CUDA error when using RTX 5060 Ti with Depth Anything v1 (Base and Boost)

[sergiogbrox]

Hi, I’m trying to use the DepthMap extension with my GPU RTX 5060 Ti to generate depth maps using Depth Anything v1 (Base and Boost), but I’m getting the following CUDA error:

 Python 3.10.18 (main, Jul 23 2025, 00:29:36) [MSC v.1944 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Installing requirements
Launching Web UI with arguments: --xformers --api --skip-python-version-check --gradio-allowed-path 'G:\StabilityMatrix\Data\Images'
W1028 18:34:02.210000 28520 venv\Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
G:\StabilityMatrix\Data\Packages\Stable Diffusion WebUI\venv\lib\site-packages\timm\models\layers\__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
  warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning)
G:\StabilityMatrix\Data\Packages\Stable Diffusion WebUI\venv\lib\site-packages\timm\models\registry.py:4: FutureWarning: Importing from timm.models.registry is deprecated, please import via timm.models
  warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.models", FutureWarning)
Started the depthmap API. DO NOT HOST PUBLICLY - SECURITY RISKS!
Loading weights [3267fd6443] from G:\StabilityMatrix\Data\Packages\Stable Diffusion WebUI\models\Stable-diffusion\sd\epicrealismXL_vxviiCrystalclear.safetensors
Running on local URL:  http://127.0.0.1:7860
Creating model from config: G:\StabilityMatrix\Data\Packages\Stable Diffusion WebUI\repositories\generative-models\configs\inference\sd_xl_base.yaml
DepthMap v0.4.8 (e4df29bc)
Backbone: WEBUI
Disabling attention optimization
device: cuda
Loading model(s) ..
Loading model weights from  initialize network with normal
loading the model from ./models/pix2pix\latest_net_G.pth
Computing output(s) ..
  0%|          | 0/1 [00:00<?, ?it/s]wholeImage being processed in : 1536
CUDA error (C:/a/xformers/xformers/third_party/flash-attention/hopper\flash_fwd_launch_template.h:188): invalid argument

System details:

GPU: NVIDIA RTX 5060 Ti

CUDA version: 12.8

Torch: 2.8.0+cu128

xFormers: 0.0.32.post2

Python: 3.10.18

Gradio: 3.41.2

OS: Windows 11

DepthMap version: 0.4.8 (e4df29b)

WebUI version: v1.10.1

Checkpoint: 3267fd6443

The error occurs right after the model starts computing the output (around “wholeImage being processed in : 1536”).
Everything loads fine before that, including the models.

Could this be related to the new GPU architecture or FlashAttention compatibility with the RTX 50-series cards?

Please let me know if you need me to run any tests or provide additional logs.

Thanks in advance for your help! 🙏