Datasets:

---

amannagrawall002

/

github-issues

like 0

[ "Hi ! Would setting TMPDIR at the beginning of your python script/session work ? I mean, even before importing transformers, datasets, etc. and using them ? I think this would be the most robust solution given any library that uses `tempfile`. I don't think we aim to support environment variables to be changed at run time", "Hi, yeah setting the environment variable before the imports / as environment variable outside is another way to fix this. I am just arguing that `datasets` already uses its own global variable to track temporary files: `_TEMP_DIR_FOR_TEMP_CACHE_FILES`, and the creation of this global variable should respect TMPDIR instead of relying on tempfile to do so." ]

[ "@Woodywarhol9 good catch, thanks for reporting.\r\n\r\nWe are fixing this." ]

[ "Hi @klyuhang9,\r\n\r\nI'm sorry but I can't reproduce your problem:\r\n```python\r\nIn [2]: metric = load_metric('glue', 'rte')\r\nDownloading builder script: 5.76kB [00:00, 2.40MB/s]\r\n```\r\n\r\nCould you please, retry to load the metric? Sometimes there are temporary connectivity issues.\r\n\r\nFeel free to re-open this issue of the problem persists." ]

[ "Hi @klyuhang9,\r\n\r\nI'm sorry but I can't reproduce your problem:\r\n```python\r\nIn [4]: ds = load_dataset(\"glue\", \"sst2\", download_mode=\"force_redownload\")\r\nDownloading builder script: 28.8kB [00:00, 9.15MB/s] \r\nDownloading metadata: 28.7kB [00:00, 10.7MB/s] \r\nDownloading and preparing dataset glue/sst2 (download: 7.09 MiB, generated: 4.78 MiB, post-processed: Unknown size, total: 11.88 MiB) to .../.cache/huggingface/datasets/glue/sst2/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad...\r\nDownloading data: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 7.44M/7.44M [00:01<00:00, 4.12MB/s]\r\nDataset glue downloaded and prepared to .../.cache/huggingface/datasets/glue/sst2/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad. Subsequent calls will reuse this data. \r\n100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 1047.96it/s]\r\n\r\nIn [5]: ds\r\nOut[5]: \r\nDatasetDict({\r\n train: Dataset({\r\n features: ['sentence', 'label', 'idx'],\r\n num_rows: 67349\r\n })\r\n validation: Dataset({\r\n features: ['sentence', 'label', 'idx'],\r\n num_rows: 872\r\n })\r\n test: Dataset({\r\n features: ['sentence', 'label', 'idx'],\r\n num_rows: 1821\r\n })\r\n})\r\n```\r\n\r\nPlease, note that sometimes GitHub has some temporary connectivity issues. Feel free to retry and re-open this issue if the problem persists.", "Maybe it's because we are in China.", "Are you able to access the URL in your web browser?", "> Are you able to access the URL in your web browser?\r\n\r\nYes, with or without a VPN, we (people in China) can access the URL. And we can even use wget to download these files. We can download the pretrained language model automatically with the code.\r\nHowever, we CANNOT access glue.py & metric.py automatically. Every time, it will raise ConnectionError, and we have to download datasets manually (SQuAD is extremely hard to preprocess) and replace metric.py with scipy.metrics. If this problem is solved, many Chinese will save a lot of time.", "> ConnectionError: Couldn't reach https://raw.githubusercontent.com/huggingface/datasets/2.0.0/datasets/glue/glue.py\r\n> \r\n> I don't know why; it is ok when I use\r\n\r\nIf you would query the question `ConnectionError: Couldn't reach` in www.baidu.com (Chinese Google, Google is banned and some people cannot access it), you will find that there are so many questions about accessing `https://raw.githubusercontent.com`. There are some solutions like adding `185.199.108.133 raw.githubusercontent.com` to `C:/windows/systen32/drives/etc/hosts`, but it is time-consuming, hard for green-hand, and invalid sometimes." ]

[ "Follow-up: I have confirmed there are no duplicate lines via `sort amazon_reviews_us_PC_v1_00.tsv | uniq -cd` after extracting the raw file.", "Hi @trentonstrong, thanks for reporting!\r\n\r\nI confirm that loading this dataset configuration throws a `NonMatchingSplitsSizesError`:\r\n```\r\nNonMatchingSplitsSizesError: [{'expected': SplitInfo(name='train', num_bytes=350242049, num_examples=785730, dataset_name='amazon_us_reviews'), 'recorded': SplitInfo(name='train', num_bytes=3982712078, num_examples=6908554, dataset_name='amazon_us_reviews')}]\r\n```\r\n\r\nAlso thank you for your offer to fix this. You can find information about how to update the metadata JSON file here: https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md#automatically-add-code-metadata\r\n```shell\r\ndatasets-cli test datasets/amazon_us_reviews --save_infos --all_configs\r\n```\r\nPlease, feel free to open a PR with this fix. And do not hesitate to ping me if you need any help.", "No sweat. Will get it patched up ASAP." ]

[ "Hi @orkenstein, thanks for reporting.\r\n\r\nPlease note that for this case, our `datasets` library uses under the hood the Apache Arrow `unique` function: https://arrow.apache.org/docs/python/generated/pyarrow.compute.unique.html#pyarrow.compute.unique\r\n\r\nAnd currently the Apache Arrow `unique` function is only implemented for these input types (see info in their [docs](https://arrow.apache.org/docs/cpp/compute.html#array-wise-vector-functions)): Boolean, Null, Numeric, Temporal, Binary- and String-like.\r\n\r\nHowever, the data types of the `wikiann` dataset are all `list<item: string>` (see its [dataset card](https://huggingface.co/datasets/wikiann#data-fields)), and thus, not yet supported by the Apache Arrow `unique` function.", "As a workaround solution you can use pandas:\r\n```python\r\nfrom datasets import load_dataset\r\n\r\ndataset = load_dataset('wikiann', 'en', split='train')\r\ndf = dataset.to_pandas()\r\nunique_df = df[~df.tokens.apply(tuple).duplicated()] # from https://stackoverflow.com/a/46958336/17517845\r\n```\r\n\r\nNote that pandas loads the dataset in memory (this one is small so it's fine).", "@lhoestq thank you! I will fall back to this method for now" ]

[]

[ "Hi! Thanks for the detailed analysis of adding IIIF support. I like the idea of \"using IIIF through datasets scripts\" due to its ease of use. Another approach that I like is yielding image ids and using the `piffle` library (which offers a bit more flexibility) + `map` to download + cache images. We can handle bad URLs in `map` by returning `None`. Plus, we can add a `Dataset Preprocessing` section with the code that explains this approach to the card of such datasets. WDYT?\r\n\r\n> currently, IIIF is mainly used by cultural heritage organizations (museums, archives etc.) The adoption of IIIF in this sector has been growing but it's possible that adoption won't be extended to other industries which may also be a source of image data for training ML models.\r\n\r\nThis is why (currently) adding a new feature type would be overkill, IMO.\r\n" ]

[ "After some investigation, maybe the bug is in `doc-builder`.\r\n\r\nI've opened an issue there:\r\n- huggingface/doc-builder#160", "Fixed by @lewtun (thank you):\r\n- huggingface/doc-builder@31fe6c8bc7225810e281c2f6c6cd32f38828c504" ]

[ "Thnaks for reporting @RRaphaell.\r\n\r\nWe are fixing it. " ]

[ "Hi @cathyxl, thanks for reporting.\r\n\r\nIndeed, we have recently updated the loading script of that dataset (and fixed that bug as well):\r\n- #4002\r\n\r\nThat fix will be available in our next `datasets` library release. In the meantime, you can incorporate that fix by:\r\n- installing `datasets` from our GitHub repo:\r\n```bash\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\n- forcing the data files to be redownloaded\r\n```python\r\nds = load_dataset('conll2012_ontonotesv5', 'english_v4', split=\"test\", download_mode=\"force_redownload\")\r\n```\r\n\r\nFeel free to re-open this issue if the problem persists." ]

[ "Hi ! You can access the faiss index with\r\n```python\r\nfaiss_index = my_dataset.get_index(\"my_index_name\").faiss_index\r\n```\r\nand then do whatever you want with it, e.g. query it using range_search:\r\n```python\r\nthreshold = 0.95\r\nlimits, distances, indices = faiss_index.range_search(x=xq, thresh=threshold)\r\n\r\ntexts = dataset[indices]\r\n```", "wow, that's great, thank you for the explanation. (if that's not already in the documentation, could be worth adding it)\r\n\r\nwhich type of faiss index is Datasets using? I looked into faiss recently and I understand that there are several different types of indexes and the choice is important, e.g. regarding which distance metric you use (euclidian vs. cosine/dot product), the size of my dataset etc. can I chose the type of index somehow as well?", "`Dataset.add_faiss_index` has a `string_factory` parameter, used to set the type of index (see the faiss documentation about [index factory](https://github.com/facebookresearch/faiss/wiki/The-index-factory)). Alternatively, you can pass an index you've defined yourself using faiss with the `custom_index` parameter of `Dataset.add_faiss_index` \r\n\r\nHere is the full documentation of `Dataset.add_faiss_index`: https://huggingface.co/docs/datasets/v2.0.0/en/package_reference/main_classes#datasets.Dataset.add_faiss_index", "great thanks, I will try it out" ]

[ "Hi, @MoritzLaurer, thanks for reporting.\r\n\r\nNormally this is due to a mismatch between the versions of your Elasticsearch client and server:\r\n- your ES client is passing only keyword arguments to your ES server\r\n- whereas your ES server expects a positional argument called 'scheme'\r\n\r\nIn order to fix this, you should align the major versions of both Elasticsearch client and server.\r\n\r\nYou can have more info:\r\n- on this other issue page: https://github.com/huggingface/datasets/issues/3956#issuecomment-1072115173\r\n- Elasticsearch client docs: https://www.elastic.co/guide/en/elasticsearch/client/python-api/current/overview.html#_compatibility\r\n\r\nFeel free to re-open this issue if the problem persists.\r\n\r\nDuplicate of:\r\n- #3956", "1. Check elasticsearch version\r\n```\r\nimport elasticsearch\r\nprint(elasticsearch.__version__)\r\n```\r\nEx: 7.9.1\r\n2. Uninstall current elasticsearch package\r\n`pip uninstall elasticsearch`\r\n3. Install elasticsearch 7.9.1 package\r\n`pip install elasticsearch==7.9.1`" ]

[ "Thanks for the suggestion, @Dref360.\r\n\r\nWe are adding that argument. " ]

[ "I found that the problem arises because the image files in my folder are actually symbolic links (for my own reasons). After modifications, the classes can now be correctly parsed. Therefore, I close this issue.", "HI, I have a question. How much time did you load the ImageNet data files? " ]

[ "Hi @hazalturkmen, thanks for reporting.\r\n\r\nNote that your dataset repository does not contain any loading script; it only contains a data file named `tr_article_2`.\r\n\r\nWhen there is no loading script but only data files, the `datasets` library tries to infer how to load the data by looking at the data file extensions. However, your data file does not have any extension.\r\n\r\nNote that current supported data file extensions are: 'csv', 'tsv', 'json', 'jsonl', 'parquet', 'txt', 'blp', 'bmp', 'dib', 'bufr', 'cur', 'pcx', 'dcx', 'dds', 'ps', 'eps', 'fit', 'fits', 'fli', 'flc', 'ftc', 'ftu', 'gbr', 'gif', 'grib', 'h5', 'hdf', 'png', 'apng', 'jp2', 'j2k', 'jpc', 'jpf', 'jpx', 'j2c', 'icns', 'ico', 'im', 'iim', 'tif', 'tiff', 'jfif', 'jpe', 'jpg', 'jpeg', 'mpg', 'mpeg', 'msp', 'pcd', 'pxr', 'pbm', 'pgm', 'ppm', 'pnm', 'psd', 'bw', 'rgb', 'rgba', 'sgi', 'ras', 'tga', 'icb', 'vda', 'vst', 'webp', 'wmf', 'emf', 'xbm', 'xpm', 'zip'.\r\n\r\nYou have more info on our docs: [How to share a dataset](https://huggingface.co/docs/datasets/share).", "thanks for reply :)" ]

