vamx
patreon
New Chat AI Models + Auto Switching
🕑 Added 2024-02-06 11:53:20 +0000 UTCExisting Users: You don't need to update vamX.
New Users: Download vamX for AI Chat 1.34.
You must be subscribed to get new passwords.
Get your new password from: https://www.patreon.com/vamx/membership
Improved / New Chat AI Models
The Chat AI Model is the AI that writes the response. Different Chat AIs produce different responses. Chat AIs are known as LLMs, Large Language Models, or just "models" for short.
For a while our default Chat AI has been NSFW Ooba, but this model wasn't performing well compared with more recent models, and we have removed it.
Instead of NSFW Ooba, we've added 3 new NSFW Chat AIs:
Silicon Maid 7B - A very fast model, creative, and made for role play.
Ooba 20B - Good at improvising without you feeding it too much detail, works for SFW & NSFW. Can return more complex responses in some situations.
Ooba 13B - Psyfighter2, by far the most popular NSFW KoboldAI model, now faster (then from KoboldAI) since we serve it directly.
These new models mean there is a much better chance you can get the response you are looking for from the AI, by choosing the appropriate model.
Also, we've added additional servers for the new models, so there should be an overall increase in speed as the servers are less likely to be maxed out (except for the first day after this announcement, which is always very busy).
Automatically Changing Chat Models
The default Chat AI model is now:
NSFW Random AI (Changes on Retry Last)
When set to this new default, the chat model switches every time you press Retry Last. There isn't yet a way to see which model is being served when set to Random AI.
If you are pressing Retry Last, you probably didn't like the response, so Retry Last changes to a new model, and regenerates the response (if you are set to NSFW Random AI).
Comments
vamX
Yeah, something broke in the US audio generation part of the server. Should be fixed within about 6 hours. Often times the first few days after a new server update are a little rocky, with one of the servers having issues. The EU server should work fine for you meanwhile, and this should be fixed soon.
Curious George
I've noticed when using the US servers the AI will display the text okay but there is no audio. I tried changing modes and voices, but still nothing. When i us one of the European servers, then the audio is there. Any idea what could be causing this issue? I live smack dab in the middle of the US, pretty much.
vamX
We've have never been able to reproduce this issue. Also switching out of elevenlabs voices then back to them shouldn't have any effect (based on the code). Each time we send a request for a voice to elevenlabs it is sent separately without any long term connection. Have you tried simply pausing for a minute then asking again? Perhaps elevenlabs blocks occasionally, or doesn't return results when requested? It's very strange. We've tried to solve this multiple times, but simply have never even been able to see the issue.
Tro
Any idea about the Elevenlabs voices issue, where they will say a sentence or two and then just stop? I know about the workaround, where you change to a non-elevenlabs voice and then back again, but that's not easy to do when you have to scroll the list in VR.
