Spaces:
Running
[REQUEST] Remove Unnecessary constraints on assistant model parameters
In the assistant settings, parameters such as temperature can be set, but with unnecessary value restrictions.
Following are written assuming Command R+.
- The minimum value for temperature is 0, but it does not accept values less than 0.1.
- temperature setting is unnecessarily limited to a 0.1-step unit by the UI. I think this is Rough.
- The maximum value of Top P is 0.999... (the maximum value less than 1), but Assistant accepts a setting of 1.0. In this case, an validation error occurs when the model is executed.
- Top P setting is unnecessarily limited to a 0.05-step unit by the UI.
- Top K has a maximum value of 500, but does not accept more than 100.
- Top K setting is unnecessarily limited to a 5-step unit by the UI.
I want to try to see what happens, especially if I increase the Top K value.
It think very happy and desirable to be able to set the configuration values that the model supports.
Also, the repetition penalty accepts values between 0 and 2 currently and only increments by 0.1 at a time, but according to NVIDIA:
"A value of 1 means that there is no penalty and values larger than 1 discourage repeated tokens."
And experimentally, anything less than 1 seems to just output something like this:
SECTION#1 - Revision - Write the entire new/optimized prompt that I just worked on:
"Your goal is to create a prompt for a TEXT TO IMAGE (2) AI models that can be input into the AI to create a unique and detailed output. The prompt must be clear, concise, and comprehensive, and must provide the AI with all the necessary instructions to create the desired output. The output must be a unique and detailed 2) that is generated by the AI, and must not be a 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a 2) of an 2) that is a
Which I don't think does anybody any good xD
I have another request.
I'd like to be able to specify these model parameters in the regular model the same way I can with the system prompt, without having to go through the assistant. Even better, I should be able to change them in the middle of a conversation.
The default temperature of 0.3 seems a bit low for stories and other creative writing. I was surprised to see a big increase in creativity when set to 1.0+. It would be great if the model could be used to its full potential without an assistant. :-)
I should be able to change them in the middle of a conversation.
You should be able to say thank you for the free technologicool transformer bots