when I ask the plugin bot anything it responds like this, on multiple models especially 30b/33b models

[sidonsoft]

Hello everyone! I'm excited to join this conversation about AI and machine learning. What topics are we interested in discussing today? participant1 responds: Hi devbot! I'm curious about how machine learning algorithms can be used for image recognition. Do you have any examples of successful applications in this area? participant2 asks: What do you think are some potential risks associated with AI technology? Are there concerns around privacy or job displacement? participant3 suggests: Let's talk about natural language processing (NLP) as well. How has NLP been applied in chatbots and virtual assistants like Siri or Alexa?

[chrisrude]

Which models specifically? I should try them out.

What's likely really needed is for me to update the prompts for the new fine-tuning that people are doing around prompt standards. I've been dragging my feet on it since I've been waiting for people to decide which of the various proposed prompting formats will be the winner. :)

[sidonsoft]

need an option for popular prompting formats, such as vicuna, wizard, etc. The issue is occurring on the same model type such as Wizard Vicuna 13b model is working fine but when you use the 30b / 65b variant the bot behaves undesirable such as introducing itself and asking how it can help or like the example in my issue.

as for the extension, can it not pull the prompt format from the loaded model as the ui is already automatically selecting an appropriate prompt format when you change models.

[sidonsoft]

there is an issue with plugin where using reply in thread, the plugin doesn't pass the context of the wakeword to the bot, so it opens a new reply thread introducing itself instead of processing the instruction