Output size of the matrix multiplication is larger than currently supported by the MPS backend: 72250,72250, needs to be less than 2**32 elements #141909
Assignees
Labels
high priority
module: mps
Related to Apple Metal Performance Shaders framework
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Milestone
Comments
ComfyUI Error ReportError Details
Stack TraceSystem Information
Devices
LogsAttached WorkflowPlease make sure that workflow does not contain any sensitive information such as API keys or passwords. Additional Context(Please add any additional context or steps to reproduce the error here) |
|
Just curious: did it work in 2.4? |
|
Unrelated, but a very funny (and unhelpful) error: Fixed in #141927 |
I don't know. I started using ComfyUI very recently. |
Minimal reproimport torch
x1 = torch.randn(1, 72250, 1, device="mps")
x2 = torch.randn(1, 1, 72250, device="mps")
res = torch.bmm(x1, x2)Reproed bad on v2.4.1, where PyTorch crashed with |
|
Commenting out pytorch/aten/src/ATen/native/mps/operations/LinearAlgebra.mm Lines 572 to 579 in 2d9b081 at main reproduces the error from v2.4.1. |
|
From triage discussion: we've determined it's high pri to use fallback or to use tiling in the large matrix case. |
jhavukainen
added a commit
that referenced
this issue
Dec 12, 2024
…ion (old: 2^32 total elem). Addresses #141909
aditew01
pushed a commit
to aditew01/pytorch
that referenced
this issue
Dec 18, 2024
…ytorch#143095) The previous tiling implementation worked for up to 2^32 total elements per single batch entry. This extends the functionality to support the dimensions encountered in ComfyUI (output shape: 1,72250,72250). Fixes pytorch#141909 Pull Request resolved: pytorch#143095 Approved by: https://github.com/kulinseth
pytorchbot
pushed a commit
that referenced
this issue
Jan 10, 2025
…143095) The previous tiling implementation worked for up to 2^32 total elements per single batch entry. This extends the functionality to support the dimensions encountered in ComfyUI (output shape: 1,72250,72250). Fixes #141909 Pull Request resolved: #143095 Approved by: https://github.com/kulinseth (cherry picked from commit afa313e)
kit1980
pushed a commit
that referenced
this issue
Jan 10, 2025
…144558) Extend bmm tiling to work up to 2^32 elem in any single output dim (#143095) The previous tiling implementation worked for up to 2^32 total elements per single batch entry. This extends the functionality to support the dimensions encountered in ComfyUI (output shape: 1,72250,72250). Fixes #141909 Pull Request resolved: #143095 Approved by: https://github.com/kulinseth (cherry picked from commit afa313e) Co-authored-by: Joona Havukainen <jhavukainen@apple.com>
73 tasks
|
Observing Output RC 2.6: Output Nightly: |
|
@atalman what machine you run this one on? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
🚀 The feature, motivation and pitch
Output size of the matrix multiplication is larger than currently supported by the MPS backend: 72250,72250, needs to be less than 2**32 elements
Alternatives
No response
Additional context
Reported as suggested by the error message.
I'm on a Apple M2 Max MacBook Pro with 96GB Ram.
cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @kulinseth @albanD @malfet @DenisVieriu97 @jhavukainen
The text was updated successfully, but these errors were encountered: