llama-3.1 finetuning examples breaks with python3.10

[goswamig]

System Info

(myenv) ubuntu@i~$ python --version
Python 3.10.14
(myenv) ubuntu@i~$ pip --version
pip 23.0.1 from /home/ubuntu/myenv/lib/python3.10/site-packages/pip (python 3.10)

Information

• The official example scripts
• My own modified scripts

🐛 Describe the bug

The official finetuning example with peft fails to run

Error logs

from llama_recipes.configs.datasets import samsum_dataset
from llama_recipes.utils.dataset_utils import get_dataloader

samsum_dataset.trust_remote_code = True

train_dataloader = get_dataloader(tokenizer, samsum_dataset, train_config)
eval_dataloader = get_dataloader(tokenizer, samsum_dataset, train_config, "val")

Errror

/home/ubuntu/myenv/lib/python3.10/site-packages/llama_recipes/model_checkpointing/checkpoint_handler.py:17: DeprecationWarning: `torch.distributed._shard.checkpoint` will be deprecated, use `torch.distributed.checkpoint` instead
  from torch.distributed._shard.checkpoint import (
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
Cell In[4], line 2
      1 from llama_recipes.configs.datasets import samsum_dataset
----> 2 from llama_recipes.utils.dataset_utils import get_dataloader
      4 samsum_dataset.trust_remote_code = True
      6 train_dataloader = get_dataloader(tokenizer, samsum_dataset, train_config)

ImportError: cannot import name 'get_dataloader' from 'llama_recipes.utils.dataset_utils' (/home/ubuntu/myenv/lib/python3.10/site-packages/llama_recipes/utils/dataset_utils.py)

$ pip list | grep -i llama
llama-recipes                            0.0.3

Expected behavior

https://github.com/meta-llama/llama-recipes/blob/main/recipes/quickstart/finetuning/quickstart_peft_finetuning.ipynb example should have run fine.

[goswamig]

I believe this is breaking because #650 has updated llama_recipes.utils.dataset_utils however the package is not updated https://pypi.org/project/llama-recipes/#history

[mreso]

Hi @goswamig
thank for reporting this.
You are running the notebook from the current nightly release but you have installed llama-recipes 0.03 which is not compatible anymore. You can either

1. Update llama-recipes to the newest nightly with

pip install -U git+https://github.com/meta-llama/llama-recipes

or 2.:
Use the notebook from 0.0.3 release:
https://github.com/meta-llama/llama-recipes/blob/v0.0.3/recipes/quickstart/finetuning/quickstart_peft_finetuning.ipynb

Let me know if this does not solve the issue.

[goswamig]

Can we also update the wheel package of llama-recipe with the latest change?

[mreso]

We just created a new release so this issue should be fixes.