The best way to upgrade vLLM is actually NOT to upgrade it, but to wipe it’s venv and install from scratch, in my opinion and based on my experimentation, just run:

deactivate

rm -rf .venv

uv venv --python 3.12 --seed --managed-python
source .venv/bin/activate

uv pip install vllm --torch-backend=auto

That’s it, Enjoy!

Sources: My experiments + Setup guidelines from https://docs.vllm.ai/en/latest/getting_started/installation/gpu/#using-vllm-metal