91–99 of 99 posts
Earlier quoted context omitted.
So the history prompts are collections of text/audio pairs?
history is semantic, coarse and fine. so essentially the same thing thats getting generated just using it as an input before the generation
How do I save the audio array to a file on the file system?
from scipy.io.wavfile import write as write_wav
write_wav("c:\\wherever\\whatever.wav", SAMPLE_RATE, audio_array)
from playsound import playsound
playsound("c:\\wherever\\whatever.wav")
Earlier quoted context omitted.
It takes about the same resources to run as Stable Diffusion or LLaMA. You don't even need a GPU.
No, the power and money to run high quality servers like ChatGPT4 and MidJourney or DallE. Sure there's local-able alternatives, but they're lower quality and lower bandwidth and not being used as a business proposition. Just like a private individual can own lock smith tools and play around with locks... don't go basing your business of supplying them wholesale to the general public worldwide for free.
Earlier quoted context omitted.
No, the power and money to run high quality servers like ChatGPT4 and MidJourney or DallE. Sure there's local-able alternatives, but they're lower quality and lower bandwidth and not being used as a business proposition. Just like a private individual can own lock smith tools and play around with locks... don't go basing your business of supplying them wholesale to the general public worldwide for free.
Local LLMs are in production as the backbone of highly valued businesses. That will increasingly be true with the advent of commercially licensed foundation models currently in training such as StableLM and RedPajamas.
That surprises me. Would you be able to point some out to me?
From what I've used of local LLMs they seem nearly unusable.