Kernel crashes when loading specific files
Hello,
my Python kernel regularly crashes for specific files.
Here is an example code to reproduce. I also tried to download the data first and then load it from a local path.
from datasets import load_dataset
dataset = load_dataset("EuropeanParliament/Eurovoc", "2002-07")
There are many more files (most of them do not have a preview in the Dataset card).
I do not get any information on why the kernel died.
17:26:19.288 [error] Disposing session as kernel process died ExitCode: undefined, Reason:
Any ideas how to load the whole dataset or how to specifically load 2002-07?
Best,
Johannes
Hi Johannes,
I observed the same behavior, let me try to rebuild that one and see if it can fix it.
In detail, it seems to be an error in the parquet file
Sébastien
Hi Sébastien,
thanks for your reply! I have actually checked the data intensively today and realised that there are empty rows in the corrupted jsonl files. I have removed them locally and filtered them at the same time. But that might not be directly applicable. Anyways, hope that code helps :)
def clean_jsonl_gz_file_filter_DE(input_file, output_file):
"""
Reads a .jsonl.gz file, removes rows based on specific conditions (language and non-null text),
and stores it again as a .jsonl file.
:param input_file: Path to the input .jsonl.gz file.
:param output_file: Path to store the output .jsonl file.
"""
with gzip.open(input_file, 'rt', encoding='utf-8') as infile, open(output_file, 'w', encoding='utf-8') as outfile:
for line in infile:
if line.strip():
data = json.loads(line)
# Filter rows: keep only rows where 'lang' is 'deu' and 'text' is not None
try:
if data.get('lang') == 'deu' and data.get('text') is not None:
json.dump(data, outfile)
outfile.write('\n')
except:
pass
Thanks for your comment, I cleaned up the file by deleting the entries without text and pushed it.
Seems to be fixed on my side, so I close that comment but don't hesitate to reopen it if needed.
Thanks again for your feedback
In [1]: from datasets import load_dataset
...: dataset = load_dataset("EuropeanParliament/Eurovoc", "2002-07")
Downloading readme: 100%|████████████████████████████████████████████████████████████████████████████████| 24.9k/24.9k [00:00<00:00, 19.7MB/s]
Downloading data: 100%|██████████████████████████████████████████████████████████████████████████████████| 74.9M/74.9M [00:18<00:00, 4.14MB/s]
Downloading data files: 100%|███████████████████████████████████████████████████████████████████████████████████| 1/1 [00:18<00:00, 18.59s/it]
Extracting data files: 100%|████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 1.40it/s]
Generating train split: 15463 examples [00:00, 15675.26 examples/s]
In [2]: