Non contiguous tensor value error in model.save_pretrained()

Question

Closed this issue 3 months ago · 3 comments

I encounter this problem today when I fine-tune google-t5/t5-base model and try to save the fine-tuned model.

System Info:
Transformers version: 4.43.4
pytorch version: 2.4.0
Python version: 3.8.19

Answer 1 · 2024-08-09T12:59:41.000Z

This bug is derived from transformers package. It has been solved with PR fix non contiguous tensor value error in save_pretrained. No more errors!

Answer 2 · 2024-09-23T19:52:34.000Z

I still have this issue when using transformers==4.44.2 with sentence-transformers==3.1.0

Answer 3 · 2024-11-08T08:23:42.000Z

I still have this issue when using transformers==4.44.2 with sentence-transformers==3.1.0

Happen to encounter the same issue. You may need to upgrade ur transformers to v4.45.0+ due to the fix has been checked in since this version(link).