Open
Description
Tested versions
python == 3.10.18
torch == 2.7.1
lightning == 2.5.2
pyannotate.audo = 3.3.2
System information
macOS Sequoia 15.5
Issue description
Changing the cell
import pytorch_lightning as pl
trainer = pl.Trainer(devices=1, max_epochs=1, accelerator)
trainer.fit(vad_model)to
import pytorch_lightning as pl
from torch import device
vad_model = vad_model.to(device("mps"))
trainer = pl.Trainer(devices = 1, max_epochs = 1)
trainer.fit(vad_model)so that it trains with mps instead of cpu causes a Kernel crash. This crash does not happen if you train with cpu (pl.Trainer(..., accelerator = "cpu")).
Looking at the kernel, this is the reason for the crash:
[error] Disposing session as kernel process died ExitCode: undefined,
Reason: (mpsFileLoc): /AppleInternal/Library/BuildRoots/1c8f7852-1ca9-11f0-b28b-226177e5bb69/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphUtilities.mm:233:0:
error: 'mps.multiply' op operands don't have
5062
broadcast-compatible shapes
Minimal reproduction example (MRE)
Replace training_a_model.ipynb cell specified
Metadata
Metadata
Assignees
Labels
No labels