JSON formatting issues
Hi AIxBlock Team,
Thank you for releasing this dataset. It looks useful for training conversation models.
While trying to load it with the Hugging Face datasets library, I encountered repeated failures due to malformed JSON files inside some of the ZIP archives.
The dataset cannot be fully loaded using standard Hugging Face tooling and distributed training jobs (PyTorch DDP) fail as soon as they hit one of these malformed files.
Could you please:
- Verify the JSON structure in each ZIP file to ensure every transcript file is valid JSON
- Re-upload the affected archives with corrected formatting.
- Optionally, include a dataset schema or sample JSON in the Hugging Face README so users know what the expected format is.
hi, thanks for your feedback. can u pls share the names of the zip that u got issues? i think HF has issues with archived files, which we didnt have any notification when we uploaded.
I also had errors unzipping the files. This includes the following:
- ❌ automotive_inbound
- ❌ auto_insurance_customer_service_inbound
- ❌ home_ervice_inbound%26telecom%20_outbound
I was able to successfully unzip the ff. files:
- ✅ customer_service_general_inbound
- ✅ insurance_outbound
Other files, I haven't tested.
Thanks for letting me know. Do u wanna test other files first before we reupload only those 3 zip files?
It seems that HF had a trouble with big files.
Hi, tried downloading and unzipping all files just now and had no issues. :)