[ "Issue unresolved, see [4000](https://github.com/huggingface/datasets/issues/4009#issue-1179658611)" ]

[ "Hi! Use the `ArrayXD` feature type (where X is the number of dimensions) to get correctly formated tensors. So in your case, define the dataset as follows :\r\n```python\r\nds = Dataset.from_dict({\"A\": [torch.rand((2, 2))]}, features=Features({\"A\": Array2D(shape=(2, 2), dtype=\"float32\")}))\r\n```\r\n", "Hi @mariosasko I'm facing the same issue and the only work around I've found so far is to convert my `DatasetDict` to a dictionary and then create new objects with `Dataset.from_dict`.\r\n```\r\ndataset = load_dataset(\"my_dataset.py\")\r\ndataset = dataset.map(lambda example: blabla(example))\r\ndict_dataset_test = dataset[\"test\"].to_dict()\r\n...\r\ndataset_test = Dataset.from_dict(dict_dataset_test, features=Features(features))\r\n```\r\nHowever, converting a `Dataset` object to a dict takes quite a lot of time and memory... Is there a way to directly create an `Array2D` without having to transform the original `Dataset` to a dict?", "Hi! Yes, you can directly pass the `Features` dictionary as `features` in `map` to cast the column to `Array2D`:\r\n```python\r\ndataset = dataset.map(lambda example: blabla(example), features=Features(features))\r\n```\r\nOr you can use `cast` after `map` to do that:\r\n```python\r\ndataset = dataset.map(lambda example: blabla(example))\r\ndataset = dataset.cast(Features(features))\r\n```", "Fantastic thank you @mariosasko\r\nThe first option you suggested is indeed way faster 😃 " ]

[ "I don't think it's an issue with the dataset-viewer. Maybe @lhoestq or @albertvillanova could confirm.\r\n\r\n```python\r\n>>> from datasets import load_dataset, DownloadMode\r\n>>> import itertools\r\n>>> # without streaming\r\n>>> dataset = load_dataset(\"yelp_review_full\", name=\"yelp_review_full\", split=\"train\", download_mode=DownloadMode.FORCE_REDOWNLOAD)\r\n\r\nDownloading builder script: 4.39kB [00:00, 5.97MB/s]\r\nDownloading metadata: 2.13kB [00:00, 3.14MB/s]\r\nDownloading and preparing dataset yelp_review_full/yelp_review_full (download: 187.06 MiB, generated: 496.94 MiB, post-processed: Unknown size, total: 684.00 MiB) to /home/slesage/.cache/huggingface/datasets/yelp_review_full/yelp_review_full/1.0.0/13c31a618ba62568ec8572a222a283dfc29a6517776a3ac5945fb508877dde43...\r\nDownloading data: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.10k/1.10k [00:00<00:00, 1.39MB/s]\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/slesage/hf/datasets/src/datasets/load.py\", line 1687, in load_dataset\r\n builder_instance.download_and_prepare(\r\n File \"/home/slesage/hf/datasets/src/datasets/builder.py\", line 605, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"/home/slesage/hf/datasets/src/datasets/builder.py\", line 1104, in _download_and_prepare\r\n super()._download_and_prepare(dl_manager, verify_infos, check_duplicate_keys=verify_infos)\r\n File \"/home/slesage/hf/datasets/src/datasets/builder.py\", line 676, in _download_and_prepare\r\n verify_checksums(\r\n File \"/home/slesage/hf/datasets/src/datasets/utils/info_utils.py\", line 40, in verify_checksums\r\n raise NonMatchingChecksumError(error_msg + str(bad_urls))\r\ndatasets.utils.info_utils.NonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://drive.google.com/uc?export=download&id=0Bz8a_Dbh9QhbZlU4dXhHTFhZQU0']\r\n\r\n>>> # with streaming\r\n>>> dataset = load_dataset(\"yelp_review_full\", name=\"yelp_review_full\", split=\"train\", download_mode=DownloadMode.FORCE_REDOWNLOAD, streaming=True)\r\n\r\nDownloading builder script: 4.39kB [00:00, 5.53MB/s]\r\nDownloading metadata: 2.13kB [00:00, 3.14MB/s]\r\nTraceback (most recent call last):\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/implementations/http.py\", line 375, in _info\r\n await _file_info(\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/implementations/http.py\", line 736, in _file_info\r\n r.raise_for_status()\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/aiohttp/client_reqrep.py\", line 1000, in raise_for_status\r\n raise ClientResponseError(\r\naiohttp.client_exceptions.ClientResponseError: 403, message='Forbidden', url=URL('https://doc-0g-bs-docs.googleusercontent.com/docs/securesc/ha0ro937gcuc7l7deffksulhg5h7mbp1/gklhpdq1arj8v15qrg7ces34a8c3413d/1648144575000/07511006523564980941/*/0Bz8a_Dbh9QhbZlU4dXhHTFhZQU0?e=download')\r\n\r\nThe above exception was the direct cause of the following exception:\r\n\r\nTraceback (most recent call last):\r\n File \"<stdin>\", line 1, in <module>\r\n File \"/home/slesage/hf/datasets/src/datasets/load.py\", line 1677, in load_dataset\r\n return builder_instance.as_streaming_dataset(\r\n File \"/home/slesage/hf/datasets/src/datasets/builder.py\", line 906, in as_streaming_dataset\r\n splits_generators = {sg.name: sg for sg in self._split_generators(dl_manager)}\r\n File \"/home/slesage/.cache/huggingface/modules/datasets_modules/datasets/yelp_review_full/13c31a618ba62568ec8572a222a283dfc29a6517776a3ac5945fb508877dde43/yelp_review_full.py\", line 102, in _split_generators\r\n data_dir = dl_manager.download_and_extract(my_urls)\r\n File \"/home/slesage/hf/datasets/src/datasets/utils/streaming_download_manager.py\", line 800, in download_and_extract\r\n return self.extract(self.download(url_or_urls))\r\n File \"/home/slesage/hf/datasets/src/datasets/utils/streaming_download_manager.py\", line 778, in extract\r\n urlpaths = map_nested(self._extract, path_or_paths, map_tuple=True)\r\n File \"/home/slesage/hf/datasets/src/datasets/utils/py_utils.py\", line 306, in map_nested\r\n return function(data_struct)\r\n File \"/home/slesage/hf/datasets/src/datasets/utils/streaming_download_manager.py\", line 783, in _extract\r\n protocol = _get_extraction_protocol(urlpath, use_auth_token=self.download_config.use_auth_token)\r\n File \"/home/slesage/hf/datasets/src/datasets/utils/streaming_download_manager.py\", line 372, in _get_extraction_protocol\r\n with fsspec.open(urlpath, **kwargs) as f:\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/core.py\", line 102, in __enter__\r\n f = self.fs.open(self.path, mode=mode)\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/spec.py\", line 978, in open\r\n f = self._open(\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/implementations/http.py\", line 335, in _open\r\n size = size or self.info(path, **kwargs)[\"size\"]\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/asyn.py\", line 88, in wrapper\r\n return sync(self.loop, func, *args, **kwargs)\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/asyn.py\", line 69, in sync\r\n raise result[0]\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/asyn.py\", line 25, in _runner\r\n result[0] = await coro\r\n File \"/home/slesage/.pyenv/versions/datasets/lib/python3.8/site-packages/fsspec/implementations/http.py\", line 388, in _info\r\n raise FileNotFoundError(url) from exc\r\nFileNotFoundError: https://drive.google.com/uc?export=download&id=0Bz8a_Dbh9QhbZlU4dXhHTFhZQU0&confirm=t\r\n```\r\n\r\nAnd this is before even trying to access the rows with\r\n\r\n```python\r\n>>> rows = list(itertools.islice(dataset, 100))\r\n>>> rows = list(dataset.take(100))\r\n```", "Yet another issue related to google drive not being nice. Most likely your IP has been banned from using their API programmatically. Do you know if we are allowed to host and redistribute the data ourselves ?", "Hi,\r\n\r\nFacing the same issue while loading the dataset: \r\n\r\n`Error: {NonMatchingChecksumError}Checksums didn't match for dataset source files`\r\n\r\nThanks", "> Facing the same issue while loading the dataset:\r\n> \r\n> Error: {NonMatchingChecksumError}Checksums didn't match for dataset source files\r\n\r\nThanks for reporting. I think this is the same issue. Feel free to try again later, once Google Drive stopped blocking you. You can retry by passing `download_mode=\"force_redownload\"` to `load_dataset`", "I noticed that FastAI hosts the Yelp dataset at https://s3.amazonaws.com/fast-ai-nlp/yelp_review_full_csv.tgz (from their catalog [here](https://course.fast.ai/datasets))\r\n\r\nLet's update the yelp dataset script to download from there instead of Google Drive", "I updated the link to not use Google Drive anymore, we will do a release early next week with the updated download url of the dataset :)" ]

