[Misc] Remove deprecated args in v0.10 #21349
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request removes two deprecated command-line arguments, --device and --use-v2-block-manager, from vllm/engine/arg_utils.py. The changes are straightforward and correctly implement the stated goal of cleaning up unused arguments for the v0.10 release. The removed arguments were already marked as deprecated and their removal appears safe. I don't see any issues with this change.
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
github-actions
bot
added
the
ready
Head branch was pushed to by a user without write access
kebe7jun
force-pushed
the
fix/remove-deprecated-args-in-v0.10
branch
from
a46b584 to
db75c11
Compare
mergify
bot
added
documentation
|
Please fix Lint and Deploy Charts CI |
Signed-off-by: Kebe <[email protected]>
kebe7jun
force-pushed
the
fix/remove-deprecated-args-in-v0.10
branch
from
db75c11 to
82bd90a
Compare
Fixed. |
| @@ -364,7 +363,6 @@ class EngineArgs: | |||
| max_prompt_adapter_token: int = \ | |||
| PromptAdapterConfig.max_prompt_adapter_token | |||
|
|
|||
| device: Device = DeviceConfig.device | |||
There was a problem hiding this comment.
Are we sure this is not longer supported?
Now the upstream caller still pass in the param, and will cause an error
lm_eval --model vllm --model_args "pretrained=Qwen/Qwen3-30B-A3B-FP8,max_model_len=32768,enforce_eager=True" --trust_remote_code --tasks gsm8k --num_fewshot 5 --batch_size auto File "/home/wentao/.wentao_env/lib/python3.12/site-packages/lm_eval/models/vllm_causallms.py", line 177, in __init__
self.model = LLM(**self.model_args)
^^^^^^^^^^^^^^^^^^^^^^
File "/home/wentao/vllm-source/vllm/entrypoints/llm.py", line 244, in __init__
engine_args = EngineArgs(
^^^^^^^^^^^
TypeError: EngineArgs.__init__() got an unexpected keyword argument 'device'
There was a problem hiding this comment.
Also, if we are removing this, should we also have removed the Device config from config.py?
There was a problem hiding this comment.
@yewentao256 See #18301 (comment), the device parameter will not take effect even if it is set. It is a useless parameter. If you want to specify the device, you can specify it through the CUDA_VISIABLE_DEVICES environment variable. If it is cpu mode, you need to recompile it to cpu mode.
There was a problem hiding this comment.
@hmellor Thanks for your reminder, I will take a closer look at how to safely remove DeviceConfig.device, which requires another PR.
There was a problem hiding this comment.
@yewentao256 See #18301 (comment), the
deviceparameter will not take effect even if it is set. It is a useless parameter. If you want to specify the device, you can specify it through theCUDA_VISIABLE_DEVICESenvironment variable. If it is cpu mode, you need to recompile it to cpu mode.
Make sense, thanks!
Device has been a deprecated arg for a few releases of vLLM and is now removed in 0.10.0 vllm-project/vllm#21349
These Args are no longer used and will be removed in the v0.10 release.