3 ms·

Context window depends on the model used. I used gemma4 I think and set it to 32k so startup is faster. I think even gemma supports up to 200k

• drbscl 2mo ago
  Yeah, I'm aware, I just think it's an unusual default for a harness to use
  • refsab 2mo ago
    Should be more? I envisioned the user of this as kind of local koboldcpp or lmstudio users with limited resources
    • drbscl 2mo ago
      I believe they report the configured window, right?
      • refsab 2mo ago
        yes they do. I thought you meant I should have a larger default. When you configure any model the context size that's reported is detectable