Open morgan55555 opened 2 weeks ago
just specify "-1", it will be adjusted to "-1m" which is keep loaded forever
behavior on Ollama API
Nope, it will change back to 0.
"keep_alive": ("INT", {"default": 5, "min": 0, "max": 60, "step": 5}),
Your are right, ComfyUI frontend will reset it back, missed that.
But when using ComfyUI throw API endpoints it will work(specifying -1
)
PR to fix this will be very simple, just change "min": 0
to "min": -1
and this will fix it for non-API ComfyUI mode.
i have add the support for this, please try.
Own ollama server is working on single llm model without unloading. Running comfyui-ollama with keep_alive will cause server to unload a model every time. We need ability to skip this loads/unloads for better performance.