Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

[BUG] importing transformers4rec.torch libraries gives error

See original GitHub issue

Bug description

After

!pip install transformers4rec[pytorch,nvtabular]

from transformers4rec import torch as tr

on a kaggle notebook I get the error:

TypeError: new() missing 1 required positional argument: ‘task’

Same with all other transformers4rec.torch libraries

Issue Analytics

State:
Created 10 months ago
Comments:6 (1 by maintainers)

Top GitHub Comments

1reaction

sararbcommented, Dec 13, 2022

I have just opened the PR https://github.com/NVIDIA-Merlin/Transformers4Rec/pull/576 that updates Transformers4Rec code with the latest release of Torchmetrics (0.11.0). In fact, Torchmetrics updated its classification metrics (Accuracy, Precision,…) to require a task argument that differentiates three classification tasks: ‘binary’, ‘multiclass’, or ‘multilabel’. When missing, it throws the error TypeError: __new__() missing 1 required positional argument: 'task' I hope this PR fixes the issue you are observing.

1reaction

Alfagovcommented, Dec 3, 2022

The problem is with Torchmetrics, the latest supported version I found is 0.10.3 and they just released 0.11