[BUG] Minimum model size is 4GB

[Evanev7]

Describe the bug

In the dashboard, the minimum model size is 4GB. This makes it impossible to attempt to load smaller models on devices 4GB of memory, and often makes it impractical to load them on medium (8-16GB) devices.

To Reproduce

Check the dashboard

Expected behavior

The model sizes for, say, Qwen3-0.6B-4bit should be ~0.3GB

Actual behavior

The model sizes for, say, Qwen3-0.6B-4bit are 4GB in the dashboard

Environment

uv run exo on main