0 tokens output on slight or NSFW stuff
Hello, I was experimenting with mythomax on ooba with exlamma with different prompt templates with the alpaca-input template. And i got this wierd bug where for certain rp inputs like hugs you or touches you doesn't work. Sometimes it works when l load the model again but even then it stops working after a few tries. It outputs 0 tokens. When i try to do normal conversations, it just works again.
this aint no issue from the GPTQ files itself, go ahead and file an issue on here instead
I have a question related to this...
i was going to ask if anyone knows if this model recognises and uses emojis and the same for actions denoted by action or π’π€π΅πͺπ°π―, and internal dialogue/thoughts with something like ("OMG HE TOUCHED ME!")
I've seen various formatting for actions and thoughts on those chatbot apps and websites. Seems typing something encased with asterisks basically always denotes action across several different bots and i was hoping this LLM would have the same capabilities. I'm not that bothered about thoughts as for me thats too unrealistic, hearing a third party's thoughts. But if Mytho can do emojis and/or actions that would be great!
this is a quantized version of mythomax... don't expect anything great from it
Yeh of course, there is that limitation. But im pleased to report I did get the model to use action before.
do you know the size of the recognised context window? i want to pack it with as much as possible
4096
isnt that the max seq length of the chat? or does a huge context window reduce the chat 'memory' as it were? Sorry, im a noob.
this is a quantized version of mythomax... don't expect anything great from it
Why not? What makes mythomax so special?