Tokenizer for GPT-J-6B fails when trying to fine-tune for GLUE tasks

#24
by Jojimon - opened

To profile and optimize performance, I am trying to fine-tune GLUE tasks. The tasks fails as tokenizer does not have pad_token. Tried adding pad_token, it still fails due to differences in dimension. Any help appreciated.

Pad token failure.

image.png

Dimension issue
image.png

Tried using GPT-2 tokenizer, that works with pad_token fix. It would be ideal to use GPT-J-6B tokenizer itself.

Jojimon changed discussion title from Tokenizer for GPT-J-6B fails with trying to fine-tune for GLUE tasks to Tokenizer for GPT-J-6B fails when trying to fine-tune for GLUE tasks

Sign up or log in to comment