>>105851413
>Is there an easy way to split a model between multiple GPUs?
Yeah.
Ask >>>/g/lmg for details but I'd expect vllm to literally just fucking work if it can recognize your CUDA GPUs.

A-anon? You haven't been showing up to the races recently, is something wrong?

Gacha? Umamusume? What are you talking about? Come on, I'm finally gonna beat Jovial Merryment.