Conversation

Notices

1. Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 15:21:20 JST Vivi Nella Verita :Debian_logo:
   Can anyone on FEDi get AMD GPU's to work with Stable Diffusion?
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 15:21:19 JST 

     in reply to
     @verita84 I got it to work, but have to use older pytorch version (1.13.1) since they broke something in newer ones for my card.
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 15:33:45 JST 

     in reply to
     @verita84 What card do you use?
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 15:33:46 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     Got a good link? or instructions? I am aon Debian 12
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 21:37:08 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     Radeon 65000 RTX
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 21:37:08 JST 

     in reply to
     @verita84 Then latest pytorch should work fine.
     https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs#running-natively - doesn't even mention installing ROCm, supposedly because the needed binaries are already packaged and redistributed with pytorch.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 21:41:33 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     wait, this is the docker setup that says no GPU found
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 21:41:33 JST 

     in reply to
     @verita84 Are you making a docker container from scratch or running some unofficial one? It needs the DRI device to be passed through from host.
     docker run -it --device=/dev/kfd --device=/dev/dri --security-opt seccomp=unconfined --group-add video
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 21:43:26 JST 

     in reply to

     • 
     @verita84 Ignore the TORCH_COMMAND lines since they still refer to older ROCm version; newer webui should automatically select the correct one.
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 21:47:41 JST 

     in reply to
     @verita84 Maybe it's using some additional virtualization layer, idk, I hate Docker and haven't used it since 2021. Try running it natively on the host system.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 21:47:42 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     I ran it just like that
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 21:57:32 JST 

     in reply to
     @verita84 Yes, but not in Docker. Here, generated a Tomoko just now.
     00000-2969430554-_lora_tomokoKu…
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 21:57:33 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     Docker is downloading the images now. Did you actually get it working?
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 21:58:12 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     Docker fails with this
      likes this.
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 21:59:04 JST 

     in reply to

     @verita84 Looks like it’s picking a wrong Torch version. Try removing the venv directory, adding export TORCH_COMMAND="pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.4.2" to webui-user.sh and restarting the webui.

   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 22:20:26 JST 

     in reply to
     @verita84 In the stable-diffusion-webui directory you cloned from git.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 22:20:27 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     how do you remove the venv dir? where is it
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 22:21:11 JST 

     in reply to

     • 
     @verita84 Also remove --precision=full --no-half, since 6xxx series cards support fp16 just fine.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 22:33:40 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     
     Followed the instructions to upgrade python...maybe it will work now
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 22:33:40 JST 

     in reply to
     @verita84 It's downloading the CUDA version.
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 22:34:37 JST 

     in reply to
     @verita84 Might be a hidden dir (.venv), check if it's there.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 22:36:50 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     are you supposed to start it with
     
     export TORCH_COMMAND="pip install torch==2.0.1+rocm5.4.2 torchvision==0.15.2+rocm5.4.2 --index-url https://download.pytorch.org/whl/rocm5.4.2";
     
     
     or like the doc says
     
     
     REQS_FILE='requirements.txt' python launch.py --precision full --no-half
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 22:36:50 JST 

     in reply to
     @verita84 You're supposed to start it with ./webui.sh.
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 22:41:43 JST 

     in reply to
     @verita84 Then don't launch it as root, or at least delete the lines 94 to 104 in webui.sh if you really want to run it as root.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 22:41:44 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     in docker? not according to instructions, You get this error:
     
     ERROR: This script must not be launched as root, aborting...
   •  (mint@ryona.agency)'s status on Friday, 15-Sep-2023 22:52:22 JST 

     in reply to

     @verita84 Again? Do rm -rf .venv venv, then try running python3 -m venv venv, source venv/bin/activate and then finally pip install torch==2.0.1+rocm5.4.2 torchvision==0.15.2+rocm5.4.2 --index-url https://download.pytorch.org/whl/rocm5.4.2.

   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Friday, 15-Sep-2023 22:52:23 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     it installs torch 2.0.2 in venv but then this h appens when starting webui:
   •  (mint@ryona.agency)'s status on Saturday, 16-Sep-2023 04:26:36 JST 

     in reply to

     @verita84 What happens when you run python3 and type import torch, then torch.cuda.is_available()?

