• Save checkpoint every N steps: 0
    • Generate preview every N steps: 0
    • ON "Don't cache latents"
    • ON "Train Text Encoder"
    • ON "Use 8bit Adam", note on this
    • ON "Gradient Checkpointing"
    • "Mixed Precision: fp16"
    • 8bit adam doesn't appear to help unless you also add the TORCH_COMMAND line into your launch bat as per the github page:
⎗
✓
1
set TORCH_COMMAND="pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 --extra-index-url https://download.pytorch.org/whl/cu116"
⎗
✓
1
* Runs at about 1.7it/s with a 9700k., 50% CPU usage, full VRAM usage, approx 10-40% GPU, it goes up and down.
Edit
Raw PDF PNG
Pub: 12 Nov 2022 23:47 UTC
Views: 1037
new·what·how·langs·contacts