[Doc] xpu backend requires running setvars.sh

[rscohn2]

IPEX depends on MKL, and you need to do setvars.sh to setup the LD_LIBRARY_PATH to find it. It works fine with with setvars.sh and 2024.2 oneapi.

@jikunshang: Thanks for adding XPU support!

Here is the error message:

(.venv) rscohn1@gkdse-dnp-23:~/projects/cutlass/vllm$ python examples/offline_inference.py 
WARNING 07-12 23:25:17 _custom_ops.py:14] Failed to import from vllm._C with ModuleNotFoundError("No module named 'vllm._C'")
WARNING 07-12 23:25:17 utils.py:218] Import Error for IPEX: libmkl_intel_lp64.so.2: cannot open shared object file: No such file or directory
WARNING 07-12 23:25:17 utils.py:222] not found ipex lib

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only trigger fastcheck CI to run, which consists only a small and essential subset of tests to quickly catch small errors.

Full CI run is still required to merge this PR so please make sure that you run full CI before merging or if you need more test signals.

To run full CI, you can do one of these:

• Add ready label to the PR
• Comment /ready on the PR
• Enable auto-merge.

🚀

[jikunshang]

LGTM! Thanks for this fix!

[WoosukKwon]

LGTM. Thanks for the fix!

[zhouyuan]

Note: it's recommended to picked the "paired" version of oneAPI and IPEX in the XPU docker file as it's more tested
https://github.com/vllm-project/vllm/blob/main/Dockerfile.xpu

thanks, -yuan