Ollama allow user to specify Context length for models

[cruzanstx]

Whenever I use a model via ollama it defaults to the context length of the model, there are plenty of times where I want to use way less than that size for my tasks. Also other model providers you can specify the context length why not ollama?

[yuisheaven]

I agree.
This is crucial for being able to reliably use Ollama.
Just as an example:
When using the qwen3-coder-30b-a3b model, this would usually fit into one GPU when quantized and when using up to lets say 16-32k tokens contextLength.
But as the model supports up to 256k, Kilo Code will try to load it with that and this will completely crash any normal user setup as most consumer pcs don't have enough vram/ram to load it anymore.