Add TF weights
#2
by
joaogante
HF staff
- opened
Model converted by the transformers' pt_to_tf CLI. All converted model outputs and hidden layers were validated against its Pytorch counterpart.
Maximum crossload output difference=6.485e-05; Maximum crossload hidden layer difference=2.015e-05;
Maximum conversion output difference=6.485e-05; Maximum conversion hidden layer difference=2.015e-05;
cc @nielsr [HF maintainer(s) for this repo]
Providing additional context.
The output difference exceeds the tolerance. I suspect it's happening because:
- It's a pixel dense task.
- The image resolution is higher than usual.
So, (very) small numerical differences are stacking up and leading to this.
joaogante
changed pull request status to
merged