[REQUEST] - 256k context limits.

[SabinStargem]

Due to the incoming Turbo Quant caching and models having larger native limits, I think it would be about time to add extra notches onto the context slider. Qwen3.5 can allegedly support up to 262,144 context.

[Thoron77]

There is a workaround for that. Setup your model as usual, save the preset into kcpps file (its just a json). Edit it and change 'contextsize' value to whatever context you'd like. Then in the command line run .\koboldcpp-launcher.exe --config path_to_kcpps_file, this way you can have much bigger context. The drawback is the command line part. Alternatively you can run .\koboldcpp-launcher.exe with all the parameters setting up the model manually but it can get complicated really fast, the kcpps method is much easier. Also if you load kcpps into the launcher GUI it will complain about the context size, so just remember to reedit the preset after making changes in the GUI.
Eventually when you do not wish to unpack the Kobold app I think it works the same with the packed version, just instead of .\koboldcpp-launcher.exe use .\koboldcpp.exe.

[LostRuins]

i'll raise the limits in the gui next version