signed-off-by: jthomson04 <jwillthomson19@gmail.com>
Co-authored-by: alec-flowers <aflowers@nvidia.com>
Co-authored-by: jthomson04 <jwillthomson19@gmail.com>

container/Dockerfile.vllm

@@ -15,14 +15,13 @@ ARG RUNTIME_IMAGE_TAG="12.9.0-runtime-ubuntu24.04"
 ARG CUDA_VERSION="12.9"
 
 # Make sure to update the dependency version in pyproject.toml when updating this
-ARG VLLM_REF="v0.11.2"
+ARG VLLM_REF="v0.12.0"
 # FlashInfer Ref used to install flashinfer-cubin and flashinfer-jit-cache
-ARG FLASHINF_REF="v0.5.2"
+ARG FLASHINF_REF="v0.5.3"
 
 # If left blank, then we will fallback to vLLM defaults
 ARG DEEPGEMM_REF=""
-# LMCache version - 0.3.9+ required for vLLM 0.11.2 compatibility
-ARG LMCACHE_REF="0.3.9.post2"
+ARG LMCACHE_REF="0.3.10"
 
 # sccache configuration - inherit from base build
 ARG USE_SCCACHE

container/deps/vllm/install_vllm.sh

@@ -11,7 +11,7 @@
 
 set -euo pipefail
 
-VLLM_REF="v0.11.2"
+VLLM_REF="v0.12.0"
 
 # Basic Configurations
 ARCH=$(uname -m)
@@ -22,9 +22,9 @@ INSTALLATION_DIR=/tmp
 TORCH_CUDA_ARCH_LIST="9.0;10.0" # For EP Kernels
 DEEPGEMM_REF=""
 CUDA_VERSION="12.9"
-FLASHINF_REF="v0.5.2"
+FLASHINF_REF="v0.5.3"
 # LMCache version - 0.3.9+ required for vLLM 0.11.2 compatibility
-LMCACHE_REF="0.3.9.post2"
+LMCACHE_REF="0.3.10"
 
 while [[ $# -gt 0 ]]; do
     case $1 in

pyproject.toml

@@ -56,7 +56,7 @@ trtllm =[
 vllm = [
     "uvloop",
     "nixl[cu12]<=0.7.1",
-    "vllm[flashinfer]==0.11.2",
+    "vllm[flashinfer]==0.12.0",
 ]
 
 sglang = [