Open RobinFrcd opened 1 year ago
How did u install diffusers ?
pip install git+https://github.com/ShivamShrirao/diffusers.git
pip install -U -r requirements.txt
Can u verify xformers is working ?
Sure, is there a way to verify this easily ?
Try with just import xformers
Oh yes, sure, it works fine !
The same here on RTX3080 10GB. It does not work natively under Linux. (Ubuntu or Fedora)
diffusers install: pip install git+https://github.com/ShivamShrirao/diffusers.git pip install -U -r requirements.txt
xformers self compiled with cuda 11.6 conda install -c "nvidia/label/cuda-11.6.2" cuda
or from here: conda install -y -c pytorch -c conda-forge cudatoolkit=11.6 pytorch=1.12.1 conda install -y xformers -c xformers/label/dev
nothing works:
RuntimeError: CUDA out of memory. Tried to allocate 16.00 MiB (GPU 0; 9.78 GiB total capacity; 8.06 GiB already allocated; 15.31 MiB free; 8.15 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 520.56.06 Driver Version: 520.56.06 CUDA Version: 11.8 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce ... Off | 00000000:01:00.0 Off | N/A |
| 0% 38C P8 29W / 320W | 0MiB / 10240MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
The really strange thing is that it runs under Windows 11 WSL (ubuntu) without any problems. The same launch.sh script and conda environment.
WSL installation guide from here: https://pastebin.com/uE1WcSxD I would rather run it on Linux.
I've seen that PyTorch is not compatible with CUDA 11.8
, maybe that's your issue here ?
I've seen that PyTorch is not compatible with
CUDA 11.8
, maybe that's your issue here ?
conda list cudatoolkit
# Name Version Build Channel
cudatoolkit 11.6.0 hecad31d_10 conda-forge
deleted, not working
Very similar to what I've done, I'll give it another try, maybe there's an issue with my environment with all the tries I've done !
Try with just
import xformers
When i try import.
- `diffusers` version: 0.7.0.dev0
- Platform: Linux-5.10.102.1-microsoft-standard-WSL2-x86_64-with-glibc2.31
- Python version: 3.9.13
- PyTorch version (GPU?): 1.12.1+cu116 (True)
- Huggingface_hub version: 0.10.0
- Transformers version: 4.22.2
- Using GPU in script?: <fill in>
- Using distributed or parallel set-up in script?: <fill in>
import xformers
should be done in the python interpreter. python -c "import xformers"
import xformers
should be done in the python interpreter.python -c "import xformers"
I do it in Ubuntu console, because i use with WSL2.
Hi, I'm trying to make
dreambooth
work on my 2080 TI. I've compiled xformers withTORCH_CUDA_ARCH_LIST=7.5
.My GPU has about 10.5 GiB of available VRAM. However, when I run the training on my dataset, I get
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 30.00 MiB (GPU 0; 10.74 GiB total capacity; 8.94 GiB already allocated; 9.31 MiB free; 9.03 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.
I've tried to change
--resolution=512
to--resolution=256
but it didn't help.Am I missing something here ? I thought it could run with 9.92 GiB.
Thanks !