[ "Using latest code, I am still facing the issue.\r\n\r\n```python\r\n(base) vimos@vimosmu ➜ ~ ipython\r\nPython 3.6.7 | packaged by conda-forge | (default, Nov 6 2019, 16:19:42) \r\nType 'copyright', 'credits' or 'license' for more information\r\nIPython 7.11.1 -- An enhanced Interactive Python. Type '?' for help.\r\n\r\nIn [1]: from datasets import load_dataset\r\n\r\nIn [2]: load_dataset(\"assin2\")\r\nDownloading builder script: 4.24kB [00:00, 244kB/s]\r\nDownloading metadata: 2.58kB [00:00, 2.19MB/s]\r\nUsing custom data configuration default\r\nDownloading and preparing dataset assin2/default (download: 2.02 MiB, generated: 1.21 MiB, post-processed: Unknown size, total: 3.23 MiB) to /home/vimos/.cache/huggingface/datasets/assin2/default/1.0.0/8467f7acbda82f62ab960ca869dc1e96350e0e103a1ef7eaa43bbee530b80061...\r\nDownloading data: 1.51MB [00:00, 102MB/s]\r\nDownloading data: 116kB [00:00, 63.6MB/s]\r\nDownloading data: 493kB [00:00, 95.8MB/s] \r\nDownloading data files: 100%|██████████████████████████████████████████| 3/3 [00:00<00:00, 8.27it/s]\r\n---------------------------------------------------------------------------\r\nExpectedMoreDownloadedFiles Traceback (most recent call last)\r\n<ipython-input-2-b367d1ffd68e> in <module>\r\n----> 1 load_dataset(\"assin2\")\r\n\r\n~/anaconda3/lib/python3.6/site-packages/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, keep_in_memory, save_infos, revision, use_auth_token, task, streaming, **config_kwargs)\r\n 1694 ignore_verifications=ignore_verifications,\r\n 1695 try_from_hf_gcs=try_from_hf_gcs,\r\n-> 1696 use_auth_token=use_auth_token,\r\n 1697 )\r\n 1698\r\n\r\n~/anaconda3/lib/python3.6/site-packages/datasets/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, base_path, use_auth_token, **download_and_prepare_kwargs)\r\n 604 if not downloaded_from_gcs:\r\n 605 self._download_and_prepare(\r\n--> 606 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n 607 )\r\n 608 # Sync info\r\n\r\n~/anaconda3/lib/python3.6/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos)\r\n 1102\r\n 1103 def _download_and_prepare(self, dl_manager, verify_infos):\r\n-> 1104 super()._download_and_prepare(dl_manager, verify_infos, check_duplicate_keys=verify_infos)\r\n 1105\r\n 1106 def _get_examples_iterable_for_split(self, split_generator: SplitGenerator) -> ExamplesIterable:\r\n\r\n~/anaconda3/lib/python3.6/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)\r\n 675 if verify_infos:\r\n 676 verify_checksums(\r\n--> 677 self.info.download_checksums, dl_manager.get_recorded_sizes_checksums(), \"dataset source files\"\r\n 678 )\r\n 679\r\n\r\n~/anaconda3/lib/python3.6/site-packages/datasets/utils/info_utils.py in verify_checksums(expected_checksums, recorded_checksums, verification_name)\r\n 31 return\r\n 32 if len(set(expected_checksums) - set(recorded_checksums)) > 0:\r\n---> 33 raise ExpectedMoreDownloadedFiles(str(set(expected_checksums) - set(recorded_checksums)))\r\n 34 if len(set(recorded_checksums) - set(expected_checksums)) > 0:\r\n 35 raise UnexpectedDownloadedFile(str(set(recorded_checksums) - set(expected_checksums)))\r\n\r\nExpectedMoreDownloadedFiles: {'https://drive.google.com/u/0/uc?id=1kb7xq6Mb3eaqe9cOAo70BaG9ypwkIqEU&export=download', 'https://drive.google.com/u/0/uc?id=1J3FpQaHxpM-FDfBUyooh-sZF-B-bM_lU&export=download', 'https://drive.google.com/u/0/uc?id=1Q9j1a83CuKzsHCGaNulSkNxBm7Dkn7Ln&export=download'}\r\n```", "That's true. Steps to reproduce the bug on Google Colab:\r\n\r\n```\r\ngit clone https://github.com/huggingface/datasets.git\r\ncd datasets\r\npip install -e .\r\npython -c \"from datasets import load_dataset; print(load_dataset('assin2')['train'][0])\"\r\n```\r\n\r\nHowever the dataset will load without any problems if you just install version 2.0.0:\r\n\r\n ```\r\npip install datasets\r\npython -c \"from datasets import load_dataset; print(load_dataset('assin2')['train'][0])\"\r\n```\r\n\r\nAny thoughts @lhoestq ?", "Right indeed ! Let me open a PR to fix this.\r\nThe dataset_infos.json file that stores some metadata about the dataset to download (and is used to verify it was correctly downloaded) hasn't been updated correctly", "Not sure what the status of this is, but personally I am still getting this error, with glue.", "Can you open a new issue if you got an error with glue please ?", "Have posted at #4241" ]

[ "Hi! Replacing `nlp.<obj>` with `datasets.<obj>` in the script should fix the problem. `nlp` has been renamed to `datasets` more than a year ago, so please use `datasets` instead to avoid weird issues.", "Thank You! Was able to solve with the help of this.", "But I request you to please fix the same in the dataset hub explorer as well...", "May I ask how to get this dataset?" ]

[ "Hi @i-am-neo,\r\n\r\nThe audio support is an extra feature of `datasets` and therefore it must be installed as an additional optional dependency:\r\n```shell\r\npip install datasets[audio]\r\n```\r\nAdditionally, for specific MP3 support (which is not the case for AMI dataset, that contains WAV audio files), there is another third-party dependency on `torchaudio`.\r\n\r\nYou have all the information in our docs: https://huggingface.co/docs/datasets/audio_process#installation", "Thanks @albertvillanova . Unfortunately the error persists after installing ```datasets[audio]```. Can you direct towards a solution?\r\n\r\n```\r\npip3 install datasets[audio]\r\n```\r\n### log\r\nRequirement already satisfied: datasets[audio] in ./.virtualenvs/hubert/lib/python3.7/site-packages (1.18.3)\r\nRequirement already satisfied: numpy>=1.17 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (1.21.5)\r\nRequirement already satisfied: xxhash in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (3.0.0)\r\nRequirement already satisfied: fsspec[http]>=2021.05.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (2022.2.0)\r\nRequirement already satisfied: dill in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (0.3.4)\r\nRequirement already satisfied: pandas in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (1.3.5)\r\nRequirement already satisfied: huggingface-hub<1.0.0,>=0.1.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (0.4.0)\r\nRequirement already satisfied: packaging in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (21.3)\r\nRequirement already satisfied: multiprocess in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (0.70.12.2)\r\nRequirement already satisfied: pyarrow!=4.0.0,>=3.0.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (7.0.0)\r\nRequirement already satisfied: tqdm>=4.62.1 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (4.63.1)\r\nRequirement already satisfied: aiohttp in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (3.8.1)\r\nRequirement already satisfied: importlib-metadata in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (4.11.3)\r\nRequirement already satisfied: requests>=2.19.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (2.27.1)\r\nRequirement already satisfied: librosa in ./.virtualenvs/hubert/lib/python3.7/site-packages (from datasets[audio]) (0.9.1)\r\nRequirement already satisfied: pyyaml in ./.virtualenvs/hubert/lib/python3.7/site-packages (from huggingface-hub<1.0.0,>=0.1.0->datasets[audio]) (6.0)\r\nRequirement already satisfied: typing-extensions>=3.7.4.3 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from huggingface-hub<1.0.0,>=0.1.0->datasets[audio]) (4.1.1)\r\nRequirement already satisfied: filelock in ./.virtualenvs/hubert/lib/python3.7/site-packages (from huggingface-hub<1.0.0,>=0.1.0->datasets[audio]) (3.6.0)\r\nRequirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from packaging->datasets[audio]) (3.0.7)\r\nRequirement already satisfied: idna<4,>=2.5 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->datasets[audio]) (3.3)\r\nRequirement already satisfied: certifi>=2017.4.17 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->datasets[audio]) (2021.10.8)\r\nRequirement already satisfied: charset-normalizer~=2.0.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->datasets[audio]) (2.0.12)\r\nRequirement already satisfied: urllib3<1.27,>=1.21.1 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->datasets[audio]) (1.26.9)\r\nRequirement already satisfied: attrs>=17.3.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from aiohttp->datasets[audio]) (21.4.0)\r\nRequirement already satisfied: frozenlist>=1.1.1 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from aiohttp->datasets[audio]) (1.3.0)\r\nRequirement already satisfied: aiosignal>=1.1.2 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from aiohttp->datasets[audio]) (1.2.0)\r\nRequirement already satisfied: yarl<2.0,>=1.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from aiohttp->datasets[audio]) (1.7.2)\r\nRequirement already satisfied: asynctest==0.13.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from aiohttp->datasets[audio]) (0.13.0)\r\nRequirement already satisfied: multidict<7.0,>=4.5 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from aiohttp->datasets[audio]) (6.0.2)\r\nRequirement already satisfied: async-timeout<5.0,>=4.0.0a3 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from aiohttp->datasets[audio]) (4.0.2)\r\nRequirement already satisfied: zipp>=0.5 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from importlib-metadata->datasets[audio]) (3.7.0)\r\nRequirement already satisfied: decorator>=4.0.10 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (5.1.1)\r\nRequirement already satisfied: soundfile>=0.10.2 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (0.10.3.post1)\r\nRequirement already satisfied: numba>=0.45.1 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (0.55.1)\r\nRequirement already satisfied: pooch>=1.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (1.6.0)\r\nRequirement already satisfied: resampy>=0.2.2 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (0.2.2)\r\nRequirement already satisfied: audioread>=2.1.5 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (2.1.9)\r\nRequirement already satisfied: joblib>=0.14 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (1.1.0)\r\nRequirement already satisfied: scipy>=1.2.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (1.7.3)\r\nRequirement already satisfied: scikit-learn>=0.19.1 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from librosa->datasets[audio]) (1.0.2)\r\nRequirement already satisfied: python-dateutil>=2.7.3 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from pandas->datasets[audio]) (2.8.2)\r\nRequirement already satisfied: pytz>=2017.3 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from pandas->datasets[audio]) (2022.1)\r\nRequirement already satisfied: setuptools in ./.virtualenvs/hubert/lib/python3.7/site-packages (from numba>=0.45.1->librosa->datasets[audio]) (60.10.0)\r\nRequirement already satisfied: llvmlite<0.39,>=0.38.0rc1 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from numba>=0.45.1->librosa->datasets[audio]) (0.38.0)\r\nRequirement already satisfied: appdirs>=1.3.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from pooch>=1.0->librosa->datasets[audio]) (1.4.4)\r\nRequirement already satisfied: six>=1.5 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from python-dateutil>=2.7.3->pandas->datasets[audio]) (1.16.0)\r\nRequirement already satisfied: threadpoolctl>=2.0.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from scikit-learn>=0.19.1->librosa->datasets[audio]) (3.1.0)\r\nRequirement already satisfied: cffi>=1.0 in ./.virtualenvs/hubert/lib/python3.7/site-packages (from soundfile>=0.10.2->librosa->datasets[audio]) (1.15.0)\r\nRequirement already satisfied: pycparser in ./.virtualenvs/hubert/lib/python3.7/site-packages (from cffi>=1.0->soundfile>=0.10.2->librosa->datasets[audio]) (2.21)\r\n\r\n### reload\r\n```\r\npython3 -c \"from datasets import load_dataset; print(load_dataset('ami', 'headset-single', split='validation')[0])\"\r\n```\r\n\r\n### log\r\nDownloading and preparing dataset ami/headset-single (download: 10.71 GiB, generated: 49.99 MiB, post-processed: Unknown size, total: 10.76 GiB) to /home/neo/.cache/huggingface/datasets/ami/headset-single/1.6.2/2accdf810f7c0585f78f4bcfa47684fbb980e35d29ecf126e6906dbecb872d9e...\r\nAMI corpus cannot be downloaded using multi-processing. Setting number of downloaded processes `num_proc` to 1. \r\n100%|██████████████████████████████████████████████████████| 136/136 [00:00<00:00, 33542.59it/s]\r\n100%|█████████████████████████████████████████████████████████| 136/136 [00:06<00:00, 22.28it/s]\r\n100%|████████████████████████████████████████████████████████| 18/18 [00:00<00:00, 21558.39it/s]\r\n100%|█████████████████████████████████████████████████████████| 18/18 [00:00<00:00, 2996.41it/s]\r\n100%|████████████████████████████████████████████████████████| 16/16 [00:00<00:00, 23431.87it/s]\r\n100%|█████████████████████████████████████████████████████████| 16/16 [00:00<00:00, 2697.52it/s]\r\nTraceback (most recent call last):\r\n File \"<string>\", line 1, in <module>\r\n File \"/home/neo/.virtualenvs/hubert/lib/python3.7/site-packages/datasets/load.py\", line 1707, in load_dataset\r\n use_auth_token=use_auth_token,\r\n File \"/home/neo/.virtualenvs/hubert/lib/python3.7/site-packages/datasets/builder.py\", line 595, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/home/neo/.virtualenvs/hubert/lib/python3.7/site-packages/datasets/builder.py\", line 690, in _download_and_prepare\r\n ) from None\r\nOSError: Cannot find data file. \r\nOriginal error:\r\nsndfile library not found\r\n\r\n### just to double-check as per your docs\r\n```\r\npip3 install librosa torchaudio\r\n```\r\n\r\n### logs\r\nRequirement already satisfied: librosa in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (0.9.1)\r\nRequirement already satisfied: torchaudio in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (0.11.0+cu113)\r\nRequirement already satisfied: audioread>=2.1.5 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (2.1.9)\r\nRequirement already satisfied: joblib>=0.14 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (1.1.0)\r\nRequirement already satisfied: packaging>=20.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (21.3)\r\nRequirement already satisfied: scikit-learn>=0.19.1 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (1.0.2)\r\nRequirement already satisfied: scipy>=1.2.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (1.7.3)\r\nRequirement already satisfied: decorator>=4.0.10 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (5.1.1)\r\nRequirement already satisfied: resampy>=0.2.2 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (0.2.2)\r\nRequirement already satisfied: pooch>=1.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (1.6.0)\r\nRequirement already satisfied: numpy>=1.17.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (1.21.5)\r\nRequirement already satisfied: soundfile>=0.10.2 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (0.10.3.post1)\r\nRequirement already satisfied: numba>=0.45.1 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from librosa) (0.55.1)\r\nRequirement already satisfied: torch==1.11.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from torchaudio) (1.11.0+cu113)\r\nRequirement already satisfied: typing-extensions in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from torch==1.11.0->torchaudio) (4.1.1)\r\nRequirement already satisfied: setuptools in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from numba>=0.45.1->librosa) (60.10.0)\r\nRequirement already satisfied: llvmlite<0.39,>=0.38.0rc1 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from numba>=0.45.1->librosa) (0.38.0)\r\nRequirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from packaging>=20.0->librosa) (3.0.7)\r\nRequirement already satisfied: requests>=2.19.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from pooch>=1.0->librosa) (2.27.1)\r\nRequirement already satisfied: appdirs>=1.3.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from pooch>=1.0->librosa) (1.4.4)\r\nRequirement already satisfied: six>=1.3 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from resampy>=0.2.2->librosa) (1.16.0)\r\nRequirement already satisfied: threadpoolctl>=2.0.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from scikit-learn>=0.19.1->librosa) (3.1.0)\r\nRequirement already satisfied: cffi>=1.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from soundfile>=0.10.2->librosa) (1.15.0)\r\nRequirement already satisfied: pycparser in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from cffi>=1.0->soundfile>=0.10.2->librosa) (2.21)\r\nRequirement already satisfied: charset-normalizer~=2.0.0 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->pooch>=1.0->librosa) (2.0.12)\r\nRequirement already satisfied: certifi>=2017.4.17 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->pooch>=1.0->librosa) (2021.10.8)\r\nRequirement already satisfied: urllib3<1.27,>=1.21.1 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->pooch>=1.0->librosa) (1.26.9)\r\nRequirement already satisfied: idna<4,>=2.5 in /home/neo/.virtualenvs/hubert/lib/python3.7/site-packages (from requests>=2.19.0->pooch>=1.0->librosa) (3.3)\r\n\r\n### try loading again\r\n```\r\npython3 -c \"from datasets import load_dataset; print(load_dataset('ami', 'headset-single', split='validation')[0])\"\r\n```\r\n\r\n### same error\r\nDownloading and preparing dataset ami/headset-single (download: 10.71 GiB, generated: 49.99 MiB, post-processed: Unknown size, total: 10.76 GiB) to /home/neo/.cache/huggingface/datasets/ami/headset-single/1.6.2/2accdf810f7c0585f78f4bcfa47684fbb980e35d29ecf126e6906dbecb872d9e...\r\nAMI corpus cannot be downloaded using multi-processing. Setting number of downloaded processes `num_proc` to 1. \r\n100%|██████████████████████████████████████████████████████| 136/136 [00:00<00:00, 33542.59it/s]\r\n100%|█████████████████████████████████████████████████████████| 136/136 [00:06<00:00, 22.28it/s]\r\n100%|████████████████████████████████████████████████████████| 18/18 [00:00<00:00, 21558.39it/s]\r\n100%|█████████████████████████████████████████████████████████| 18/18 [00:00<00:00, 2996.41it/s]\r\n100%|████████████████████████████████████████████████████████| 16/16 [00:00<00:00, 23431.87it/s]\r\n100%|█████████████████████████████████████████████████████████| 16/16 [00:00<00:00, 2697.52it/s]\r\nTraceback (most recent call last):\r\n File \"<string>\", line 1, in <module>\r\n File \"/home/neo/.virtualenvs/hubert/lib/python3.7/site-packages/datasets/load.py\", line 1707, in load_dataset\r\n use_auth_token=use_auth_token,\r\n File \"/home/neo/.virtualenvs/hubert/lib/python3.7/site-packages/datasets/builder.py\", line 595, in download_and_prepare\r\n dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n File \"/home/neo/.virtualenvs/hubert/lib/python3.7/site-packages/datasets/builder.py\", line 690, in _download_and_prepare\r\n ) from None\r\nOSError: Cannot find data file. \r\nOriginal error:\r\nsndfile library not found\r\n", "Hi @i-am-neo, thanks again for your detailed report.\r\n\r\nOur `datasets` library support for audio relies on a third-party Python library called `librosa`, which is installed when you do:\r\n```shell\r\npip install datasets[audio]\r\n```\r\n\r\nHowever, the `librosa` library has a dependency on `soundfile`; and `soundfile` depends on a non-Python package called `sndfile`. \r\n\r\nOn Linux (which is your case), this must be installed manually using your operating system package manager, for example:\r\n```shell\r\nsudo apt-get install libsndfile1\r\n```\r\n\r\nPlease, let me know if this works and if so, I will update our docs with all this information.", "@albertvillanova thanks, all good. The key is ```libsndfile1``` - it may help others to note that in your docs. I had installed libsndfile previously." ]

