InvokeAI
InvokeAI copied to clipboard
invoke-ai
[bug]: Gets randomly stuck at "Generating Running VAE Decoder"
Is there an existing issue for this problem?
- [x] I have searched the existing issues
Operating system
Windows
GPU vendor
Nvidia (CUDA)
GPU model
RTX 4090
GPU VRAM
24GB
Version number
5.6.0
Browser
Microsoft Edge 131.0.2903.146
Python dependencies
accelerate 1.0.1 compel 2.0.2 cuda 12.4 diffusers 0.31.0 numpy 1.26.3 opencv 4.9.0.80 onnx 1.16.1 pillow 10.2.0 python 3.11.11 torch 2.4.1+cu124 torchvision 0.19.1+cu124 transformers 4.46.3 xformers Not Installed
What happened
Invoke will occassionally get stuck at "Generating Running VAE Decoder" when attempting to generate. GPU-load will go to 100% when this happens and I need to restart Invoke to break out of it.
This started occuring immediately after upgrading from 5.5.0 to 5.6.0. So far it feels like this only happens after changing model/checkpoint although I haven't confirmed this.
What you expected to happen
For generating to work without getting stuck
How to reproduce the problem
No confirmed steps to reproduce, although it feels like frequent model/checkpoint changes increases the likelihood of problem occuring
Additional context
No response
Discord username
No response
Having the same exact issue to a literal t. Tried enabling the low vram parameters in yaml file and it didn't fix anything. EDIT: Thought it was maybe an issue with the models I was using. Same issues with very popular stable diffusion models like Juggernaut XL
Also having this same exact issue. VAE decoder stalls after running several prompts. Have to shutdown Invoke and reopen.
i have this issue too, the issue seems to be with the model caching system, this problem comes in when you keep switching models or loras, i notice that the vram usage keep going up when the models are switched, then its the memory limit. The application is not actually getting stuck its that its running out of VRAM and running the VAE from the CPU ram which causes it to be very slow for us this seems like it is stuck. If you leave it for a minute, it will cause the image to be complete. But instead i just prefer to close the app which causes the cache to be cleaned and restarting with a fresh slate.
