Dataset Viewer issue: TransformRowsProcessingError

#1
by gorovuha - opened

The dataset viewer is not working.

Error details:

Error code:   TransformRowsProcessingError

cc @albertvillanova @lhoestq @severo .

The error:

{
  "error": "list index out of range",
  "cause_exception": "IndexError",
  "cause_message": "list index out of range",
  "cause_traceback": [
    "Traceback (most recent call last):\n",
    " File \"/src/services/worker/src/worker/job_manager.py\", line 125, in process\n job_result = self.job_runner.compute()\n",
    " File \"/src/services/worker/src/worker/job_runners/config/parquet_and_info.py\", line 1392, in compute\n compute_config_parquet_and_info_response(\n",
    " File \"/src/services/worker/src/worker/job_runners/config/parquet_and_info.py\", line 1263, in compute_config_parquet_and_info_response\n fill_builder_info(builder, hf_endpoint=hf_endpoint, hf_token=hf_token, validate=validate)\n",
    " File \"/src/services/worker/src/worker/job_runners/config/parquet_and_info.py\", line 687, in fill_builder_info\n first_url = urls[0]\n",
    "IndexError: list index out of range\n"
  ]
}

It seems like a bug on our side. Raising here: https://github.com/huggingface/dataset-viewer/blob/52b249eb0c2bd584d3c3777898ad85be001ff2f6/services/worker/src/worker/job_runners/config/parquet_and_info.py#L687

I opened an issue: https://github.com/huggingface/dataset-viewer/issues/2672

cc @lhoestq

This errors is due to a config having zero file. Indeed the train/ directory asked in this config in README.md doesn't exist:

  - config_name: train
    data_files:
      - split: train
        path: train/train-*
      - split: validation
        path: train/validation-*
      - split: test
        path: train/test-*

Thank you very much! Unfortunately, there were corrupted images, I have deleted them and the problem disappeared! Thank you for fast response.

Btw, I have another problem with Dataviewer
Rows from parquet row groups are too big to be read: 302.07 MiB (max=286.10 MiB)
Error code: TooBigContentError
So preview doesn't work, but except this everything is fine?

Yes, it's another known issue, when a row contains a lot of data: https://github.com/huggingface/dataset-viewer/issues/2215

Sorry, I didn't get how should I change config features?

I merged your pr but still have the same error Error code: TooBigContentError
Is the problem with heavy images or with long captions?

Hmmm the issue is only for the train split. https://huggingface.co/datasets/gorovuha/ru_image_captioning/viewer/default/test and https://huggingface.co/datasets/gorovuha/ru_image_captioning/viewer/default/validation seem to work.

the problem is generally with long captions, or data embedded in the cells. Heavy images: no, because we create image files, so: we just transfer the URL

So I should fine too long captions, shorten them and update dataset card to make preview work, right?

Yes. No need to update the dataset card, I think.

Thank you! I'll fix it

Actually, I was curious about one thing. In my metadata file I have to descriptions for each photo, but after creating DatasetDict the second one is lost. Even if I have list of to string in my metadata.csv, there is still only one description in the dataset card. What am I doing wrong?

image.png

image.png

image.png

gorovuha changed discussion status to closed

Sign up or log in to comment