Why does my kernal keep dying when I try to import Hugging Face BERT models to Amazon SageMaker?

Question

When I try to import Hugging Face BERT models to the **conda_pytorch_p36** kernal of my Amazon SageMaker Notebook instance using the following pip command, the kernal always dies:
```
! pip install transformers
```
The result is the same for Hugging Face BERT, RoBERTa, and GPT2 models on ml.c5.2xlarge and ml.c5d.4xlarge Amazon SageMaker instances.

Why is this happening, and how do I resolve the issue?

Accepted Answer

This issue occurs when the latest sentence piece breaks. The workaround is to force install sentencepiece==0.1.91.

```
pip install sentencepiece==0.1.91

```

Answer

Any update on this I am trying to download "GPT-J-6B" model on the sagemaker and getting the dying kernel again and again. 
I have explained my issue on [this question](https://repost.aws/questions/QUGqUOzFKiR_2QxZ8nyhJ2Rw/can-not-load-gpt-j-6-b-on-32-gb-instance-from-hugging-face).

And installing `sentencepieve==0.1.91` gives the error. I have also checked from the pypi, the syntax and versions are correct but returns:
*subprocess-exited-with-error in pip install* error.

Any update on this?

(Please view the link given above for my reproducible code)
Thanks.

Why does my kernal keep dying when I try to import Hugging Face BERT models to Amazon SageMaker?

Conteúdo relevante