[Misc] Remove Entrypoint Hijack for vLLM / 0.20.0 Changes#3082
Conversation
Signed-off-by: Alex Brooks <albrooks@redhat.com>
|
Please turn this into ready as upstream pr has been merged. |
|
Thanks @lishunyang12! Actually since we are going straight to |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3d9f1d1727
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| [project.scripts] | ||
| vllm = "vllm_omni.entrypoints.cli.main:main" | ||
| vllm-omni = "vllm_omni.entrypoints.cli.main:main" |
There was a problem hiding this comment.
Add vLLM minimum version when dropping CLI override
Removing the vllm console-script entrypoint makes vllm ... --omni depend entirely on whatever vllm executable is already installed, but this package still does not declare a vllm dependency/version floor in its install metadata. In environments that have an older vLLM (or no vLLM upgrade), users will now hit missing/unknown --omni behavior after upgrading vllm-omni, which is a regression from the previous interception model; please enforce a minimum compatible vLLM version (or fail fast with a clear runtime check).
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Not a really new problem, but I think we can pin vLLM in requirements now if we want 🙂 maybe would be nice to do that and update the instructions so that it can just be installed directly
Signed-off-by: Alex Brooks <albrooks@redhat.com>
Signed-off-by: Alex Brooks <albrooks@redhat.com>
9cf03d0 to
70dfcb4
Compare
| If you do not need to modify source code of vLLM, you can directly install the stable 0.20.0 release version of the library | ||
|
|
||
| ```bash | ||
| uv pip install vllm==0.19.0+rocm700 --extra-index-url https://wheels.vllm.ai/rocm/0.19.0/rocm700 |
There was a problem hiding this comment.
Looks like the rocm links are actually wrong, should be https://wheels.vllm.ai/rocm/0.19.0/rocm721 (since 0.18 was the last one 700 was prebuilt), so changed it here
Signed-off-by: Alex Brooks <albrooks@redhat.com>
Signed-off-by: Alex Brooks <albrooks@redhat.com>
| ServingPooling( | ||
| engine_client, | ||
| state.openai_serving_models, | ||
| state.openai_serving_render, |
There was a problem hiding this comment.
I think this is unused and that it is being silently absorbed by *args in the superclass constructors, so removing it for clarity
Signed-off-by: Alex Brooks <albrooks@redhat.com>
Signed-off-by: Alex Brooks <albrooks@redhat.com>
26d3b57 to
913ef90
Compare
Signed-off-by: Alex Brooks <albrooks@redhat.com>
Signed-off-by: Alex Brooks <albrooks@redhat.com>
…ct#3082) Signed-off-by: Alex Brooks <albrooks@redhat.com> Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com>
…ct#3082) Signed-off-by: Alex Brooks <albrooks@redhat.com> Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com>
…ct#3082) Signed-off-by: Alex Brooks <albrooks@redhat.com> Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com> Signed-off-by: sphinxkkkbc <binchengkang8@gmail.com>
…ct#3082) Signed-off-by: Alex Brooks <albrooks@redhat.com> Co-authored-by: SYLAR <125541396+lishunyang12@users.noreply.github.com>
Purpose
Removes entrypoint hijacking for vLLM to prevent installation issues between vLLM / vLLM Omni, e.g., vLLM Omni uninstalling also uninstalling the entrypoint for vLLM, vLLM Omni needing to be installed after vLLM for interception to work correctly, etc.
Should not be merged before vllm-project/vllm#40744, otherwise it'll break
vllm <command> --omni(thoughvllm-omniwill still work)This should also fix issues like #2850
CC @redwhitecat @hsliuustc0106 @lishunyang12