[Feature]: Supporting Guided Decoding via AsyncLLMEngine

[DhruvaBansal00]

🚀 The feature, motivation and pitch

Today, Guided Decoding can only be accessed if we use the OpenAI compatible API interface, since that is the only way of passing in a GuidedDecodingRequest object into the generation query.

It should theoretically also be possible to support this via the AsyncLLMEngine.generate() interface. Anything blocking from supporting this today? I don't see it being possible to pass in a GuidedDecodingRequest into the generate method today.

Alternatives

No response

Additional context

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.

[njhill]

@joerunde

[joerunde]

@DhruvaBansal00 mind taking a look at https://github.com/vllm-project/vllm/pull/8252/files#r1747793503?