While it's somewhat limited for a #LLM, Granite3.1 MoE 3B Q8 runs pretty well on the #RaspberryPi 5.

Going to try using this with some #n8n pipelines.

I've been experimenting with blending some of the Koroko text to voice models in an attempt to improve the voice quality.

Not OpenAI or Eleven Labs quality, but I think it's pretty decent for a local model.

Let me know your thoughts.

youtu.be/nKMHIsINScg?...

Making progress on my #RaspberryPi mini-rack.