[Bug]: Can't call numpy() on Tensor that requires grad

[TroySeins]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What happened?

When training an embedding, the moment something wants to get saved this error occurs and the embedding is broken.

Steps to reproduce the problem

1. Go to .... Training
2. Press .... Train Embedding
3. ... Wait for this error after the default setting of 500 steps.

What should have happened?

The embedding or the intermediate step should have saved.

Commit where the problem happens

804d9fb..ac08562

What platforms do you use to access UI ?

MacOS

What browsers do you use to access the UI ?

Apple Safari

Command Line Arguments

my run_webui_mac.sh
===================

#!/usr/bin/env bash -l

# This should not be needed since it's configured during installation, but might as well have it here.
conda env config vars set PYTORCH_ENABLE_MPS_FALLBACK=1

# Activate conda environment
conda activate web-ui

# Pull the latest changes from the repo
git pull --rebase

# Run the web ui
python webui.py --precision full --no-half --use-cpu Interrogate GFPGAN CodeFormer ESRGAN SCUNet --disable-safe-unpickle $@

# Deactivate conda environment
conda deactivate

Additional information, context and logs

No response

[rafaelmotac]

Same thing here. Every time I start an Embedding Training, after finish or interrupted the same error appears and the .pt file is broken.

OS: Mac OS Arch: aarch64 Device: Macbook Pro M1 Max 32GB RAM

[julianko13]

Update torch to 1.13 should solve this.