使用mps后无法运行

[Raclez]

Description of the bug | 错误描述

NotImplementedError: The operator 'aten::upsample_bicubic2d.out' is not currently implemented for the MPS device. If you want this op to be added in priority during the prototype phase of this feature, please comment on https://github.com/pytorch/pytorch/issues/77764. As a temporary fix, you can set the environment variable PYTORCH_ENABLE_MPS_FALLBACK=1 to use the CPU as a fallback for this op. WARNING: this will be slower than running natively on MPS.

How to reproduce the bug | 如何复现

怎么解决

Operating system | 操作系统

MacOS

Python version | Python 版本

3.10

Software version | 软件版本 (magic-pdf --version)

0.6.x

Device mode | 设备模式

mps

[myhloli]

mps的支持依赖pytorch官方，遇到这个报错请参考官方修复方案

As a temporary fix, you can set the environment variable PYTORCH_ENABLE_MPS_FALLBACK=1 to use the CPU as a fallback for this op.

[Raclez]

export PYTORCH_ENABLE_MPS_FALLBACK=1后就报错这个所以我才问 @Error: command buffer exited with error status. The Metal Performance Shaders operations encoded on it may not have completed. Error: (null) Internal Error (0000000e:Internal Error) <AGXG13GFamilyCommandBuffer: 0x10952cd10> label = device = <AGXG13GDevice: 0x31af29600> name = Apple M1 commandQueue = <AGXG13GFamilyCommandQueue: 0x31af73a00> label = device = <AGXG13GDevice: 0x31af29600> name = Apple M1 retainedReferences = 1

[myhloli]

那只能全局退回cpu了 修改下magic-pdf.json中

{
"device-mode":"cpu"
}

[myhloli]

MPS is not stable，we will remove mps support from main-line.maybe cuda is a nice way to use this project.

[868618]

MPS is not stable，we will remove mps support from main-line.maybe cuda is a nice way to use this project.

大哥，要是mps not stable，我们这些个Mac用户咋办啊，用CPU跑很慢的，我这边有几万个PDF队列，请官方予以考虑

[myhloli]

MPS is not stable，we will remove mps support from main-line.maybe cuda is a nice way to use this project.

大哥，要是mps not stable，我们这些个Mac用户咋办啊，用CPU跑很慢的，我这边有几万个PDF队列，请官方予以考虑

目前确实是mps兼容性问题比较多，受限于开发人员精力有限，后续将不再支持mps。如果只有一个批次的几万本pdf，可以考虑租用支持cuda的云服务，单卡4090大概一天可以处理10w页左右，一周时间差不多可以处理完几万本。