Bug when creating transformer pipeline in sample-apps/transformers

[brandhsu]

Describe the bug

Not a vespa related bug, however, the transformer pipeline cannot be created in sample-apps/transformers.

To Reproduce Issue is here: https://github.com/vespa-engine/sample-apps/blob/master/transformers/src/python/setup-model.py#L16

Fix

pipeline = transformers.pipeline("text-classification", model=model_name)

Environment

• OS: macOS 10.15.7
• Python: 3.9.7
  • PyTorch: 1.10.1
  • Transformers: 4.15.0

[kkraune]

CC @jobergum

[jobergum]

Thanks, fixed by https://github.com/vespa-engine/sample-apps/pull/817

[brandhsu]

Yes, also saw some other tutorials following the same style of pipeline declaration that would also encounter the same error https://github.com/vespa-engine/sample-apps/search?q=transformers.Pipeline

[jobergum]

The DPR model is pinned to a given transformer version, as it does not work with recent updates to the transformer library.

[jobergum]

Thanks for reporting this @Brandhsu !

[brandhsu]

Yes @jobergum, also there is a similar issue with the pyvespa package, I encountered it when doing the stateless model serving.

[kkraune]

@thigm85 please take a look at the pyvespa problem