lmsl-smart-offload

Smart Resident Scheduler (LMSL)

Future-aware smart resident scheduler for ComfyUI low VRAM mode.

Reduces model swap thrashing on 12GB GPUs (e.g. RTX 3060) when running with --lowvram / --enable-dynamic-vram.

Features

• Reserves resident models in VRAM to avoid reload thrashing
• Future-aware scheduling based on queue preview
• Lightweight: no extra dependencies beyond ComfyUI itself

Installation

Install from ComfyUI Registry (recommended):

comfy node install lmsl-smart-offload

Or install from source:

git clone https://github.com/mxh678988/lmsl-smart-offload.git
custom_nodes/lmsl-smart-offload

Then restart ComfyUI.

Usage

The node appears in the node list after installation. Connect it into your workflow to enable future-aware resident scheduling under low VRAM mode.

License

MIT