Datasets:

Modalities:
Tabular
Text
Formats:
parquet
Languages:
code
ArXiv:
Libraries:
Datasets
Dask
License:

Stream specific data_dir

#4
by yashnbx - opened

Is there someway we can stream code for a specific data_dir, I tried running the following code but it seems to iterate from start of items ("Assembly").

dataset = load_dataset(
    "bigcode/the-stack-dedup",
    data_dir = "data/python",
    use_auth_token = token,
    streaming = True,
    cache_dir = "./",
    split="train"
)

If there is some way I can contribute towards this I'd love to be useful.

Screenshot 2022-11-12 at 11.10.07 AM.png

Could it be an issue with how you're creating the dataframe? I just tried your code and it returns python data

image.png

I think this was caused since I attempted downloading in the same cache_dir, it works now. Thanks!

Very interested in exploring this and building new datasets.

Glad to hear it's now working. I look forward to seeing what you build with the Stack!

christopher changed discussion status to closed

i get the same issue, when the set the data_dir with 'data/go', however, the data I got was BABP

BigCode org

Could you send some details on how to reproduce that issue?
In the following script the data looks correct:
Screen Shot 2022-12-13 at 10.51.01 AM.png

Sign up or log in to comment