[New Model]:Can you support the VLA series models? For example, openVLA.

[psong123]

The model to consider.

https://huggingface.co/openvla/openvla-7b

The vllm documentation does not support the VLA series large models. Can you support the VLA series models?

The closest model vllm already supports.

VLM series models

What's your difficulty of supporting the model you want?

No response

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.

[PieterBecking]

Would be awesome!

[zmf2022]

please accelerate vla models, such as pi0 or gr00tN1.5

[Komershan]

Hi! I have a vacation in a week, so I would actually like to try contributing on this issue, also because I have an interest in VLA models. Unfortunately, I don't have much experience in this, so I would check similar issues (for VLM models I guess), an then try to contribute.

[Komershan]

@DarkLight1337 Is it okay if I try to resolve this issue?

[DarkLight1337]

Sure!

[DavidHe137]

@Komershan Any updates on this? I'd like to give it a shot

[Komershan]

@DavidHe137 Hey! Thanks for jumping in on this issue — I actually started working on integrating OpenVLA a while back, but got caught up with a big work project and had to pause.
I’d still love to help if you’re open to collaborating — maybe we can split the work between different VLA models.
I’d be happy to get in touch and maybe set up a bit of joint work if that’s possible!

[yongming-qin]

I'm also working on integrating VLA models into vLLM. From my testing, not all architectures fit well — models like OpenVLA, RT-2-X, and RoboCat can be adapted as pure decoder-style LLMs, but Pi-Zero and GR00T N1 rely on diffusion and aren’t ideal for vLLM. Even OpenVLA needs some extra handling for proper integration.

I’m midway through implementing OpenVLA and will share updates once I have a preliminary result.