Is there a way to add classifier free guidance (CFG) to vllm while maintaining super fast inference?

[salahzoubi]

I've seen there's recently been a PR that incorporates classifier-free guidance to HF model.generate function and it's super useful for generation. Is there a way to incorporate this in to the vllm sampling mechanism?

[LouisCastricato]

Following up on this

[unyqhz]

Following up on this

[Vermeille]

What would it take to implement it?

[Vermeille]

ping

[github-actions]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!