From ac9370b1c9a3762059461c037b6992605e305313 Mon Sep 17 00:00:00 2001
From: Eric Curtin <eric.curtin@docker.com>
Date: Tue, 3 Mar 2026 11:53:36 +0000
Subject: [PATCH] Bump vllm to 0.16.0 and install from wheel

Install vllm 0.16.0 from the macOS wheel on PyPI instead of
building from source, and update the optional dependency lower
bound to match.

Signed-off-by: Eric Curtin <eric.curtin@docker.com>
---
 install.sh     | 12 ++----------
 pyproject.toml |  2 +-
 2 files changed, 3 insertions(+), 11 deletions(-)

diff --git a/install.sh b/install.sh
index a7344266..97543dc1 100755
--- a/install.sh
+++ b/install.sh
@@ -123,16 +123,8 @@ main() {
 
   ensure_venv "$venv"
 
-  local vllm_v="0.14.1"
-  local url_base="https://github.com/vllm-project/vllm/releases/download"
-  local filename="vllm-$vllm_v.tar.gz"
-  curl -OL $url_base/v$vllm_v/$filename
-  tar xf $filename
-  cd vllm-$vllm_v
-  uv pip install -r requirements/cpu.txt --index-strategy unsafe-best-match
-  uv pip install .
-  cd -
-  rm -rf vllm-$vllm_v*
+  local vllm_v="0.16.0"
+  uv pip install "vllm==$vllm_v"
 
   if [[ -n "$local_lib" && -f "$local_lib" ]]; then
     uv pip install .
diff --git a/pyproject.toml b/pyproject.toml
index 763c53d2..aef6ab0e 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -45,7 +45,7 @@ paged = [
     # Paged attention Metal kernel (opt-in, experimental)
     "kernels>=0.4.5; platform_system == 'Darwin' and platform_machine == 'arm64'",
 ]
-vllm = ["vllm>=0.14.0"]
+vllm = ["vllm>=0.16.0"]
 stt = [
     # Speech-to-text audio processing (Whisper models)
     "librosa>=0.10.2",