Datasets:

---

sphyuung

/

datasets-github-issues

like 0

[ { "id": 1935892871, "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement", "name": "enhancement", "color": "a2eeef", "default": true, "description": "New feature or request" } ]

[]

[ "Hi,\r\n\r\nI've downloaded an Arrow-formatted dataset offline using the hugggingface's datasets library by:\r\n\r\nimport json\r\nfrom datasets import load_dataset\r\n\r\ndataset = load_dataset(\"pixparse/cc3m-wds\")\r\ndataset.save_to_disk(\"./cc3m_1\")\r\n\r\n\r\nnow I need to convert it to WebDataset's TAR format for offline data ingestion.\r\nIs there a straightforward method to achieve this conversion without an internet connection? Can I simply convert it by\r\n\r\ntar -cvf\r\n\r\n\r\nbtw, when I tried:\r\n\r\nimport webdataset as wds\r\nfrom huggingface_hub import get_token\r\nfrom torch.utils.data import DataLoader\r\n\r\nhf_token = get_token()\r\nurl = \"https://huggingface.co/datasets/timm/imagenet-12k-wds/resolve/main/imagenet12k-train-{{0000..1023}}.tar\"\r\nurl = f\"pipe:curl -s -L {url} -H 'Authorization:Bearer {hf_token}'\"\r\ndataset = wds.WebDataset(url).decode()\r\ndataset.save_to_disk(\"./cc3m_webdataset\")\r\n\r\n\r\nerror occured:\r\n\r\nAttributeError: 'WebDataset' object has no attribute 'save_to_disk'\r\n\r\n\r\nThanks a lot!\r\n\r\nMotivation\r\n\r\nConverting Arrow to WebDataset TAR Format\r\n\r\nYour contribution\r\n\r\nNo clue yet\r\n\r\n\r\nاحصل على Outlook لـ iOS<https://aka.ms/o0ukef>\r\n________________________________\r\nمن: katie312 ***@***.***>\r\n‏‏تم الإرسال: Friday, December 27, 2024 4:41:21 AM\r\nإلى: huggingface/datasets ***@***.***>\r\nنسخة: Subscribed ***@***.***>\r\n‏‏الموضوع: [huggingface/datasets] Converting Arrow to WebDataset TAR Format for Offline Use (Issue #7347)\r\n\r\n\r\nFeature request\r\n\r\nHi,\r\n\r\nI've downloaded an Arrow-formatted dataset offline using the hugggingface's datasets library by:\r\n\r\nimport json\r\nfrom datasets import load_dataset\r\n\r\ndataset = load_dataset(\"pixparse/cc3m-wds\")\r\ndataset.save_to_disk(\"./cc3m_1\")\r\n\r\n\r\nnow I need to convert it to WebDataset's TAR format for offline data ingestion.\r\nIs there a straightforward method to achieve this conversion without an internet connection? Can I simply convert it by\r\n\r\ntar -cvf\r\n\r\n\r\nbtw, when I tried:\r\n\r\nimport webdataset as wds\r\nfrom huggingface_hub import get_token\r\nfrom torch.utils.data import DataLoader\r\n\r\nhf_token = get_token()\r\nurl = \"https://huggingface.co/datasets/timm/imagenet-12k-wds/resolve/main/imagenet12k-train-{{0000..1023}}.tar\"\r\nurl = f\"pipe:curl -s -L {url} -H 'Authorization:Bearer {hf_token}'\"\r\ndataset = wds.WebDataset(url).decode()\r\ndataset.save_to_disk(\"./cc3m_webdataset\")\r\n\r\n\r\nerror occured:\r\n\r\nAttributeError: 'WebDataset' object has no attribute 'save_to_disk'\r\n\r\n\r\nThanks a lot!\r\n\r\nMotivation\r\n\r\nConverting Arrow to WebDataset TAR Format\r\n\r\nYour contribution\r\n\r\nNo clue yet\r\n\r\n—\r\nReply to this email directly, view it on GitHub<https://github.com/huggingface/datasets/issues/7347>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AQJDZ2X2RUIIULBJEF5R2HL2HSV4DAVCNFSM6AAAAABUH5QSLCVHI2DSMVQWIX3LMV43ASLTON2WKOZSG43DAMRYGIZTGOI>.\r\nYou are receiving this because you are subscribed to this thread.Message ID: ***@***.***>\r\n", "> now I need to convert it to WebDataset's TAR format for offline data ingestion.\r\n\r\nyou can directly download the .TAR files from HF using e.g. `huggingface-cli download` and load them in webdataset :)", "الفله سنه والطبقه يوم\r\n\r\nاحصل على Outlook لـ iOS<https://aka.ms/o0ukef>\r\n________________________________\r\nمن: Quentin Lhoest ***@***.***>\r\n‏‏تم الإرسال: Friday, December 27, 2024 4:14:43 PM\r\nإلى: huggingface/datasets ***@***.***>\r\nنسخة: hamad350 ***@***.***>; Comment ***@***.***>\r\n‏‏الموضوع: Re: [huggingface/datasets] Converting Arrow to WebDataset TAR Format for Offline Use (Issue #7347)\r\n\r\n\r\nnow I need to convert it to WebDataset's TAR format for offline data ingestion.\r\n\r\nyou can directly download the .TAR files from HF using e.g. huggingface-cli download and load them in webdataset :)\r\n\r\n—\r\nReply to this email directly, view it on GitHub<https://github.com/huggingface/datasets/issues/7347#issuecomment-2563691570>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AQJDZ2R5M3Z7L2MZZYARYID2HVHEHAVCNFSM6AAAAABUH5QSLCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNRTGY4TCNJXGA>.\r\nYou are receiving this because you commented.Message ID: ***@***.***>\r\n", "> > now I need to convert it to WebDataset's TAR format for offline data ingestion.\r\n> \r\n> you can directly download the .TAR files from HF using e.g. `huggingface-cli download` and load them in webdataset :)\r\n\r\nThanks a lot! I completely forgot to use Hugging Face-CLI download. Thanks for the reminding!" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7342). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7341). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7340). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7339). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[]

