preferences that are model related need to be stored with the selected model #35
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
parameters that are related to the model behaviour need to be stored per model and thinking effortand rehydrated on model switch, so that for example activation of ssd streaming and context limit and other settings are aligned with the selected model. Deepseek V4 Flash will have very different settings from GLM 5.2, but also Flash with no thinking and Flash with max thinking will differ.
So the panel for those settings that are token generation related need to have a selection at the top with the model and thinking effort and then present the values set up for that combination. the default should be the currently selected model and the default thinking mode for that model.
some of tjhe generation parameters are outside that, for example the system prompt is a global setting and should move to its own section for prompt related stuff, and the default thinking mode belongs to the model selection in the top.
"Acceleration & memory" is also model dependent, but not thinking dependent. this should have a model selection at the top and store the values per model, and the model selection should default to the model selected in "Model & lifecycle".
The goal is to have the model selection and thinking mode selection in preferences (and also in chat window - thinking can be switched there!) to load the correct preferences and use them in generation. So switching thinking could for example use different context sizes and temperature.
implemented