[ "Good catch ! Yes I think passing `format=\"wav\"` is the right thing to do", "Thanks @polinaeterna for reporting this issue.\r\n\r\nIn relation to the decoding of MP3 audio files without torchaudio, I remember Patrick made some tests and these had quite bad performance. That is why he proposed to support MP3 files only with torchaudio. But yes, nice to give an alternative to non-torchaudio users (with a big warning on performance).", "> I remember Patrick made some tests and these had quite bad performance. That is why he proposed to support MP3 files only with torchaudio.\r\n\r\nYeah, I know, but as far as I understand, some users just categorically don't want to have torchaudio in their environment. Anyway, it's just a more or less random example, they can use any library they like following the same logic (I'm just not a big expert in decoding utils so if you can give me some presentation / resources about that I would really appreciate it 🤗)" ]

[ "Same thing occurs when streaming files loaded from disk.", "Hi ! Thanks for reporting, could this be related to https://github.com/huggingface/datasets/issues/3950 ?\r\n\r\nCurrently streaming datasets only works in single process, but we're working on having in work in distributed setups as well :) (EDIT: done)", "Hi, thanks for your reply. It seems related :)", "+1", "Please update `datasets` if you're having this issue. What version are you using ?" ]

[ "Hi! This behavior stems from this line: https://github.com/huggingface/datasets/blob/799b817d97590ddc97cbd38d07469403e030de8c/src/datasets/arrow_dataset.py#L1919\r\nBasically, the `Image`/`Audio` columns are decoded only if the `format_type` attribute is `None` (`set_format`/`with_format` and `set_transform`/`with_transform` assign a non-`None` value to it) and the `input_columns` param is not specified (see https://github.com/huggingface/datasets/issues/3756). We will remove these limitations soon.\r\n\r\n\r\n\r\n" ]

[]

[ "There is an open PR to do that: #3364. I just haven't had time to finish it... ", "> There is an open PR to do that: #3364. I just haven't had time to finish it...\r\n\r\n😬 thanks..." ]

[ "Hi ! I didn't managed to reproduce the issue. When you kill the process, is there any stacktrace that shows at what point in the code python is hanging ?", "Hi @lhoestq , I've traced the issue back to file locking. It's similar to this thread, using Lustre filesystem as well. https://github.com/huggingface/datasets/issues/329 . In this case the user was able to modify and add -o flock option while mounting and it solved the problem. \r\nHowever in other cases such as mine, we do not have the permissions to modify the commands while mounting. I'm still trying to figure out a workaround. Any ideas how can we use a mounted Lustre filesystem with no flock option?\r\n", "Hi @kelvinAI , I've had this issue on our institution's system which uses Lustre (in addition to our compute nodes being siloed off from external network access). The workaround I made for downloading/loading datasets was to set the `$HFHOME` environment variable to a location on the node's local storage (SSD), effectively a location that gets cleared regularly and sometimes gets used for temporary or cached files which is pretty common, e.g. \"scratch\" storage. Maybe your sysadmins, if you have them, could point you to subdirectories on a node that aren't linked to the Lustre filesystem. After downloading to scratch I found that the transformers, modules, and metrics cached folders were fine to move to my user drives on the Lustre filesystem but cached datasets that had fingerprints still had some issues with filelock, so it would help to use the function `my_dataset.save_to_disk('path/on/lustre_fs')` and static class function `Dataset.load_from_disk('path/on/lustre_fs')`. In rough steps:\r\n\r\n1. Initially download to scratch storage with `ds = datasets.load_dataset(dataset_name)`\r\n2. Call `ds.save_to_disk(my_path_on_lustre)` with a path in your user space on the Lustre filesystem\r\n3. Load datasets with `from datasets import Dataset; new_ds = Dataset.load_from_disk(my_path_on_lustre)`\r\n\r\nObviously this hinges on there existing scratch storage on the nodes you're using. Fingers crossed.", "Hi @jpmcd , thanks for sharing your experience. For my case, the Lustre filesystem (with more storage space) is the scratch storage like the one you've mentioned. We have a local storage for each user but unfortunately there's not enough space in it to 'cache' huge datasets, hence that is why I tried changing HF_HOME to point to the scratch disk with more space and encountered the flock issue. Unfortunately I'm not aware of any viable solution to this for now so I simply fall back to using torch dataset. ", "@jpmcd your comment saved me from pulling my hair out in frustration. Setting `HF_HOME` to a directory that's not on Lustre works like a charm. ✨ " ]

[]

[ "Hi ! To be able to run the tests, you need to install all the test dependencies and additional ones with\r\n```\r\npip install -e .[tests]\r\npip install -r additional-tests-requirements.txt --no-deps\r\n```\r\n\r\nIn particular, you probably need to `sacrebleu`. It looks like it wasn't able to instantiate `sacrebleu.TER` properly." ]

[ "Hi ! Thanks for reporting. We're using filelock` as our locking mechanism.\r\n\r\nCan you try deleting the .lock file mentioned in the logs and try again ? Make sure that no other process is generating the `cnn_dailymail` dataset.\r\n\r\nIf it doesn't work, could you try to set up a lock using the latest version of filelock` and see if it works ?\r\n\r\n```\r\npip install filelock\r\n```\r\nhere is a code example from the `filelock` documentation that you can try:\r\n\r\n```python\r\nfrom filelock import FileLock\r\n\r\nlock = FileLock(\"test.txt.lock\")\r\nwith lock:\r\n with open(\"test.txt\", \"a\") as f:\r\n f.write(\"foo\")\r\n```", "I ran into this problem on my school server as well? Any update on how we can solve it? Thanks! ", "Have you tried running the code above to check if FileLock works in your setup ? You may also be interested in checking the https://github.com/tox-dev/filelock repository for issues", "Can you try using a different cache directory ? Maybe there are permissions issues with the default one.\r\n\r\nYou can do so by passing `cache_dir=...` to load_dataset()" ]

