'Trainer' object has no attribute 'do_grad_scaling'

Short Course Q&A Finetuning Large Language Models
1

On colab, error with ‘Trainer’ object has no attribute ‘do_grad_scaling’. Why no such error on official jupyter lab server?

1 Like
2

Hi @SBFALLOUT,

This can be because one of the packages used for this notebook has a different version on the platform and, possibly the latest, version on in the colab environment. The author might have moved around the things a little bit which is why you are getting this error.

In order to avoid such inconveniences in running things in a colab or locally, please make sure you are using exact package versions as they are being used in the platform.

Best,
Mubsi

1 Like
3

I have the same problem despite working on jupyter notebook.In the utilities.py file there is no “do_grad_scaling” attribute. how can I get the latest version of file?

Thanks,
Volkan

1 Like
4

Hi @Volkan_Celik,

Please take note of this comment.

5

How does one find the package versions ?

6

!pip list doesn’t work

8

Try !pip install transformers==4.32.1
This works for me, I hope this will work for you as well.

3 Likes
10

Thank you！It works for me too

11

The stack trace indicates that the error is in utilities.py file, near line 276

I have replaced this

      if self.do_grad_scaling:
          self.scaler.scale(loss).backward()
      else:
          self.accelerator.backward(loss)

with this

      #if self.do_grad_scaling:
      #    self.scaler.scale(loss).backward()
      #else:
      self.accelerator.backward(loss)
1 Like
13

Any body can share the exact packages to install to avoid any error.
and the exact utilities.py file

14 
# python 3.9.
datasets==2.14.4
transformers[torch]==4.31.0
tokenizers==0.13.3
#python-configuration[yaml]
python-configuration
torch==2.0.1
scipy==1.11.1
zstandard==0.21.0
accelerate==0.21.0
numpy==1.24.3
urllib3==1.26.0
ipywidgets==8.0.7
#lamini==0.0.20
lamini==2.0.1
 -e ./L6/lm-evaluation-harness/.

# ran this in the L6 directory where it is used
#!git clone https://github.com/EleutherAI/lm-evaluation-harness
#!pip install -e lm-evaluation-harness/.
1 Like