Program hangs when creating a transformers.TrainingArguments object

🤗Transformers
1

Hi! I’m trying to create a transformers.TrainingArguments object in a JupyterLab notebook as follows

transformers.TrainingArguments(
    per_device_train_batch_size=8,
    gradient_accumulation_steps=16,
    warmup_steps=100,
    num_train_epochs=2,
    learning_rate=2e-5,
    fp16=True,
    logging_steps=1,
    output_dir="lora-dolly",
    save_total_limit=3,
)

but the program gets stuck trying to run this cell.

My deps are:

  • torch == 1.13.1+cu117
  • transformers == 4.29.0.dev0

This is really confusing since I’m not loading any model into memory, just trying to initialize a TrainingArguments object. Wonder if anyone has seen similar issues before?

2

Hey, may I know if you fixed this?
I’ve faced the same bug.

3

Same issue on databricks notebook
transformers==4.40.0
torch==2.2.2