[projects][vllm] Use custom index instead of manual wheel downloading

sasha0552 · Jun 3, 2024 · fd2b60c · fd2b60c
1 parent b801896
commit fd2b60c
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/airootfs/root/customize_airootfs/scripts/1000-vllm.sh.jinja2 b/airootfs/root/customize_airootfs/scripts/1000-vllm.sh.jinja2
@@ -2,11 +2,11 @@
 set -eu
 
 {% if platform == "cuda" %}
-# install vllm
-pipx install "{{ gh_release("vllm", 1) }}"
+# use patched vllm
+export PIP_EXTRA_INDEX_URL="https://sasha0552.github.io/vllm-ci/"
 
-# install triton with pascal patches
-pipx runpip vllm install --force-reinstall --no-deps --upgrade "{{ gh_release("vllm", 0) }}"
+# install vllm
+pipx install vllm
 
 # install nvidia-pstate
 pipx inject vllm nvidia-pstate