Centralize default deps versions in a single source-of-truth file

[doringeman]

Default versions for vLLM, SGLang, and llama-server are currently duplicated across multiple files (Dockerfile, Makefile .github/workflows/release.yml, scripts, README). This makes bumping a version error-prone as they easily drift out of sync (e.g. vLLM was 0.12.0 in CI but 0.17.0 in the Dockerfile).

Introduce a single source-of-truth file (e.g. .versions or .env) that defines the defaults, and have each consumer read from it:

• Makefile: VLLM_VERSION ?= $(shell ...)
• Dockerfile: remove ARG defaults; always pass via --build-arg
• CI workflow: read from the file in a setup step, remove hardcoded input defaults
• Shell scripts: source the file or read the relevant value

This ensures a version bump is a one-line change in one file.

[vb-hmn]

I would like to take this issue up.

[doringeman]

Take it away @vb-hmn! 😁

[ayushgharat234]

Are you working on this @vb-hmn ? I was wanting to contribute

[vb-hmn]

@ayushgharat234 Yes I have it almost ready

[ayushgharat234]

@ayushgharat234 Yes I have it almost ready

Good luck

[alimx07]

Hi @vb-hmn , are you still working on this? I noticed the repository hasn’t been forked yet.
If not, @doringeman, I’d be happy to take on this issue.

[vb-hmn]

I got caught up on work commitments, I wont be able to finish this. Feel free to assign to someone who can.

[ayushgharat234]

Can someone assign this to me

[ayushgharat234]

I got caught up on work commitments, I wont be able to finish this. Feel free to assign to someone who can.

Can you tell someone to /assign it to me?

[doringeman]

Hello, anyone can take it, just discuss beforehand. You can take a look at other issues, too, even if they're not marked as "good first issue", because they still might be good first issues! 😁