[ "the dataset viewer is working on this dataset. I imagine the issue is that we would expect to be able to listen to the audio files in the `Please Record Your Voice file` column, right?\r\n\r\nmaybe @lhoestq or @albertvillanova could help\r\n\r\n<img width=\"1019\" alt=\"Capture d’écran 2022-03-24 à 17 36 20\" src=\"https://user-images.githubusercontent.com/1676121/159966006-57dcf8f7-b65f-4200-ac8c-66859318a8bb.png\">\r\n", "The structure of the dataset is not supported. Only the CSV file is parsed and the audio files are ignored.\r\n\r\nWe're working on supporting audio datasets with a specific structure in #3963 ", "Got it.", "Current error:\r\n\r\n```\r\nError code: StreamingRowsError\r\nException: LibsndfileError\r\nMessage: Error opening <File-like object HfFileSystem, datasets/kingabzpro/Urdu-ASR-flags@6a8878cfe3a41343fa86ec8b4254209fe56a0f0d/Please Record Your Voice/0.wav>: Format not recognised.\r\nTraceback: Traceback (most recent call last):\r\n File \"/src/services/worker/src/worker/utils.py\", line 263, in get_rows_or_raise\r\n return get_rows(\r\n File \"/src/services/worker/src/worker/utils.py\", line 204, in decorator\r\n return func(*args, **kwargs)\r\n File \"/src/services/worker/src/worker/utils.py\", line 241, in get_rows\r\n rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py\", line 1357, in __iter__\r\n example = _apply_feature_types_on_example(\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py\", line 1051, in _apply_feature_types_on_example\r\n decoded_example = features.decode_example(encoded_example, token_per_repo_id=token_per_repo_id)\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/datasets/features/features.py\", line 1902, in decode_example\r\n return {\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/datasets/features/features.py\", line 1903, in <dictcomp>\r\n column_name: decode_nested_example(feature, value, token_per_repo_id=token_per_repo_id)\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/datasets/features/features.py\", line 1325, in decode_nested_example\r\n return schema.decode_example(obj, token_per_repo_id=token_per_repo_id)\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/datasets/features/audio.py\", line 187, in decode_example\r\n array, sampling_rate = sf.read(f)\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/soundfile.py\", line 285, in read\r\n with SoundFile(file, 'r', samplerate, channels,\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/soundfile.py\", line 658, in __init__\r\n self._file = self._open(file, mode_int, closefd)\r\n File \"/src/services/worker/.venv/lib/python3.9/site-packages/soundfile.py\", line 1216, in _open\r\n raise LibsndfileError(err, prefix=\"Error opening {0!r}: \".format(self.name))\r\n soundfile.LibsndfileError: Error opening <File-like object HfFileSystem, datasets/kingabzpro/Urdu-ASR-flags@6a8878cfe3a41343fa86ec8b4254209fe56a0f0d/Please Record Your Voice/0.wav>: Format not recognised.\r\n```\r\n\r\nMaybe switch to a discussion here? https://huggingface.co/datasets/kingabzpro/Urdu-ASR-flags/discussions. cc @albertvillanova " ]

[ "Thanks for reporting @qqaatw.\r\n\r\nYes, we should definitely adapt that file for `datasets`. " ]

[ "Hi ! You can download the `oscar.py` file from this repository at `/datasets/oscar/oscar.py`.\r\n\r\nThen you can load the dataset by passing the local path to `oscar.py` to `load_dataset`:\r\n```python\r\nload_dataset(\"path/to/oscar.py\", \"unshuffled_deduplicated_it\")\r\n```", "it works,but another error occurs.\r\n```\r\nConnectionError: Couldn't reach https://s3.amazonaws.com/datasets.huggingface.co/oscar/1.0/unshuffled/deduplicated/it/it_sha256.txt (SSLError(MaxRetryError(\"HTTPSConnectionPool(host='s3.amazonaws.com', port=443): Max retries exceeded with url: /datasets.huggingface.co/oscar/1.0/unshuffled/deduplicated/it/it_sha256.txt (Caused by SSLError(SSLEOFError(8, 'EOF occurred in violation of protocol (_ssl.c:1129)')))\")))\r\n```\r\nI can access `https://s3.amazonaws.com/datasets.huggingface.co/oscar/1.0/unshuffled/deduplicated/it/it_sha256.txt` and `https://aws.amazon.com/cn/s3/` directly, so why it reports a SSLError, should I need tomodify the host file？", "Could it be an issue with your python environment or your version of OpenSSL ?", "you are so wise!\r\nit report [ConnectionError] in python 3.9.7\r\nand works well in python 3.8.12\r\n\r\nI need you help again: how can I specify the path for download files?\r\nthe data is too large and my C hardware is not enough", "Cool ! And you can specify the path for download files with to the `cache_dir` parameter:\r\n```python\r\nfrom datasets import load_dataset\r\ndataset = load_dataset('oscar', 'unshuffled_deduplicated_it', cache_dir='path/to/directory')", "It takes me some days to download data completely, Despise sometimes it occurs again, change py version is feasible way to avoid this ConnectionEror.\r\nparameter `cache_dir` works well, thanks for your kindness again!" ]

[ "I guess the cache got corrupted due to a previous issue with Google Drive service.\r\n\r\nThe cache should be regenerated, e.g. by passing `download_mode=\"force_redownload\"`.\r\n\r\nCC: @severo ", "Note that the dataset preview uses its own cache, not `datasets`' cache. So `download_mode=\"force_redownload\"` doesn't help. But yes indeed the cache must be refreshed.\r\n\r\nThe CNN Dailymail dataste is currently hosted on Google Drive, which is an unreliable host and we've had many issues with it. Unless we found another most reliable host for the data, we will keep running into issues from time to time.\r\n\r\nAt Hugging Face we're not allowed to host the CNN Dailymail data by ourselves AFAIK", "Yes @lhoestq, I didn't explain myself well: my previous message was addressed to @severo. ", "I remove the tag dataset-viewer, since it's more an issue with the hosting on Google Drive", "Sounds good. I was looking for another host of this dataset but couldn't find any (yet)", "It seems like the issue is with the streaming mode, not with the hosting:\r\n\r\n```python\r\n>>> import datasets\r\n>>> dataset = datasets.load_dataset('cnn_dailymail', name=\"3.0.0\", split=\"train\", streaming=True, download_mode=\"force_redownload\")\r\nDownloading builder script: 9.35kB [00:00, 10.2MB/s]\r\nDownloading metadata: 9.50kB [00:00, 12.2MB/s]\r\n>>> len(list(dataset))\r\n0\r\n>>> dataset = datasets.load_dataset('cnn_dailymail', name=\"3.0.0\", split=\"train\", streaming=False)\r\nReusing dataset cnn_dailymail (/home/slesage/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0/3cb851bf7cf5826e45d49db2863f627cba583cbc32342df7349dfe6c38060234)\r\n>>> len(dataset)\r\n287113\r\n```\r\n\r\nNote, in particular, that the streaming mode is failing silently, returning 0 row while I would have expected an exception instead. The result is that the dataset viewer shows `No data` instead of a detailed error.\r\n\r\n<img width=\"1511\" alt=\"Capture d’écran 2022-04-12 à 11 50 46\" src=\"https://user-images.githubusercontent.com/1676121/162935341-d50f1e73-d053-41d4-917f-e79708a0ca23.png\">\r\n", "Well this is because the host (Google Drive) returns a document that is not the actual data, but an error page", "Do you think that `datasets` should detect this anyway and throw an exception?", "Yes it definitely should ! I don't have the bandwidth to work on this right now though", "Indeed, streaming was not supported: tgz archives were not properly iterated.\r\n\r\nI've opened a PR to support streaming.\r\n\r\nHowever, keep in mind that Google Drive will keep generating issues from time to time, like 403,..." ]

[ "Hi @cahya-wirawan, thanks for reporting.\r\n\r\nYour dataset is working OK in streaming mode:\r\n```python\r\nIn [1]: from datasets import load_dataset\r\n ...: ds = load_dataset(\"indonesian-nlp/eli5_id\", split=\"train\", streaming=True)\r\n ...: item = next(iter(ds))\r\n ...: item\r\nUsing custom data configuration indonesian-nlp--eli5_id-9fe728a7e760fb7b\r\n\r\nOut[1]: \r\n{'q_id': '1oy5tc',\r\n 'title': 'dalam sepak bola apa gunanya menyia-nyiakan dua permainan pertama dengan terburu-buru - di tengah - bukan permainan terburu-buru biasa saya mendapatkannya',\r\n 'selftext': '',\r\n 'document': '',\r\n 'subreddit': 'explainlikeimfive',\r\n 'answers': {'a_id': ['ccwtgnz', 'ccwtmho', 'ccwt946', 'ccwvj0u'],\r\n 'text': ['Jaga pertahanan tetap jujur, rasakan operan terburu-buru, buka permainan yang lewat. Pelanggaran yang terlalu satu dimensi akan gagal. Dan mereka yang bergegas ke tengah kadang-kadang dapat dibuka lebar-lebar untuk ukuran yard yang besar.',\r\n 'Jika Anda melempar bola sepanjang waktu, maka pertahanan akan beradaptasi untuk selalu menutupi umpan. Dengan melakukan permainan lari sederhana sesekali, Anda memaksa pertahanan untuk tetap dekat dan menjaga dari lari. Terkadang, pelanggaran dapat membuat pertahanan lengah dengan berpura-pura berlari dan membebaskan penerima mereka. Selain itu, Anda tidak perlu mendapatkan yard besar di setiap permainan. Terkadang, paling baik mendapatkan beberapa yard sekaligus. Selama Anda mendapatkan yang pertama, Anda dalam kondisi yang baik.',\r\n 'Dalam kebanyakan kasus, O-Line seharusnya membuat lubang untuk dilalui kembali. Jika Anda menjalankan terlalu banyak permainan ke luar / melempar, pertahanan akan mengejar. Juga, 2 permainan 5 yard memberi Anda satu set down baru.',\r\n 'Saya Anda tidak suka jenis drama itu, tonton CFL. Kami hanya mendapatkan 3 down sehingga Anda tidak bisa menyia-nyiakannya. Lebih banyak lagi yang lewat.'],\r\n 'score': [3, 2, 2, 2]},\r\n 'title_urls': {'url': []},\r\n 'selftext_urls': {'url': []},\r\n 'answers_urls': {'url': []}}\r\n```\r\nTherefore, it should be properly rendered in the previewer. Let me ping @severo to have a look at it.", "Thanks @albertvillanova for checking it. Btw, I have another dataset indonesian-nlp/lfqa_id which has the same issue. However, this dataset is still private, is it the reason why the preview doesn't work?", "Yes, preview is not supported on private datasets yet. We are working on that though...", "Thanks for the confirmation ", "Fixed. Thanks for your feedback." ]

[ "Hi @lewtun, thanks for reporting.\r\n\r\nIt seems that our library fails at inferring the dtype of the columns:\r\n- `milestone`\r\n- `performed_via_github_app` \r\n\r\n(and assigns them `null` dtype)." ]

[]

[ "Hi! Yes, that makes sense! Would you like to submit a PR to fix this?", "Created PR https://github.com/huggingface/datasets/pull/3971" ]

