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.

Trying to load bert vocab would result in unpickling error

See original GitHub issue

Hey,

I was trying to serve a finetuned bert model. in the logs of torchserve I noticed that the workers are not being loaded.

The exception I got is

2020-04-22 14:06:23,581 [INFO ] W-9004-sentiment-analysis_1.0-stdout org.pytorch.serve.wlm.WorkerLifeCycle -     self.source_vocab = torch.load(self.manifest['model']['sourceVocab'])
2020-04-22 14:06:23,581 [INFO ] W-9007-sentiment-analysis_1.0-stdout org.pytorch.serve.wlm.WorkerLifeCycle - _pickle.UnpicklingError: invalid load key, '['.

The vocab file I used is this one

Also i noticed that even trying torch.load('vocab.txt') would throw the same error. Am i missing something with how to load the vocab ?

Thank you

Issue Analytics

State:
Created 3 years ago
Comments:11 (3 by maintainers)

Top GitHub Comments

3reactions

MFreidankcommented, Apr 25, 2020

@wassimseif For transformers models, you will need to provide custom handler code, the default text_handler does not support transformers models as of now. For more details on how this would work, see: https://medium.com/@freidankm_39840/deploy-huggingface-s-bert-to-production-with-pytorch-serve-27b068026d18

2reactions

wassimseifcommented, Apr 27, 2020

Hey @vgoklani Thanks !

Oh-My-Zsh + A LOT of configurations & long restless nights