EmptyDatasetError when loading dataset downloaded with HuggingFace cli

### Describe the bug

I am using a cluster that does not have access to the internet when given a job. I tried downloading the dataset using the huggingface-cli command and then loading it with load_dataset but I get an error:

```raise EmptyDatasetError(f"The directory at {base_path} doesn't contain any data files") from None```

The dataset I'm using is "lmsys/chatbot_arena_conversations". The folder structure is

- README.md
- data
- train-00000-of-00001-cced8514c7ed782a.parquet

### Steps to reproduce the bug

1. Download dataset using HuggingFace CLI: ```huggingface-cli download lmsys/chatbot_arena_conversations --local-dir ./lmsys/chatbot_arena_conversations```
2. In Python
```
from datasets import load_dataset
load_dataset("lmsys/chatbot_arena_conversations")
```

### Expected behavior

Should return a Dataset Dict in the form of
```
DatasetDict({
train: Dataset({
features: [...],
num_rows: 33,000
})
})
```
### Environment info

Python 3.11.5
Datasets 2.18.0
Transformers 4.38.2
Pytorch 2.2.0
Pyarrow 15.0.1
Rocky Linux release 8.9 (Green Obsidian)