[ { "id": 1935892871, "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement", "name": "enhancement", "color": "a2eeef", "default": true, "description": "New feature or request" } ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7328). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7325). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "Hi @AndreaFrancis and @lhoestq ! Thanks for looking at the code and for all the changes and suggestions. I have worked on all your suggestions. I need to work a bit more on the tests, but I created a first version and uploaded a simple pdf to use for testing. The pdf has a couple of pages some of which have images, figures, and tables (this is exactly the pdfs for which I think keeping the pdf format is very interesting because they have multiple types of content). Will try to finish the tests as soon as possible." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "pip install datasets==2.18.0", "Hi ! I think you need to update axolotl" ]

[]

[]

[ "Now i have other error" ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7319). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[ { "id": 1935892871, "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement", "name": "enhancement", "color": "a2eeef", "default": true, "description": "New feature or request" } ]

[]

[ "#self-assign", "Awesome ! Let me know if you have any question or if I can help :)\r\n\r\ncc @AndreaFrancis as well for viz", "Other candidates libraries for the Pdf type: PyMuPDF pypdf and pdfplumber\r\n\r\nEDIT: Pymupdf looks like a good choice when it comes to maturity + performance + versatility BUT the license is maybe an issue, and pypdf, pypdfium2 or pdfplumber are good options imo", "Related to https://github.com/huggingface/datasets/issues/7058", "PyMuPDF is AGPL licensed, so we can't use it. I will move forward with [pdfplumber](https://github.com/jsvine/pdfplumber?tab=readme-ov-file#python-library).", "Hi both! I have made a pull request with a first basic implementation of the Pdf feature. I followed closely what I saw on the Video and Image features. It is my first time contributing so any comments are very welcomed. I think it would be useful to outline together what additional things we can implement (e.g. enabling parsing of the pdf). Thanks :) " ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7317). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7316). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "Closes #6152 ", "@mariosasko I hope this resolves #6152 " ]

[]

[]

[ "Hello ! when you `cast_column` you need the paths to be absolute paths or relative paths to your working directory, not the original dataset directory.\r\n\r\nThough I'd recommend structuring your dataset as an AudioFolder which automatically links a metadata.jsonl or csv to the audio files via relative paths **within** the dataset repository: https://huggingface.co/docs/datasets/v3.2.0/en/audio_load#audiofolder", "@lhoestq thank you, but there are two problems with using AudioFolder:\r\n1. It is said that AudioFolder requires metadata.csv. However, my datset is too large and contains nested and np.ndarray fields, so I can't use csv.\r\n2. It is said that I need to load the dataset with `load_dataset(\"audiofolder\", ...)`. However, if possible, I want my dataset to be loaded as usual with `load_dataset(dataset_name)` after I upload if to HF.", "You can use metadata.jsonl if you have nested data :)\r\n\r\nAnd actually if you have a dataset structured as an AudioFolder then `load_dataset(dataset_name)` does work after uploading to HF" ]

[]

[]

[]

[ { "id": 1935892871, "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement", "name": "enhancement", "color": "a2eeef", "default": true, "description": "New feature or request" } ]

[]

[]

[ { "id": 1935892871, "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement", "name": "enhancement", "color": "a2eeef", "default": true, "description": "New feature or request" } ]

[]

[ "Hi ! What about having audio + start + duration columns and enable something like this ?\r\n\r\n```python\r\nfor example in ds:\r\n array = example[\"audio\"].read(start=example[\"start\"], frames=example[\"duration\"])\r\n```", "Hi @lhoestq, this would work with a file-based dataset but would be terrible for a sharded one as it would duplicate the large audio file many times. Also, very long audio files are not embedded very well in the parquet file, even with large_binary(). It crashed a few times for me until I switched to one sample == one file :-( " ]

[]

[]

[ "Hi @diarray-hub , thanks for opening the issue :) Let me ping @lhoestq and @severo from the dataset viewer team :hugs: ", "amazing :)", "Hi ! why not modify the manifest.json file directly ? this way users see in the viewer the dataset as is instead which makes it easier to use using e.g. the `datasets` library", "Can I create and push the dataset with the dataset library while also pushing the dataset directory, mainting its structure and all the files as with git? ", "(I transferred to the issue to the `datasets` repo as it's not related to `huggingface_hub`)", "> Can I create and push the dataset with the dataset library while also pushing the dataset directory, mainting its structure and all the files as with git?\r\n\r\nyes push_to_hub simply uploads Parquet files in a directory named \"data\" in the git repository\r\n", "That's the problem actually, I need that the data stays in the same format and the directory they are in keep the same structure in order to go quick with Nemo training so users of Nvidia's Nemo framework don't need to write any preprocessing code before starting training. That's why I used git instead of push_to_hub so me and other users working with Nemo can just:\r\n1. git clone\r\n2. asr_model.setup_training_data(train_data_config={'manifest_filepath': training_manifest_filepath})\r\n\r\nAnd start training already. It may be not very kind of me to prioritize users of a specific framework but I noticed that it take much more code to convert an huggingFace dataset with the parquet file to Nemo manifest format than the inverse :haha: ", "Happy to help if you think the Nemo dataset format should be supported in `datasets` (and therefore in the HF Viewer that is based on `datasets`). Maybe the Nemo team could help as well\r\n\r\nThough I'm not sure if there is only one but actually many formats/structure in Nemo depending on the task ?", "Yeah, you're right Quentin, it depends of the task. This one is for ASR. And, yes maybe they can help. I noticed that they already share their models through HF. Maybe someone in your teams already have a contact point there. Anyway it's not really a big issues since people can easily understand the dataset and its format with the dataset card but it's a little annoying for those who wanna visually explore each features with the viewer as for regular HF datasets", "In that case I'd recommend you to upload the dataset in Nemo format and \r\n1) add the \"nemo\" tag\r\n2) add how to use the dataset in Nemo in the dataset README.md\r\n\r\nThe viewer is likely to show the audio content by default but without the transcriptions. You can also configure the viewer to show the transcriptions instead (without the audio).", "I already did, it's just a little bit \"dommage\" (Hope you'll understand, you speak french right? Cause I don't know any english word for this) that I have to choose which one the viewer displays. But it's no problem for the usability of the dataset. Thanks Quentin :+1: ", "It's \"dommage\" for now, but feel free to ping the Nemo people if you think there is room for making this better together :)\r\n\r\nKinda related, but the `datasets` AudioFolder structure looks similar and simply asks for a `metadata.jsonl` with a field named \"file_name\" to link the transcriptions to the audio files - you could also add this file to your repository to make the viewer show audio + transcripts.\r\n\r\nAlternatively maybe we can expand the AudioFolder configuration to allow you to set the metadata file to be the \"manifest.json\" and the linking field to be \"audio_file_name\" (we just need to agree on something general - not just for Nemo)", "Right, actually that was my idea when I opened this issues. That's what I suggested, taking my case as an exemple but you should think of a more general approach like adding a field to configure the viewer as you wish in the metadata (in the dataset card) or a config.yaml or json file. With a level of abstraction like the solution I proposed ot even higher abstraction, it would allow for more customizability :)" ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7309). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7304). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "Hi ! Can you try with a more recent version of `datasets` ? Also you might need to pass trust_remote_code=True since it's a script based dataset" ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7302). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "No need for a specific version of huggingface_hub to avoid a breaking change no (it's a server-side change)" ]

[]

[]

[ "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7301). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[ "May I request a review @lhoestq", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7300). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Oops my bad ! thanks for reporting" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Updated! 😄 ", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7293). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.", "@lhoestq, can you help with this failing test please? 🙏 " ]

[]

[]

[ "Hi ! If the dataset owner uses `push_to_hub()` instead of `save_to_disk()` and upload the local files it will fix the issue.\r\nRight now `datasets` sees the train/test/valid pickle files but they are not supported file formats.", "Alternatively you can load the arrow file instead:\r\n\r\n```python\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('OpenMol/PubChemSFT', data_files='stage1/*.arrow')\r\n```", "Thanks! I'll have a try." ]

[]

[]

[ "Hi ! `datasets` uses Arrow as storage backend which is agnostic to deep learning frameworks like torch. If you want to get torch tensors back, you need to do `dataset = dataset.with_format(\"torch\")`", "> Hi ! `datasets` uses Arrow as storage backend which is agnostic to deep learning frameworks like torch. If you want to get torch tensors back, you need to do `dataset = dataset.with_format(\"torch\")`\r\n\r\nIt does work! Thanks for your suggestion!" ]

[]

[]

[]

[]

[]

[ "i think this issue is more for https://github.com/huggingface/dataset-viewer" ]

[]

[]

[]

[ { "id": 1935892871, "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement", "name": "enhancement", "color": "a2eeef", "default": true, "description": "New feature or request" } ]

[]

[ "Hi ! You can already configure the README.md to have multiple sets of splits, e.g.\r\n\r\n```yaml\r\nconfigs:\r\n- config_name: my_first_set_of_split\r\n data_files:\r\n - split: train\r\n path: *.csv\r\n- config_name: my_second_set_of_split\r\n data_files:\r\n - split: train\r\n path: train-*.csv\r\n - split: test\r\n path: test-*.csv\r\n```", "Hi - I had something slightly different in mind:\r\n\r\nCurrently the yaml splits specified like this only allow specifying which filenames to pass to each split.\r\nBut what if I have a situation where I know which individual *training examples* I want to put in each split.\r\n\r\nI could build split-specific files, however for large datasets with overlapping (e.g. multiple sets of) splits this could result in significant duplication of data.\r\n\r\nI can see that this could actually be very much intended (i.e. to discourage overlapping splits), but wondered whether some support for handling splits based on individual identifiers is something that could be considered. ", "This is not supported right now :/ Though you can load the data in two steps like this\r\n```python\r\nfrom datasets import load_dataset\r\n\r\nfull_dataset = load_dataset(\"username/dataset\", split=\"train\")\r\nmy_first_set_indices = load_dataset(\"username/dataset\", \"my_first_set_of_split\", split=\"train\")\r\n\r\nmy_first_set = full_dataset.select(my_first_set_indices[\"indices\"])\r\n```\r\n\r\nyou can create such a dataset by adapting this code for example\r\n```python\r\n# upload the full dataset\r\nfull_dataset.push_to_hub(\"username/dataset\")\r\n# then upload the indices for each set\r\nDatasetDict({\r\n \"train\": Dataset.from_dict({\"indices\": [0, 1, 2, 3]}),\r\n \"test\": Dataset.from_dict({\"indices\": [4, 5]}),\r\n}).push_to_hub(\"username/dataset\", \"my_first_set_of_split\")" ]

[ { "id": 1935892871, "node_id": "MDU6TGFiZWwxOTM1ODkyODcx", "url": "https://api.github.com/repos/huggingface/datasets/labels/enhancement", "name": "enhancement", "color": "a2eeef", "default": true, "description": "New feature or request" } ]

[]

[]

[]

[]

[]

[]

[]

[ "@lhoestq ", "The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_7284). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update." ]