Aug 14th is the next Build & Learn stream on Music generation, but I got already got ACE-Step 1.5 setup in a Flyte workflow over the weekend. Last time I compared any other OSS models it was by far the best, but i'm going to try several others out over the next couple weeks! github.com/ace-step/ACE-Step Let me know if you have any song prompts you'd like me to try on it!

Will try it out this week! Im working on some music for a game and would be great to experiment with this and other OS models in the workflow.

Look at their license too, if i recall it was open to use in applications like that but def double check.

🔬

Yeah I'm just using Flyte devbox on the single GPU on spark. So it all runs there and I remote in with tailscale.

Maybe still beneficial running the actual GPU time as a local process but triggering tasks and logging all from flyte?

Actually yeah you should be able to do something like that. Just like using Flyte to manage agent runs on a hosted API or locall LLM on Ollama

Diff rhythm and Stable Audio were pretty strong, that was actually some nice ambient

Reminder this Friday we'll be discussing Open Source music generation. You all seem pretty interested in this subject with 90 RSVPs already! I encourage you to check out the models yourself and show off what you learned during the stream. (ACE-Step if a good one) Or give me some music prompts you want me to run!