NotImplementedError when loading ultrachat_200k dataset from HuggingFace in final chapter code

[Devpatel1012]

Hi,

I'm running the code from the final chapter of the Hands-On LLM book, specifically this snippet that loads the ultrachat_200k dataset and formats it using the TinyLlama chat template:

Code:
from transformers import AutoTokenizer
from datasets import load_dataset

template_tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0")

def format_prompt(example):
chat = example["messages"]
prompt = template_tokenizer.apply_chat_template(chat, tokenize=False)
return {"text": prompt}

dataset = (
load_dataset("HuggingFaceH4/ultrachat_200k", split="test_sft")
.shuffle(seed=42)
.select(range(3_000))
)
dataset = dataset.map(format_prompt)

Error:
NotImplementedError: Loading a dataset cached in a LocalFileSystem is not supported.

Environment:
Python 3.11

datasets version: [please specify, e.g., 2.19.0]

OS: Colab

Platform: Colab

I believe the error is caused by the datasets library not being able to handle caching in certain environments (e.g., temporary FS in Colab or Docker).

Could you suggest a workaround or update the code snippet to handle this?

Thanks for the excellent book and guidance!

[MaartenGr]

I can't seem to reproduce this issue in a Google Colab. I opened the notebook and ran the code to load in the dataset, which worked without issues. Can you explain a bit more why this isn't working for you?