-
Notifications
You must be signed in to change notification settings - Fork 2.1k
Add vLLM transformers backend to online methods #3773
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
@qgallouedec @kashif this works, apparently my compiled dev version for vLLM was causing all the issue, updating v1 solved it! |
@@ -393,6 +393,14 @@ class GRPOConfig(TrainingArguments): | |||
"contention with training." | |||
}, | |||
) | |||
vllm_model_impl: str = field( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
can you also add these doc in the docstrings further up in the file?
@@ -292,6 +292,14 @@ class ScriptArguments: | |||
"'trace'." | |||
}, | |||
) | |||
vllm_model_impl: str = field( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It's missing in the docstring above
@@ -164,6 +164,14 @@ class may differ from those in [`~transformers.TrainingArguments`]. | |||
"(`pip install vllm`)." | |||
}, | |||
) | |||
vllm_model_impl: str = field( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
same
docs/source/vllm_integration.md
Outdated
|
||
## vLLM with Transformers Backend | ||
|
||
vLLM now supports transformers backend for model implementations. Simply passing in `transformers` in `vllm_model_impl` in configurations or through argument parser will set use transformers backend. See an example below. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Does it support VLMs? Limitations?
Additionally, maybe linking the blog: https://blog.vllm.ai/2025/04/11/transformers-backend.html
(in case these ideas can be added in 1-2 sentences)
@kashif can you merge 🙏🏻 |
Add vLLM transformers backend to online methods GRPO and online DPO.
This has two limitations:
VLLM_USE_PRECOMPILED=1 uv pip install -e .
I will check the issues, you can test with
while serving with