[Installation]: v0.5.3.post1+Gaudi-1.18.0 has unknown manifest with 2.4.1 pytorch image

[vbedida79]

Your current environment

The output of `python collect_env.py`

How you are installing vllm

We are using v0.5.3.post1+Gaudi-1.18.0 and image https://github.com/HabanaAI/vllm-fork/blob/v0.5.3.post1%2BGaudi-1.18.0/Dockerfile.hpu to run a sample Vllm server with Gaudi on OpenShift

The image vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.4.1:latest used in Dockerfile.hpu does not exist 24.04 link and 22.04 link. Hence image build fails with manifest unknown.
The instructions in release tag (Run Docker image) https://github.com/HabanaAI/vllm-fork/releases/tag/v0.5.3.post1%2BGaudi-1.18.0 also point to this image instead of 2.4.0.

This is updated with 2.4.0 image in main branch, but not the latest release v0.5.3.post1+Gaudi-1.18.0.

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.