   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 04:26:37 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 

     @mint

     Same

     ################################################################ Cannot locate TCMalloc (improves CPU memory usage) Python 3.8.16 (default, Jun 12 2023, 18:09:05) [GCC 11.2.0] Version: v1.6.0 Commit hash: 5ef669de080814067961f28357256e8fe27544f4 Traceback (most recent call last): File "launch.py", line 48, in <module> main() File "launch.py", line 39, in main prepare_environment() File "/dockerx/stable-diffusion-webui/modules/launch_utils.py", line 356, in prepare_environment raise RuntimeError( RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 04:33:36 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 

     @mint

     (venv) root@nexus:/dockerx/stable-diffusion-webui# python3 Python 3.8.16 (default, Jun 12 2023, 18:09:05) [GCC 11.2.0] :: Anaconda, Inc. on linux Type "help", "copyright", "credits" or "license" for more information. >>> import torch terminate called after throwing an instance of 'std::bad_alloc' what(): std::bad_alloc Aborted (core dumped)
   •  (mint@ryona.agency)'s status on Saturday, 16-Sep-2023 04:33:36 JST 

     in reply to
     @verita84 Weird. Try updaing Python to 3.10, since that's what webui was tested on.
   •  (mint@ryona.agency)'s status on Saturday, 16-Sep-2023 04:44:06 JST 

     in reply to

     @verita84 What if you add export HSA_OVERRIDE_GFX_VERSION=10.3.0 to webui-user.sh?

   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 04:44:07 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     
     It comes with 3,3.8,3.9
     
     no updates availble for new one
   •  (mint@ryona.agency)'s status on Saturday, 16-Sep-2023 04:52:10 JST 

     in reply to

     @verita84 What Torch version shows up in pip list when running in venv (source venv/bin/activate)?

   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 04:52:11 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 

     @mint

     Cannot locate TCMalloc (improves CPU memory usage) Python 3.8.16 (default, Jun 12 2023, 18:09:05) [GCC 11.2.0] Version: v1.6.0 Commit hash: 5ef669de080814067961f28357256e8fe27544f4 Traceback (most recent call last): File "launch.py", line 48, in <module> main() File "launch.py", line 39, in main prepare_environment() File "/dockerx/stable-diffusion-webui/modules/launch_utils.py", line 356, in prepare_environment raise RuntimeError( RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
   •  (mint@ryona.agency)'s status on Saturday, 16-Sep-2023 04:58:05 JST 

     in reply to
     @verita84 Looks fine, not sure why it wouldn't pick up your card. Can you install rocminfo? It might be already packaged for Debian unless it's still stuck in sid.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 04:58:06 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 

     @mint

     (venv) root@nexus:/dockerx/stable-diffusion-webui# pip list Package Version ------------------- ---------------- certifi 2022.12.7 charset-normalizer 2.1.1 cmake 3.25.0 filelock 3.9.0 idna 3.4 Jinja2 3.1.2 lit 15.0.7 MarkupSafe 2.1.2 mpmath 1.2.1 networkx 3.0 numpy 1.24.1 Pillow 9.3.0 pip 22.0.4 pytorch-triton-rocm 2.0.1 requests 2.28.1 setuptools 56.0.0 sympy 1.11.1 torch 2.0.1+rocm5.4.2 torchvision 0.15.2+rocm5.4.2 typing_extensions 4.4.0 urllib3 1.26.13 WARNING: You are using pip version 22.0.4; however, version 23.2.1 is available. You should consider upgrading via the '/dockerx/stable-diffusion-webui/venv/bin/python3 -m pip install --upgrade pip' command.
   •  (mint@ryona.agency)'s status on Saturday, 16-Sep-2023 05:02:55 JST 

     in reply to

     • 
     @verita84 I'll try to help further tomorrow, since it's getting pretty late.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 13:19:44 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     ok it was working fine, GPU was making lots of image. Then I decided to try and run it as a regular user and then started from scratch and everything fucked up again. No idea
      likes this.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 13:19:44 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     Update, needed to install python 10.6, works =)
      likes this.
   • Vivi Nella Verita :Debian_logo: (verita84@poster.place)'s status on Saturday, 16-Sep-2023 13:19:45 JST Vivi Nella Verita :Debian_logo:

     in reply to

     • 
     @mint
     
     
     Thanks for all the help, it works! I started from scratch using Debian's Native Python 3.11