[ "Hi! Instead of @nateraw's `image-folder`, I suggest using the newly released `imagefolder` dataset:\r\n```python\r\n>>> from datasets import load_dataset\r\n>>> data_files={'train': ['/ssd/datasets/imagenet/pytorch/train/**'], 'validation': ['/ssd/datasets/imagenet/pytorch/val/**']}\r\n>>> ds = load_dataset('imagefolder', data_files=data_files, cache_dir='./', task='image-classification')\r\n```\r\n\r\n\r\nLet us know if that resolves the issue.", "> Hi! Instead of @nateraw's `image-folder`, I suggest using the newly released `imagefolder` dataset:\r\n> \r\n> ```python\r\n> >>> from datasets import load_dataset\r\n> >>> data_files={'train': ['/ssd/datasets/imagenet/pytorch/train/**'], 'validation': ['/ssd/datasets/imagenet/pytorch/val/**']}\r\n> >>> ds = load_dataset('imagefolder', data_files=data_files, cache_dir='./', task='image-classification')\r\n> ```\r\n> \r\n> Let us know if that resolves the issue.\r\n\r\nSorry, replied late.\r\nThanks a lot! It's worked for me. But it seems much slower than before, and now gets stuck.....\r\n\r\n```\r\n>>> from datasets import load_dataset\r\n>>> data_files={'train': ['/ssd/datasets/imagenet/pytorch/train/**'], 'validation': ['/ssd/datasets/imagenet/pytorch/val/**']}\r\n>>> ds = load_dataset('imagefolder', data_files=data_files, cache_dir='./', task='image-classification')\r\nResolving data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1281167/1281167 [00:02<00:00, 437283.97it/s]\r\nResolving data files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 50001/50001 [00:00<00:00, 89094.29it/s]\r\nUsing custom data configuration default-baebca6347576b33\r\nDownloading and preparing dataset image_folder/default to ./image_folder/default-baebca6347576b33/0.0.0/ee92df8e96c6907f3c851a987be3fd03d4b93b247e727b69a8e23ac94392a091...\r\nDownloading data files #0: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 82289.56obj/s]\r\nDownloading data files #1: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:01<00:00, 73559.11obj/s]\r\nDownloading data files #2: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 81600.46obj/s]\r\nDownloading data files #3: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:01<00:00, 79691.56obj/s]\r\nDownloading data files #4: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 82341.37obj/s]\r\nDownloading data files #5: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:01<00:00, 75784.46obj/s]\r\nDownloading data files #6: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 81466.18obj/s]\r\nDownloading data files #7: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 82320.27obj/s]\r\nDownloading data files #8: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:01<00:00, 78094.00obj/s]\r\nDownloading data files #9: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 84057.59obj/s]\r\nDownloading data files #10: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 83082.31obj/s]\r\nDownloading data files #11: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:01<00:00, 79944.21obj/s]\r\nDownloading data files #12: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 84569.77obj/s]\r\nDownloading data files #13: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 84949.63obj/s]\r\nDownloading data files #14: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80073/80073 [00:00<00:00, 80666.53obj/s]\r\nDownloading data files #15: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 80072/80072 [00:01<00:00, 76723.20obj/s]\r\n^[[Bloading data files #8: 94%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████▌ | 75061/80073 [00:00<00:00, 82609.89obj/s]\r\nDownloading data files #9: 85%|████████████████████████████████████████████████████████████████████████████████████████████████████▍ | 68120/80073 [00:00<00:00, 83868.54obj/s]\r\nDownloading data files #9: 96%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏ | 76784/80073 [00:00<00:00, 84722.34obj/s]\r\nDownloading data files #10: 75%|███████████████████████████████████████████████████████████████████████████████████████▋ | 59995/80073 [00:00<00:00, 84148.19obj/s]\r\nDownloading data files #10: 97%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████ | 77412/80073 [00:00<00:00, 85724.53obj/s]\r\nDownloading data files #11: 71%|███████████████████████████████████████████████████████████████████████████████████▎ | 57032/80073 [00:00<00:00, 79930.58obj/s]\r\nDownloading data files #11: 92%|███████████████████████████████████████████████████████████████████████████████████████████████████████████ | 73277/80073 [00:00<00:00, 78091.27obj/s]\r\nDownloading data files #12: 86%|█████████████████████████████████████████████████████████████████████████████████████████████████████ | 69125/80073 [00:00<00:00, 84723.02obj/s]\r\nDownloading data files #12: 97%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████▋ | 77803/80073 [00:00<00:00, 85351.59obj/s]\r\nDownloading data files #13: 75%|████████████████████████████████████████████████████████████████████████████████████████▏ | 60356/80073 [00:00<00:00, 84833.35obj/s]\r\nDownloading data files #13: 97%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████ | 77368/80073 [00:00<00:00, 84475.10obj/s]\r\nDownloading data files #14: 72%|████████████████████████████████████████████████████████████████████████████████████▍ | 57751/80073 [00:00<00:00, 80727.33obj/s]\r\nDownloading data files #14: 92%|████████████████████████████████████████████████████████████████████████████████████████████████████████████▏ | 74022/80073 [00:00<00:00, 78703.16obj/s]\r\nDownloading data files #15: 78%|███████████████████████████████████████████████████████████████████████████████████████████▋ | 62724/80072 [00:00<00:00, 78387.33obj/s]\r\nDownloading data files #15: 99%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████▎ | 78933/80072 [00:01<00:00, 79353.63obj/s]\r\n```", "Wait a long time, it completed. I don't know why it's so slow...", "You can pass `ignore_verifications=True` in `load_dataset` to make it fast (to skip checksum verification). I'll add this tip to the docs.", "> You can pass `ignore_verifications=True` in `load_dataset` to make it fast (to skip checksum verification). I'll add this tip to the docs.\r\n\r\nThanks！It's worked well.", "> You can pass `ignore_verifications=True` in `load_dataset` to make it fast (to skip checksum verification). I'll add this tip to the docs.\r\n\r\nI find current `load_dataset` loads ImageNet still slowly, even add `ignore_verifications=True`.\r\nFirst loading, it costs about 20 min in my servers.\r\n```\r\nreal\t19m23.023s\r\nuser\t21m18.360s\r\nsys\t7m59.080s\r\n```\r\n\r\nSecond reusing, it costs about 15 min in my servers.\r\n```\r\nreal\t15m20.735s\r\nuser\t12m22.979s\r\nsys\t5m46.960s\r\n```\r\n\r\nI think it's too much slow, is there other method to make it faster?", "And in transformers the [ViT example](https://github.com/huggingface/transformers/blob/main/examples/pytorch/image-classification/run_image_classification.py), could you make some changes ? Like the `collect_fn`\r\n```python\r\ndef collate_fn(examples):\r\n pixel_values = torch.stack([example[\"pixel_values\"] for example in examples])\r\n labels = torch.tensor([example[\"labels\"] for example in examples])\r\n return {\"pixel_values\": pixel_values, \"labels\": labels}\r\n```\r\nHow to know the keys of example?", "Loading the image files slowly, is it because the multiple processes load files at the same time?", "Could you please share the output you get after the second loading? Also, feel free to interrupt (`KeyboardInterrupt`) the process while waiting for it to end and share a traceback to show us where the process hangs. \r\n\r\n> And in transformers the [ViT example](https://github.com/huggingface/transformers/blob/main/examples/pytorch/image-classification/run_image_classification.py), could you make some changes ? Like the `collect_fn`\r\n> \r\n> ```python\r\n> def collate_fn(examples):\r\n> pixel_values = torch.stack([example[\"pixel_values\"] for example in examples])\r\n> labels = torch.tensor([example[\"labels\"] for example in examples])\r\n> return {\"pixel_values\": pixel_values, \"labels\": labels}\r\n> ```\r\n> \r\n> How to know the keys of example?\r\n\r\nWhat do you mean by \"could you make some changes\".The `ViT` script doesn't remove unused columns by default, so the keys of an example are equal to the columns of the given dataset.\r\n\r\n", "> Could you please share the output you get after the second loading? Also, feel free to interrupt (`KeyboardInterrupt`) the process while waiting for it to end and share a traceback to show us where the process hangs.\r\n> \r\n> > And in transformers the [ViT example](https://github.com/huggingface/transformers/blob/main/examples/pytorch/image-classification/run_image_classification.py), could you make some changes ? Like the `collect_fn`\r\n> > ```python\r\n> > def collate_fn(examples):\r\n> > pixel_values = torch.stack([example[\"pixel_values\"] for example in examples])\r\n> > labels = torch.tensor([example[\"labels\"] for example in examples])\r\n> > return {\"pixel_values\": pixel_values, \"labels\": labels}\r\n> > ```\r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > \r\n> > How to know the keys of example?\r\n> \r\n> What do you mean by \"could you make some changes\".The `ViT` script doesn't remove unused columns by default, so the keys of an example are equal to the columns of the given dataset.\r\n\r\nThanks for your reply!\r\n\r\n1. I did not record the second output, so I run it again. \r\n```\r\n(merak) txacs@master:/dat/txacs/test$ time python test.py \r\nResolving data files: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1281167/1281167 [00:02<00:00, 469497.89it/s]\r\nResolving data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 50001/50001 [00:00<00:00, 70123.73it/s]\r\nUsing custom data configuration default-baebca6347576b33\r\nReusing dataset image_folder (./image_folder/default-baebca6347576b33/0.0.0/ee92df8e96c6907f3c851a987be3fd03d4b93b247e727b69a8e23ac94392a091)\r\n100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:10<00:00, 5.37s/it]\r\nLoading cached processed dataset at ./image_folder/default-baebca6347576b33/0.0.0/ee92df8e96c6907f3c851a987be3fd03d4b93b247e727b69a8e23ac94392a091/cache-cd3fbdc025e03f8c.arrow\r\nLoading cached processed dataset at ./image_folder/default-baebca6347576b33/0.0.0/ee92df8e96c6907f3c851a987be3fd03d4b93b247e727b69a8e23ac94392a091/cache-b5a9de701bbdbb2b.arrow\r\nDatasetDict({\r\n train: Dataset({\r\n features: ['image', 'labels'],\r\n num_rows: 1281167\r\n })\r\n validation: Dataset({\r\n features: ['image', 'labels'],\r\n num_rows: 50000\r\n })\r\n})\r\n\r\nreal\t10m10.413s\r\nuser\t9m33.195s\r\nsys\t2m47.528s\r\n```\r\nAlthough it cost less time than the last, but still slowly.\r\n\r\n2. Sorry, forgive my poor statement. I solved it, updating to new script 'run_image_classification.py'.", "Thanks for rerunning the code to record the output. Is it the `\"Resolving data files\"` part on your machine that takes a long time to complete, or is it `\"Loading cached processed dataset at ...\"˙`? We plan to speed up the latter by splitting bigger Arrow files into smaller ones, but your dataset doesn't seem that big, so not sure if that's the issue.", "> Thanks for rerunning the code to record the output. Is it the `\"Resolving data files\"` part on your machine that takes a long time to complete, or is it `\"Loading cached processed dataset at ...\"˙`? We plan to speed up the latter by splitting bigger Arrow files into smaller ones, but your dataset doesn't seem that big, so not sure if that's the issue.\r\n\r\nSounds good! The main position, which costs long time, is from program starting to `\"Resolving data files\"`. I hope you can solve it early, thanks!", "I'm getting this problem. Script has been stuck at this part for the past 15 or so minutes:\r\n \r\n`Resolving data files: 100%|█████████████████████████████████████████| 107/107 [00:00<00:00, 472.74it/s]`\r\n\r\nI had everything working fine on an AWS EC2 node with a single GPU. Then I created an image based on the single GPU machine, and spun up a new one with 4 GPUs, so I got all of the training data ready at .cache. \r\n\r\nTurned off all checks with `verification_mode='no_checks'`. Logged in with huggingface-cli again just to be sure.\r\n\r\nInterrupting shows the code is stuck here:\r\n\r\n```\r\nFile \"/opt/conda/envs/pytorch/lib/python3.10/site-packages/datasets/arrow_reader.py\", line 200, in _read_files\r\n pa_table: Table = self._get_table_from_filename(f_dict, in_memory=in_memory)\r\n File \"/opt/conda/envs/pytorch/lib/python3.10/site-packages/datasets/arrow_reader.py\", line 336, in _get_table_from_filename\r\n table = ArrowReader.read_table(filename, in_memory=in_memory)\r\n File \"/opt/conda/envs/pytorch/lib/python3.10/site-packages/datasets/arrow_reader.py\", line 357, in read_table\r\n return table_cls.from_file(filename)\r\n File \"/opt/conda/envs/pytorch/lib/python3.10/site-packages/datasets/table.py\", line 1059, in from_file\r\n table = _memory_mapped_arrow_table_from_file(filename)\r\n File \"/opt/conda/envs/pytorch/lib/python3.10/site-packages/datasets/table.py\", line 66, in _memory_mapped_arrow_table_from_file\r\n pa_table = opened_stream.read_all()\r\n```\r\n\r\nIs it just going to take a while or am I going to run out of money? :sweat_smile: \r\n\r\nedit: ping @mariosasko " ]

[ "Hi ! It looks like an issue with pyarrow, could you try updating pyarrow and try again ?", "@albertvillanova did you find a solution to this?", "I´m getting the same problem with some files, @albertvillanova did you find a solution to this?" ]

[ "Hi @amirj, thanks for reporting.\r\n\r\nAt first sight, your issue seems a version incompatibility between your Elasticsearch client and your Elasticsearch server.\r\n\r\nFeel free to have a look at Elasticsearch client docs: https://www.elastic.co/guide/en/elasticsearch/client/python-api/current/overview.html#_compatibility\r\n> Language clients are forward compatible; meaning that clients support communicating with greater or equal minor versions of Elasticsearch. Elasticsearch language clients are only backwards compatible with default distributions and without guarantees made.", "@albertvillanova It doesn't seem a version incompatibility between the client and server, since the following code is working:\r\n\r\n```\r\nfrom elasticsearch import Elasticsearch\r\nes_client = Elasticsearch(\"http://localhost:9200\")\r\ndataset.add_elasticsearch_index(column=\"e1\", es_client=es_client, es_index_name=\"e1_index\")\r\n```", "Hi @amirj, \r\n\r\nI really think it is a version incompatibility issue between your Elasticsearch client and server:\r\n- Your Elasticsearch server NodeConfig expects a positional argument named 'scheme'\r\n- Whereas your Elasticsearch client passes only keyword arguments: `NodeConfig(**options)`\r\n\r\nMoreover:\r\n- Looking at your stack trace, I deduce you are using Elasticsearch client **\"8\"** major version:\r\n - the Elasticsearch file \"elasticsearch/_sync/client/utils.py\" was created in version \"8.0.0a1\": https://github.com/elastic/elasticsearch-py/commit/21fa13b0f03b7b27ace9e19a1f763d40bd2e2ba4\r\n - you can check your Elasticsearch client version by running this Python code:\r\n ```python\r\n import elasticsearch\r\n print(elasticsearch.__version__)\r\n ```\r\n\r\n- However, in the *Environment info*, you informed that the major version of your Eleasticsearch cluster server is **\"7\"** (\"7.10.2-SNAPSHOT\")\r\n\r\nCould you please align the Elasticsearch client/server major versions (as pointed out in Elasticsearch docs) and check if the problem persists?", "I'm closing this issue, @amirj.\r\n\r\nFeel free to re-open it if the problem persists. \r\n\r\n", "```\r\nfrom elasticsearch import Elasticsearch\r\nes = Elasticsearch([{'host': 'localhost', 'port': 9200}])\r\n```\r\n```\r\nTypeError Traceback (most recent call last)\r\n<ipython-input-8-675c6ffe5293> in <module>\r\n 1 #es = Elasticsearch([{'host':'localhost', 'port':9200}])\r\n 2 from elasticsearch import Elasticsearch\r\n----> 3 es = Elasticsearch([{'host': 'localhost', 'port': 9200}])\r\n\r\nC:\\ProgramData\\Anaconda3\\lib\\site-packages\\elasticsearch\\_sync\\client\\__init__.py in __init__(self, hosts, cloud_id, api_key, basic_auth, bearer_auth, opaque_id, headers, connections_per_node, http_compress, verify_certs, ca_certs, client_cert, client_key, ssl_assert_hostname, ssl_assert_fingerprint, ssl_version, ssl_context, ssl_show_warn, transport_class, request_timeout, node_class, node_pool_class, randomize_nodes_in_pool, node_selector_class, dead_node_backoff_factor, max_dead_node_backoff, serializer, serializers, default_mimetype, max_retries, retry_on_status, retry_on_timeout, sniff_on_start, sniff_before_requests, sniff_on_node_failure, sniff_timeout, min_delay_between_sniffing, sniffed_node_callback, meta_header, timeout, randomize_hosts, host_info_callback, sniffer_timeout, sniff_on_connection_fail, http_auth, maxsize, _transport)\r\n 310 \r\n 311 if _transport is None:\r\n--> 312 node_configs = client_node_configs(\r\n 313 hosts,\r\n 314 cloud_id=cloud_id,\r\n\r\nC:\\ProgramData\\Anaconda3\\lib\\site-packages\\elasticsearch\\_sync\\client\\utils.py in client_node_configs(hosts, cloud_id, **kwargs)\r\n 99 else:\r\n 100 assert hosts is not None\r\n--> 101 node_configs = hosts_to_node_configs(hosts)\r\n 102 \r\n 103 # Remove all values which are 'DEFAULT' to avoid overwriting actual defaults.\r\n\r\nC:\\ProgramData\\Anaconda3\\lib\\site-packages\\elasticsearch\\_sync\\client\\utils.py in hosts_to_node_configs(hosts)\r\n 142 \r\n 143 elif isinstance(host, Mapping):\r\n--> 144 node_configs.append(host_mapping_to_node_config(host))\r\n 145 else:\r\n 146 raise ValueError(\r\n\r\nC:\\ProgramData\\Anaconda3\\lib\\site-packages\\elasticsearch\\_sync\\client\\utils.py in host_mapping_to_node_config(host)\r\n 209 options[\"path_prefix\"] = options.pop(\"url_prefix\")\r\n 210 \r\n--> 211 return NodeConfig(**options) # type: ignore\r\n 212 \r\n 213 \r\n\r\nTypeError: __init__() missing 1 required positional argument: 'scheme'\r\n```", "I am facing the same issue, and version is same for the both i.e(8.1.3)", "@raj713335, thanks for reporting.\r\n\r\nPlease note that in your code example, you are not using our `datasets` library. \r\n\r\nThus, I think you should report that issue to `elasticsearch` library: https://github.com/elastic/elasticsearch-py\r\n\r\n", "it is simple hack which shock you just replace https to http in scheme\r\n\r\n**In My Case:** ->\r\n\r\n`es = Elasticsearch([{'host': 'localhost', 'port': 9200, \"scheme\": \"http\"}])\r\n if es.ping():\r\n print('Connected to ES!')\r\n else:\r\n print('Could not connect!')\r\n sys.exit()`" ]

[ "Hi @MatanBenChorin, thanks for reporting.\r\n\r\nPlease, take into account that the preview may take some time until it properly renders (we are working to reduce this time).\r\n\r\nMaybe @severo can give more details on this.", "Hi, \r\nThank you", "Thanks for reporting. We are looking at it and will give updates here.", "I imagine the dataset has been moved to https://huggingface.co/datasets/tdklab/Hebrew_Squad_v1, which still has an issue:\r\n\r\n```\r\nServer Error\r\n\r\nStatus code: 400\r\nException: NameError\r\nMessage: name 'HebrewSquad' is not defined\r\n```", "The issue is not related to the dataset viewer but to the loading script (cc @albertvillanova @lhoestq @mariosasko)\r\n\r\n```python\r\n>>> import datasets as ds\r\n>>> hf_token = \"hf_...\" # <- required because the dataset is gated\r\n>>> d = ds.load_dataset('tdklab/Hebrew_Squad_v1', use_auth_token=hf_token)\r\n...\r\nNameError: name 'HebrewSquad' is not defined\r\n```", "Yes indeed there is an error in [Hebrew_Squad_v1.py:L40](https://huggingface.co/datasets/tdklab/Hebrew_Squad_v1/blob/main/Hebrew_Squad_v1.py#L40)\r\n\r\nHere is the fix @MatanBenChorin :\r\n\r\n```diff\r\n- HebrewSquad(\r\n+ HebrewSquadConfig(\r\n```" ]

[ "Can you assign this task to me? @nreimers @mariosasko ", "Hi! Sure! Let us know if you need any pointers." ]

[ "Hi, @ravindra-ut.\r\n\r\nI'm sorry but I can't reproduce your problem:\r\n```python\r\nIn [1]: from datasets import load_dataset\r\n\r\nIn [2]: ds = load_dataset(\"glue\", \"sst2\")\r\nDownloading builder script: 28.8kB [00:00, 11.6MB/s] \r\nDownloading metadata: 28.7kB [00:00, 12.9MB/s] \r\nDownloading and preparing dataset glue/sst2 (download: 7.09 MiB, generated: 4.81 MiB, post-processed: Unknown size, total: 11.90 MiB) to .../.cache/huggingface/datasets/glue/sst2/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad...\r\nDownloading data: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 7.44M/7.44M [00:01<00:00, 5.82MB/s]\r\nDataset glue downloaded and prepared to .../.cache/huggingface/datasets/glue/sst2/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad. Subsequent calls will reuse this data. \r\n100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 895.96it/s]\r\n\r\nIn [3]: ds\r\nOut[2]: \r\nDatasetDict({\r\n train: Dataset({\r\n features: ['sentence', 'label', 'idx'],\r\n num_rows: 67349\r\n })\r\n validation: Dataset({\r\n features: ['sentence', 'label', 'idx'],\r\n num_rows: 872\r\n })\r\n test: Dataset({\r\n features: ['sentence', 'label', 'idx'],\r\n num_rows: 1821\r\n })\r\n})\r\n``` \r\n\r\nMoreover, I see in your traceback that your error was for an URL at https://firebasestorage.googleapis.com\r\nHowever, the URLs were updated on Sep 16, 2020 (`datasets` version 1.0.2) to https://dl.fbaipublicfiles.com: https://github.com/huggingface/datasets/commit/2f03041a21c03abaececb911760c3fe4f420c229\r\n\r\nCould you please try to update `datasets`\r\n```shell\r\npip install -U datasets\r\n```\r\nand then force redownload\r\n```python\r\nds = load_dataset(\"glue\", \"sst2\", download_mode=\"force_redownload\")\r\n```\r\nto update the cache?\r\n\r\nPlease, feel free to reopen this issue if the problem persists." ]

[ "Thanks for reporting this second issue as well. We definitely want to make streaming datasets fully working in a distributed setup and with the best performance. Right now it only supports single process.\r\n\r\nIn this issue it seems that the streaming capabilities that we offer to dataset builders are not transferred to the forked process (so it fails to open remote files and start streaming data from them). In particular `open` is supposed to be mocked by our `xopen` function that is an extended open that supports remote files. Let me try to fix this" ]

