xpu-master docker xpu-flex fails README.md example and torchvision warning

[retoXD]

OpenCL Error

I built the image xpu-flex image in xpu-master/docker and it fails the sample code found in README.md (Inference on GPU) RuntimeError: An OpenCL error occurred: -6. I fixed this by adding

FROM intel-extension-for-pytorch:xpu-flex

RUN apt update && apt -y upgrade

Torchvision Warning

It would be nice if the instructions (docker/README.md) mention that this torchvision warning is not a problem. It took me too much time to find the issue where this was mentioned.

root@1f3f1787e64f:/# python -c "import torch; import intel_extension_for_pytorch as ipex; print(torch.__version__); print(ipex.__version__); [print(f'[{i}]: {ipex.xpu.get_device_properties(i)}') for i in range(ipex.xpu.device_count())];"
/usr/local/lib/python3.10/dist-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: 
  warn(f"Failed to load image Python extension: {e}")
1.13.0a0+git6c9b55e
1.13.120+xpu
[0]: _DeviceProperties(name='Intel(R) Arc(TM) A770 Graphics', platform_name='Intel(R) Level-Zero', dev_type='gpu, support_fp64=0, total_memory=15473MB, max_compute_units=512)

Performance Reference

After ultimately getting it to work with stable diffusion using https://github.com/vladmandic/automatic (dev) performance was good.

Prompt: cute dog riding an apple
Steps: 25 | Sampler: DPM++ 2M Karras | CFG scale: 7 | Seed: 1 | Size: 512x512 | Model hash: 6ce0161689 | Model: v1-5-pruned-emaonly | Clip skip: 1 | Version: cc685a8 | Parser: Full parser

Time taken: 2m 28.33s |

GPU active 5424 MB reserved 5908 MB | System peak 5280 MB total 15474 MB

(Variation Seed: 1, Batch Count: 8, Batch Size 8)

RTX 3060 12GB takes 3m 11s, so the A770 is substantially faster ( 2m 28s).

[jingxu10]

Thanks a lot for the great feedback!

Could you elaborate a little bit more for the first issue?
apt upgrade fixed the issue?
Where is FROM intel-extension-for-pytorch:xpu-flex used?

[retoXD]

Yes, starting a container using

IMAGE_NAME=intel-extension-for-pytorch:xpu-flex

VIDEO=$(getent group video | sed -E 's,^video:[^:]*:([^:]*):.*$,\1,')
RENDER=$(getent group render | sed -E 's,^render:[^:]*:([^:]*):.*$,\1,')

test -z "$RENDER" || RENDER_GROUP="--group-add ${RENDER}"

docker run --rm \
    -v <your-local-dir>:/workspace \
    --group-add ${VIDEO} \
    ${RENDER_GROUP} \
    --device=/dev/dri \
    --ipc=host \
    -e http_proxy=$http_proxy \
    -e https_proxy=$https_proxy \
    -e no_proxy=$no_proxy \
    -it $IMAGE_NAME bash

I got the above-mentioned error when running the example code. Running apt update && apt -y upgrade fixes the issue so I added this to my own Dockerfile using the intel-extension-for-pytorch:xpu-flex images, so I didn't have to edit the Dockerfile.xpu file.

[jingxu10]

Got it. Thanks. We will double confirm the docker image for out-of-box usage.

[retoXD]

Thanks! One more note, the repo mentioned in https://github.com/intel/intel-extension-for-pytorch/tree/xpu-master/tests/gpu git clone https://github.com/intel-innersource/frameworks.ai.pytorch.ipex-gpu.git is private.

[jingxu10]

Thanks a lot!

[tye1]

@retoXD thanks, the link issue has been fixed in 00a2734.

[Nuullll]

FYI, the RuntimeError: An OpenCL error occurred: -6 error was related to the intel-opencl-icd package. See vladmandic/sdnext#1474 (comment)

Just confirmed: the issue is gone for me with intel-opencl-icd >= 23.13.26032.26-627~22.04.

And the build script for xpu-master dockerfile specifies an out-of-date version: 23.05.25593.18-601~22.04

intel-extension-for-pytorch/docker/build.sh

Line 34 in 00a2734

--build-arg ICD_VER=23.05.25593.18-601~22.04 \