The weights for Kimi K3 have been released!
And we've got a Public Endpoint where you can get started!
If you build anything with it, let us know in the # built-on-runpod channel on Discord: discord.com/invite/runpod
Day 2 at @Ai4Conferences, and someone spent 25 minutes on her feet and built our GPU out of LEGO!
Definitely faster than most Dockerfile setups we've seen 😂 .
PS: If you're in Vegas, come say hi tomorrow!
Your models run. You rest.
If you're at @Ai4Conferences in Las Vegas this week, come check out our relaxation station.
Because 3 days on a conference floor is a lot.
Wan 2.2 is the most capable open-source video model out right now.
We wrote the guide to running it on Runpod with ComfyUI – text-to-video, image-to-video, LoRA support, all covered:
Thinking about fine-tuning DeepSeek V3? Even with LoRA it needs ~24x H100s.
Before you spin up a cluster, it's worth asking whether a distilled 7B to 70B variant gets you there on one GPU.
We wrote up both paths to help you choose: