Running Lightning upgrades permanently causes SHA256 checks to fail in future

[kyleboddy]

In the warnings, if you actually run this command to upgrade the loaded checkpoint:

Lightning automatically upgraded your loaded checkpoint from v1.5.4 to v2.0.6. To apply the upgrade to your files permanently, run python -m pytorch_lightning.utilities.upgrade_checkpoint --file ../.cache/torch/whisperx-vad-segmentation.bin

Then in the future you will have SHA256 model checksum issues:

RuntimeError: Model has been downloaded but the SHA256 checksum does not not match. Please retry loading the model.

Then you have to delete the models in ~/.cache/torch and it will run fine.

Unsure what the workaround is if you want to permanently upgrade.

[AdolfVonKleist]

@kyleboddy you probably solved this already, but the sha256 check is run 'externally' by the library against the baked in s3 url. This is why the checksum fails after upgrade. Either the check needs to be skipped, or preferably updated against the updated checkpoint:

whisperX/whisperx/vad.py

Line 46 in 06e30b2

if hashlib.sha256(model_bytes).hexdigest() != VAD_SEGMENTATION_URL.split('/')[-2]:

[FreeTymeKiyan]

So no fix?

[FreeTymeKiyan]

I removed ~/.cache/torch/whisperx-vad-segmentation.bin and it resumed working.

[transfluxus]

yes, but this comes after running that permanent update
Lightning automatically upgraded your loaded checkpoint from v1.5.4 to v2.4.0. To apply the upgrade to your files permanently, run python -m pytorch_lightning.utilities.upgrade_checkpoint ../../.cache/torch/whisperx-vad-segmentation.bin

[diaskz]

yes, but this comes after running that permanent update Lightning automatically upgraded your loaded checkpoint from v1.5.4 to v2.4.0. To apply the upgrade to your files permanently, run python -m pytorch_lightning.utilities.upgrade_checkpoint ../../.cache/torch/whisperx-vad-segmentation.bin

1 RUN ( with activated you venv)

python -m pytorch_lightning.utilities.upgrade_checkpoint ../../.cache/torch/whisperx-vad-segmentation.bin

2 RUN
sha256sum ../../.cache/torch/whisperx-vad-segmentation.bin

3 Modify ( cat ../venv/lib/python3.12/site-packages/whisperx/vad.py)
nano ../venv/lib/python3.12/site-packages/whisperx/vad.py

and in 19 line change

from:
VAD_SEGMENTATION_URL = "https://whisperx.s3.eu-west-2.amazonaws.com/model_weights/segmentation/0b5b3216d60a2d32fc086b47ea8c67589aaeb26b7e07fcbe620d6d0b83e209ea/pytorch_model.bin"

to:
VAD_SEGMENTATION_URL = "https://whisperx.s3.eu-west-2.amazonaws.com/model_weights/segmentation/ed7051b55315bfb3581da02c63c3e55fbb1eb93b951127d591a2fb95e4d956c6/pytorch_model.bin"

and try to re-run-code