[bug]: “InvokeAI 6.4 and above shuts down after generating about 250 images.” #8543
Description
Is there an existing issue for this problem?
- I have searched the existing issues
Install method
Invoke's Launcher
Operating system
macOS
GPU vendor
Apple Silicon (MPS)
GPU model
Mac Studio M2 Max
GPU VRAM
64
Version number
6.5
Browser
No response
System Information
No response
What happened
After generating approximately 200–300 images (either via the Web UI or Desktop app), InvokeAI suddenly crashes without error dialogs, returning: Process exited with code: 0
In some cases, the following stack trace appears:
BrokenPipeError: [Errno 32] Broken pipe
...
self.sp = self.status_printer(self.fp)
...
File "/invokeai/backend/stable_diffusion/diffusers_pipeline.py", line 395, in latents_from_embeddings
for i, t in enumerate(self.progress_bar(timesteps)):
Additionally, in the UI, this exception occurs:
{
"name": "TypeError",
"message": "e.map is not a function",
"stack": "... App-BAHkxqK1.js"
}
• Crash does not correlate with a specific prompt or model.
• The issue occurs regardless of the frontend used (desktop app or Web UI).
• The issue already existed in version 6.4 and is still present.
• Memory usage is monitored and stable before crash.
• Logs indicate full VRAM use (VRAM: 100%) just before termination.
• Happened both with the --max_loaded_models=1 setting and without.
• Reproducible after consistent long runs, regardless of whether using scheduler queues or direct prompt
• Model cache logs show MPS device memory being filled progressively:
VRAM: 100% just before crash
What you expected to happen
not to crash :-)
How to reproduce the problem
Steps to Reproduce
1. Launch InvokeAI (UI or Web)
2. Queue or manually trigger 200+ image generations in a session
3. Wait for process crash or silent termination
Additional context
• The system is stable, has plenty of resources, and the issue is not linked to any specific model or prompt.
• No memory leak or OOM errors are observed.
• Precision set to float32 for compatibility with MPS but the bug is still here
Discord username
No response