Why the he** does my cuda stream get disabled?

[UnconnectedBedna]

I feel its better to ask than to make an issue, unless we find out it actually is an issue, and not me screwing something up..

$ ./webui.sh

################################################################
Install script for stable-diffusion + Web UI
Tested on Debian 11 (Bullseye), Fedora 34+ and openSUSE Leap 15.4 or newer.
################################################################

################################################################
Running on <redacted> user
################################################################

################################################################
Repo already cloned, using it as install directory
################################################################

################################################################
python venv already activate or run without venv: /home/<redacted>/AI/stable-diffusion-webui-forge/venv
################################################################

################################################################
Launching launch.py...
################################################################
glibc version is 2.40
Check TCMalloc: libtcmalloc_minimal.so.4
libtcmalloc_minimal.so.4 is linked with libc.so,execute LD_PRELOAD=/usr/lib/libtcmalloc_minimal.so.4
Python 3.10.10 (main, Sep  2 2024, 09:34:27) [GCC 14.2.1 20240805]
Version: f2.0.1v1.10.1-previous-635-gf5330788
Commit hash: f53307881bfd824dbdce6ac0d4bba04d9a74ab36
CUDA 12.1
Faceswaplab : Use GPU requirements
Checking faceswaplab requirements
Install protobuf>=3.20.2
Installing sd-webui-faceswaplab requirement: protobuf>=3.20.2
1.211446364000949
Launching Web UI with arguments: --xformers --cuda-malloc --always-high-vram --cuda-stream --pin-shared-memory
Using cudaMallocAsync backend.
Total VRAM 24150 MB, total RAM 31195 MB
pytorch version: 2.3.1+cu121
xformers version: 0.0.27
Set vram state to: HIGH_VRAM
Always pin shared GPU memory
Device: cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync
VAE dtype preferences: [torch.bfloat16, torch.float32] -> torch.bfloat16
CUDA Using Stream: True
Using xformers cross attention
Using xformers attention for VAE
ControlNet preprocessor location: /home/<redacted>/AI/stable-diffusion-webui-forge/models/ControlNetPreprocessor
CHv1.8.13: Get Custom Model Folder
15:59:49 - ReActor - STATUS - Running v0.7.1-b2 on Device: CUDA
Loading additional modules ... done.
CHv1.8.13: Set Proxy: 
2025-01-21 15:59:58,695 - ControlNet - INFO - ControlNet UI callback registered.
Model selected: {'checkpoint_info': {'filename': '/home/<redacted>/AI/stable-diffusion-webui-forge/models/Stable-diffusion/XL/juggernautXL_juggXIByRundiffusion.safetensors', 'hash': 'ed515f41'}, 'additional_modules': ['/home/<redacted>/AI/stable-diffusion-webui-forge/models/VAE/sdxl_vae.safetensors'], 'unet_storage_dtype': None}
Using online LoRAs in FP16: False
Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 20.2s (prepare environment: 3.4s, launcher: 0.3s, import torch: 3.3s, load scripts: 1.5s, initialize google blockly: 8.5s, create ui: 1.9s, gradio launch: 1.3s).
Environment vars changed: {'stream': False, 'inference_memory': 1020.0, 'pin_shared_memory': False}

They get changed somewhere the last second:
Environment vars changed: {'stream': False, 'inference_memory': 1020.0, 'pin_shared_memory': False}

Is pytorch version: 2.3.1+cu121 to old or smthn?

Also, reactor...
I have not done git pull since I realized it was removed from github, can I without screwing something up on that side?
Or should I stash stuff first?

Edit
Lol, when you make a post, you naturally finally figure out what it is... The settings go back to Queue and CPU in the gui, I just did not realize when I was loading into an sdxl model..
But this creates more questions. How I set that in the settings? I can't find it anywhere.
And if I change it in the gui:

Environment vars changed: {'stream': True, 'inference_memory': 1650.0, 'pin_shared_memory': True}
[GPU Setting] You will use 93.17% GPU memory (22500.00 MB) to load weights, and use 6.83% GPU memory (1650.00 MB) to do matrix computation.

And then go back to sdxl or sd:

Environment vars changed: {'stream': False, 'inference_memory': 1650.0, 'pin_shared_memory': False}
[GPU Setting] You will use 93.17% GPU memory (22500.00 MB) to load weights, and use 6.83% GPU memory (1650.00 MB) to do matrix computation.
Environment vars changed: {'stream': False, 'inference_memory': 1650.0, 'pin_shared_memory': False}
[GPU Setting] You will use 93.17% GPU memory (22500.00 MB) to load weights, and use 6.83% GPU memory (1650.00 MB) to do matrix computation.

Two times, witch is also odd...
It also does not seem to save the env settings if I shut down with them active and load back into a flux model.

More questions:
Can I set "Never OOM integrated" to be enabled by default? (both settings)

I guess all of this boils down to QOL questions, can I fix configure this?