Kanana serving docs: vLLM version difference for RoPE/YaRN config

by seongsubae - opened Jan 15

Jan 15

Hi Kanana Team,

Thanks for the Kanana model upgrade release, and for sharing the serving/deployment guide promptly!

I noticed that long-context RoPE/YaRN configuration depends on the vLLM version. For example, on vLLM v0.13.0 (and generally since v0.11.1), vllm serve no longer accepts --rope-scaling (“unrecognized arguments”), so users need an alternative approach (e.g., --hf-overrides or config-based overrides). If the docs could mention this version difference (or provide an up-to-date recommended command), it would make deployment smoother when upgrading.

Best,
Seongsu

ryan-u

Kakao Corp. org Jan 15

Hi Seongsu,

Thank you so much for your interest in the Kanana model and for taking the time to review our deployment guide.
We really appreciate you pointing out the compatibility issue with newer vLLM versions. Based on your suggestion, we have updated the documentation to use the --hf-overrides approach instead of the deprecated argument. This should ensure a much smoother deployment process for everyone upgrading to the latest versions.
Thanks again for your valuable contribution!

Best regards,
The Kanana Team

ryan-u changed discussion status to closed Jan 15

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment