Same issue with YOLOv5 on MPS noted in #77748 (comment). I see buffer is not large enough. Must be 25600 bytes

same issue

This issue is fixed in PR #78496 (nightly build 1.13.0.dev20220531 or later).

Same issue with YOLOv5 on device "mps"
MPSNDArray.mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 19200 bytes' #78492

This issue is fixed in PR #78496 (nightly build 1.13.0.dev20220531 or later).
This issue has not been fixed..

@jerjer1223 @GerardWalsh can you please reinstall nightly and see if this resolves ultralytics/yolov5#8102

Hello I am still receiving this error. What do I have to do to resolve this bug/issue? Thanks.

pytorch nightly version: torch-1.13.0.dev20220607

/AppleInternal/Library/BuildRoots/b6051351-c030-11ec-96e9-3e7866fcf3a1/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 160000 bytes

I'm also still receiving this error, a fix would be appreciated.

PyTorch version 1.13.0.dev20220607

Fusing layers...
YOLOv5s summary: 213 layers, 7225885 parameters, 0 gradients
/AppleInternal/Library/BuildRoots/b6051351-c030-11ec-96e9-3e7866fcf3a1/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 25600 bytes

@glenn-jocher

can you please reinstall nightly and see if this resolves ultralytics/yolov5#8102

No it does not

python yolov5/detect.py --source 0 --device='mps'                                       
detect: weights=yolov5/yolov5s.pt, source=0, data=yolov5/data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=mps, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=yolov5/runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False
YOLOv5 🚀 v6.1-246-g2dd3db0 Python-3.8.13 torch-1.13.0.dev20220607 MPS

Fusing layers... 
YOLOv5s summary: 213 layers, 7225885 parameters, 0 gradients
1/1: 0...  Success (inf frames 1280x720 at 30.00 FPS)

/AppleInternal/Library/BuildRoots/b6051351-c030-11ec-96e9-3e7866fcf3a1/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 25600 bytes
'
zsh: abort      python yolov5/detect.py --source 0 --device='mps'

Collecting environment information...
PyTorch version: 1.13.0.dev20220607
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 12.4 (arm64)
GCC version: Could not collect
Clang version: 13.1.6 (clang-1316.0.21.2.5)
CMake version: Could not collect
Libc version: N/A

Python version: 3.8.13 | packaged by conda-forge | (default, Mar 25 2022, 06:04:14)  [Clang 12.0.1 ] (64-bit runtime)
Python platform: macOS-12.4-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

Versions of relevant libraries:
[pip3] numpy==1.22.4
[pip3] torch==1.13.0.dev20220607
[pip3] torchvision==0.14.0.dev20220608
[conda] numpy                     1.22.4           py38he1fcd3f_0    conda-forge
[conda] pytorch                   1.13.0.dev20220607         py3.8_0    pytorch-nightly
[conda] torchvision               0.14.0.dev20220603          pypi_0    pypi

Same issue try running YOLOv5s with mps on M1 Pro

YOLOv5 🚀 2022-6-12 Python-3.9.12 torch-1.13.0.dev20220612 MPS

/AppleInternal/Library/BuildRoots/b6051351-c030-11ec-96e9-3e7866fcf3a1/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:782: failed assertion [MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 19200 bytes

same issue when running with stable baselines 3 contrib PPO recurrent

/AppleInternal/Library/BuildRoots/b6051351-c030-11ec-96e9-3e7866fcf3a1/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 576 bytes
'

YOLOv5 🚀 v6.1-253-g75bbaa8 Python-3.10.4 torch-1.13.0.dev20220616 MPS

/AppleInternal/Library/BuildRoots/b6051351-c030-11ec-96e9-3e7866fcf3a1/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 25600 bytes

I confirm I'm experiencing the same YOLOv5 Apple MPS bug with torch 1.12 on MacBook M1: Error: buffer is not large enough. Must be 25600 bytes

$ glennjocher@Glenns-MacBook-Air yolov5 % python detect.py --device mps

detect: weights=yolov5s.pt, source=data/images, data=data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=mps, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False
YOLOv5 🚀 v6.1-386-g858a1a3 Python-3.9.13 torch-1.12.0 MPS

Fusing layers... 
YOLOv5s summary: 213 layers, 7225885 parameters, 0 gradients
/AppleInternal/Library/BuildRoots/b6051351-c030-11ec-96e9-3e7866fcf3a1/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error: buffer is not large enough. Must be 25600 bytes

This issue is fixed in PR #78496 (nightly build 1.13.0.dev20220531 or later).

@albanD @razarmehr is this error supposed to exist in torch 1.12? I see the fix was was in a 1.13 nightly.

Hi,

I'm not sure if this made it for the release no.
If you're using MPS a lot, I would recommend using the nightly though as we did quite a few fixes that didn't make it to 1.12.

I'm not sure if this made it for the release no. If you're using MPS a lot, I would recommend using the nightly though as we did quite a few fixes that didn't make it to 1.12.

Hmm #78496 was picked into release branch as e3e7531

Reopening to investigate if it still crashes on trunk, and if it is not, why Pytorch-1.12 is still affected

I'm still getting the issue on 1.13.0.dev20220712

Is there any fix?

@glenn-jocher, @daniwnwd this should be fixed in the latest PyTorch nightly (1.13.0.dev20220722). Could you please let me know if you are still seeing the issue on your end?

@DenisVieriu97
NotImplementedError: The operator 'aten::index.Tensor_out' is not current implemented for the MPS device. #82034

@DenisVieriu97 I confirm that the original buffer is not large enough error is now resolved in latest nightly.

YOLOv5 inference still fails on operator 'aten::index.Tensor_out' is not current implemented for the MPS device as mentioned by @crayon7442, but that's a separate issue, so I believe this issue can be closed now.

Thanks a lot @crayon7442 and @glenn-jocher for checking this!
index.Tensor_out is already part of https://github.com/kulinseth/pytorch and we hope to get it soon in PyTorch master

@DenisVieriu97 awesome! Thanks for the update.

I'm on the latest nightly torch-1.13.0.dev20221003 and still getting this error. Could it be cause aten::repeat_interleave.self_int fell back to CPU?

/usr/local/opt/miniforge3/lib/python3.9/site-packages/whisper/decoding.py:628:
/UserWarning: The operator 'aten::repeat_interleave.self_int' is not currently
/supported on the MPS backend and will fall back to run on the CPU. This may
/have performance implications. (Triggered internally at
//Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/mps/MPSFallback.mm:11
/.)
  audio_features = audio_features.repeat_interleave(self.n_group, dim=0)
/AppleInternal/Library/BuildRoots/a0876c02-1788-11ed-b9c4-96898e02b808/Library/
/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.
/mm:782: failed assertion `[MPSNDArray, initWithBuffer:descriptor:] Error:
/buffer is not large enough. Must be 201452 bytes
'

Edited to hardwrap the long lines

@astrowonk that's not an error, that's a warning. As the message states not all torch ops are fully converted to MPS yet.

@astrowonk that's not an error, that's a warning. As the message states not all torch ops are fully converted to MPS yet.

I think you have have to side scroll to see everything @glenn-jocher, the formatting wasn't great. " Error: buffer is not large enough. Must be 201452 bytes" is there on the last line.

and then the kernel dies (in Ipython or Jupyter)

@astrowonk , can you create a new issue with the network and command line and we will take a look. Thanks.

@astrowonk ah yes, I stand corrected!

@astrowonk , can you create a new issue with the network and command line and we will take a look. Thanks.

@kulinseth Opened #86152

I just had this issue and I found the solution. It turns out the error was because my MacOS was too old and PyTorch mps was not compatible. If nothing else it working try updating the operating system.