Is there any model I can run on single Spark to generate music?

Just for fun.. is there any model capable of creating music? Any experience?

I’ve used Ace-step. You’ll have plenty of vram to run the full xl version. Make sure you use the 5hz-lm model with it if you want lyrics ACE-Step/acestep-v15-xl-base at main ACE-Step/acestep-5Hz-lm-4B · Hugging Face

I’ve used Ace V1.5 as well. It does a fine job but it isn’t up to Suno standards. It is best if you create your own lyrics.

Yep, I’ve been using Ace-step 1.5 for a while now, albeit on an RTX 6000. Should run just fine on a Spark, too.

Ace-steps likely the best open solution right now for self hosted music generation, some of the songs are great, some are total slop, it’s a slot machine. I made a self hosted webUI for Ace-step with a radio function, here’s the repo if you wanna play with it: GitHub - DoctorMasterNewb/conjure: AI music self host · GitHub

I dug into this a little bit and Ace-Steps really is great. Here is my first post, and write up on it, if it helps at all. ACE-Step v1.5 on DGX Spark — install recipe, measured numbers, and the scheduling finding I'd have wanted before I started