TrainingArgument does not work on colab

Beginners
1

I was following chapter 3 of the NLP course on Fine-tuning a model with the Trainer API and when executing the command in colab

training_args = TrainingArguments(“test-trainer”)

I get:

ImportError Traceback (most recent call last)
in <cell line: 1>()
----> 1 training_args = TrainingArguments(output_dir=“test-trainer”)

4 frames
/usr/local/lib/python3.10/dist-packages/transformers/training_args.py in _setup_devices(self)
1670 if not is_sagemaker_mp_enabled():
1671 if not is_accelerate_available(min_version=“0.20.1”):
→ 1672 raise ImportError(
1673 “Using the Trainer with PyTorch requires accelerate>=0.20.1: Please run pip install transformers[torch] or pip install accelerate -U”
1674 )

ImportError: Using the Trainer with PyTorch requires accelerate>=0.20.1: Please run pip install transformers[torch] or pip install accelerate -U

NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
“Open Examples” button below.

I try to pip install the recommended ones but never work for me.
Help!

4 Likes
2

I experienced the same import error when running the following script from the hugging face transformer quick tour. Installing the 2 mentioned libraries did not resolve.

from transformers import TrainingArguments

training_args = TrainingArguments(
output_dir=“./pt_training”,
learning_rate=2e-5,
per_device_train_batch_size=8,
per_device_eval_batch_size=8,
num_train_epochs=2,
)

5 Likes
3

I encountered the same problem when I run the code below.

from transformers import TrainingArguments
training_args = TrainingArguments(“test-trainer”)

ERROR：Using the Trainer with PyTorch requires accelerate>=0.20.1: Please run pip install transformers[torch] or pip install accelerate -U

But installing the 2 mentioned libraries did not resolve.
I don’t know what to do now. Help!

1 Like
4

For me help reset environment and install older transformers
!pip install transformers==4.17

3 Likes
5

!pip install transformers==4.17 did not resolve the import error. I forgot to mention I am running the code in Google Colab.

6

I’m getting the same error in the colab notebook for the NLP Course: Lesson 3 - Fine-tuning model with the Trainer API. In order to get around it you can:

  1. Run pip install accelerate -U in a cell
  2. In the top menu click Runtime → Restart Runtime
  3. Do not rerun any cells with !pip install in them
  4. Rerun all the other code cells and you should be good to go!

On a side note, be sure to turn on a GPU for this notebook by clicking Edit → Notebook Settings → GPU type - from the top menu. This was the next thing that got me : )

21 Likes
8

Has this concern been resolved?

9

Thanks for this, it worked for me.

10

Yess worked for me too! Thanks,

11

Thanks it worked but can you please explain this issue.

1 Like
12

Thank you. Because of the above workaround I fixed the error

13

Also works for me! Thank you!

14

This works in colab. Thanks a lot and greatly appreciate you!

15

This resolved the issue. Thanks!

16

Worked for me, thank you!

17

Thank U for sharing :sparkles:

18

Thank you so much, it worked like a charm

19

is very nice idea

20

I am using this, but for this code
training_args = TrainingArguments(

output_dir="./results_banglabert_tier2_tier3",
per_device_train_batch_size=8,
per_device_eval_batch_size=8,
num_train_epochs=3,
logging_dir="./logs_banglabert_tier2_tier3",
logging_steps=10,
evaluation_strategy="epoch",
save_strategy="epoch",
load_best_model_at_end=True,
metric_for_best_model="accuracy"

)
I am facing this error
TypeError Traceback (most recent call last)
in <cell line: 0>()
151 # Training Arguments
152 # =======================
→ 153 training_args = TrainingArguments(
154
155 output_dir=“./results_banglabert_tier2_tier3”,

TypeError: TrainingArguments.init() got an unexpected keyword argument ‘evaluation_strategy’

1 Like
21

evaluation_strategy has been removed.

#evaluation_strategy="epoch",
eval_strategy="epoch",
1 Like