[ "Hi, thanks for reporting. This could be related to https://github.com/huggingface/datasets/issues/3148 too\r\n\r\nWe should definitely make `TorchIterableDataset` picklable by moving it in the main code instead of inside a function. If you'd like to contribute, feel free to open a Pull Request :)\r\n\r\nI'm also taking a look at your second issue, which is more technical" ]

[ "Hi @XingxingZhang, \r\n\r\nWe have already fixed this. You should update `datasets` version to at least 1.18.4:\r\n```shell\r\npip install -U datasets\r\n```\r\nAnd then force the redownload:\r\n```python\r\nload_dataset(\"...\", download_mode=\"force_redownload\")\r\n```\r\n\r\nDuplicate of:\r\n- #3773", "thanks @albertvillanova . by upgrading to 1.18.4 and using `load_dataset(\"...\", download_mode=\"force_redownload\")` fixed \r\n the bug.\r\n\r\nusing the following as you suggested in another thread can also fixed the bug\r\n```\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\n", "The latter solution (installing from GitHub) was proposed because the fix was not released yet. But last week we made the 1.18.4 patch release (with the fix), so no longer necessary to install from GitHub.\r\n\r\nYou can now install from PyPI, as usual:\r\n```shell\r\npip install -U datasets\r\n```\r\n" ]

[ "Hi @XingxingZhang, thanks for reporting.\r\n\r\nThis was due to a change in Google Drive service:\r\n- #3786 \r\n\r\nWe have already fixed it:\r\n- #3787\r\n\r\nYou should update `datasets` version to at least 1.18.4:\r\n```shell\r\npip install -U datasets\r\n```\r\nAnd then force the redownload:\r\n```python\r\nload_dataset(\"...\", download_mode=\"force_redownload\")\r\n```", "thanks @albertvillanova ", "@albertvillanova \r\nYOU Said: pip install git+ https://github.com/huggingface/datasets.git Then set: dataset=load_dataset (\"multinews\", download_mode=\"force-redownload\"). I changed the ’datautils‘ file according to this setting: traindata=load_dataset (path='wikitext ', name='wikitext-2-raw v1', split='train ', download_mode=\"force-redownload\")\r\nTestdata=load_dataset (path='wikitext ', name='wikitext-2-raw v1', split='test ', download_mode=\"force-redownload\")\r\nthen the bug is\r\n\r\n\r\nI have tried both versions\r\n\r\n" ]

[ "Thanks for reporting @qqaatw.\r\n\r\n@mishig25 @sgugger do you think this can be tweaked in the new doc framework?\r\n- From: https://github.com/huggingface/datasets/blob/v2.0.0/\r\n- To: https://github.com/huggingface/datasets/blob/2.0.0/", "@qqaatw thanks a lot for notifying about this issue!\r\n\r\nin comparison, transformers tags start with `v` like [this one](https://github.com/huggingface/transformers/blob/v4.17.0/src/transformers/models/bert/configuration_bert.py#L54).\r\n\r\nTherefore, we have to do one of 2 options below:\r\n1. Make necessary changes on doc-builder side\r\nOR\r\n2. Make [datasets tags](https://github.com/huggingface/datasets/tags) start with `v`, just like [transformers](https://github.com/huggingface/transformers/tags) (so that tag naming can be consistent amongst hf repos)\r\n\r\nI'll let you decide @albertvillanova @lhoestq @sgugger ", "I think option 2 is the easiest and would provide harmony in the HF ecosystem but we can also add a doc config parameter to decide whether the default version has a v or not if `datasets` folks prefer their tags without a v :-)", "For me it is OK to conform to the rest of libraries and tag/release with a preceding \"v\", rather than adding an extra argument to the doc builder just for `datasets`.\r\n\r\nLet me know if it is also OK for you @lhoestq. ", "https://github.com/huggingface/doc-build/commit/f41c1e8ff900724213af4c75d287d8b61ecf6141\r\n\r\nhotfix so that `datasets` docs source button works correctly on hf.co/docs/datasets", "We could add a tag for each release without a 'v' but it could be confusing on github to see both tags `v2.0.0` and `2.0.0` IMO (not sure if many users check them though). Removing the tags without 'v' would break our versioning for github datasets: the library looks for dataset scripts at the URLs like `https://raw.githubusercontent.com/huggingface/datasets/{revision}/datasets/{path}/{name}` where `revision` is equal to `datasets.__version__` (which doesn't start with a 'v') for all released versions of `datasets`.\r\n\r\nI think we could just have a parameter for the documentation - and having different URLs schemes for the source links that the users don't even see (they simply click on a button) is probably fine", "This is done in #3943 to go along with [doc-builder#146](https://github.com/huggingface/doc-builder/pull/146).\r\n\r\nNote that this will only work for future versions, so once those two are merged, the actual v2.0.0 doc should be fixed. The easiest is to cherry-pick this commit on the v2.0.0 release branch (or on a new branch created from the 2.0.0 tag, with a name that triggers the doc building job, for instance v2.0.0-release)", "Thanks for fixing @sgugger." ]

[ "Thanks for contacting @Eytan-S.\r\n\r\nI think @lvwerra could better answer this. ", "That seems to be an oversight - I originally planned to include them in the dataset and for some reason they were in the list of languages but not in the query. Since there is an issue with the deduplication step I'll rerun the pipeline anyway and will double check the query.\r\n\r\nThanks for reporting this @Eytan-S!", "Can confirm that the two languages are indeed missing from the dataset. Here are the file counts per language:\r\n```Python\r\n{'Assembly': 82847,\r\n 'Batchfile': 236755,\r\n 'C': 14127969,\r\n 'C#': 6793439,\r\n 'C++': 7368473,\r\n 'CMake': 175076,\r\n 'CSS': 1733625,\r\n 'Dockerfile': 331966,\r\n 'FORTRAN': 141963,\r\n 'GO': 2259363,\r\n 'Haskell': 340521,\r\n 'HTML': 11165464,\r\n 'Java': 19515696,\r\n 'JavaScript': 11829024,\r\n 'Julia': 58177,\r\n 'Lua': 576279,\r\n 'Makefile': 679338,\r\n 'Markdown': 8454049,\r\n 'PHP': 11181930,\r\n 'Perl': 497490,\r\n 'PowerShell': 136827,\r\n 'Python': 7203553,\r\n 'Ruby': 4479767,\r\n 'Rust': 321765,\r\n 'SQL': 655657,\r\n 'Scala': 0,\r\n 'Shell': 1382786,\r\n 'TypeScript': 0,\r\n 'TeX': 250764,\r\n 'Visual Basic': 155371}\r\n ```", "@Eytan-S check out v1.1 of the `github-code` dataset where issue should be fixed:\r\n\r\n| | Language |File Count| Size (GB)|\r\n|---:|:-------------|---------:|-------:|\r\n| 0 | Java | 19548190 | 107.7 |\r\n| 1 | C | 14143113 | 183.83 |\r\n| 2 | JavaScript | 11839883 | 87.82 |\r\n| 3 | HTML | 11178557 | 118.12 |\r\n| 4 | PHP | 11177610 | 61.41 |\r\n| 5 | Markdown | 8464626 | 23.09 |\r\n| 6 | C++ | 7380520 | 87.73 |\r\n| 7 | Python | 7226626 | 52.03 |\r\n| 8 | C# | 6811652 | 36.83 |\r\n| 9 | Ruby | 4473331 | 10.95 |\r\n| 10 | GO | 2265436 | 19.28 |\r\n| 11 | TypeScript | 1940406 | 24.59 |\r\n| 12 | CSS | 1734406 | 22.67 |\r\n| 13 | Shell | 1385648 | 3.01 |\r\n| 14 | Scala | 835755 | 3.87 |\r\n| 15 | Makefile | 679430 | 2.92 |\r\n| 16 | SQL | 656671 | 5.67 |\r\n| 17 | Lua | 578554 | 2.81 |\r\n| 18 | Perl | 497949 | 4.7 |\r\n| 19 | Dockerfile | 366505 | 0.71 |\r\n| 20 | Haskell | 340623 | 1.85 |\r\n| 21 | Rust | 322431 | 2.68 |\r\n| 22 | TeX | 251015 | 2.15 |\r\n| 23 | Batchfile | 236945 | 0.7 |\r\n| 24 | CMake | 175282 | 0.54 |\r\n| 25 | Visual Basic | 155652 | 1.91 |\r\n| 26 | FORTRAN | 142038 | 1.62 |\r\n| 27 | PowerShell | 136846 | 0.69 |\r\n| 28 | Assembly | 82905 | 0.78 |\r\n| 29 | Julia | 58317 | 0.29 |", "Thanks @lvwerra. " ]

[ "Hi @caush, thanks for reporting:\r\n\r\nIn order to load local CSV files, you can use our \"csv\" loading script: https://huggingface.co/docs/datasets/loading#csv\r\n```python\r\ndataset = load_dataset(\"csv\", data_files=[\"data/file1.csv\", \"data/file2.csv\"])\r\n```\r\nOR:\r\n```python\r\ndataset = load_dataset(\"csv\", data_dir=\"data\")\r\n```\r\n\r\nAlternatively, you may also use:\r\n```python\r\ndataset = load_dataset(\"data\")" ]

[ "Hi @Ierezell, thanks for reporting.\r\n\r\nI'm making a PR to suppress those logs from the terminal. " ]

[ "Hi @Arij-Aladel,\r\n\r\nYou are using a very old version of our library `datasets`: 1.8.0\r\nCurrent version is 2.0.0 (and the previous one was 1.18.4)\r\n\r\nPlease, try to update `datasets` library and check if the problem persists:\r\n```shell\r\n/env/bin/pip install -U datasets", "The problem I can no I have build my project on this version and old version on transformers. I have preprocessed the data again to use it. Thank for your reply" ]

[ "You are likely trying to get the `features` from a `DatasetDict`, a dictionary containing `Datasets`. You probably first want to index into a particular split from your `DatasetDict` i.e. `dataset['train'].features`. \r\n\r\nFor example \r\n\r\n```python \r\nds = load_dataset('mnist')\r\nds.features\r\n```\r\nReturns \r\n```python\r\n---------------------------------------------------------------------------\r\n\r\nAttributeError Traceback (most recent call last)\r\n\r\n[<ipython-input-39-791c1f9df6c2>](https://localhost:8080/#) in <module>()\r\n----> 1 ds.features\r\n\r\nAttributeError: 'DatasetDict' object has no attribute 'features'\r\n```\r\n\r\nIf we look at the dataset variable, we see it is a `DatasetDict`:\r\n\r\n```python \r\nprint(ds)\r\nDatasetDict({\r\n train: Dataset({\r\n features: ['image', 'label'],\r\n num_rows: 60000\r\n })\r\n test: Dataset({\r\n features: ['image', 'label'],\r\n num_rows: 10000\r\n })\r\n})\r\n```\r\n\r\nWe can grab the features from a split by indexing into `train`:\r\n```python\r\nds['train'].features\r\n{'image': Image(decode=True, id=None),\r\n 'label': ClassLabel(num_classes=10, names=['0', '1', '2', '3', '4', '5', '6', '7', '8', '9'], id=None)}\r\n```\r\n\r\nHope that helps ", "Yes, Thanks for that clarification," ]

[ "Hi @willowdong! These issues were fixed on master. We will have a new release of `datasets` later today. In the meantime, you can avoid these issues by installing `datasets` from master as follows:\r\n```bash\r\npip install git+https://github.com/huggingface/datasets.git\r\n```", "You should force redownload:\r\n```python\r\ndataset = load_dataset(\"multi_news\", download_mode=\"force_redownload\")\r\ndataset_2 = load_dataset(\"reddit_tifu\", \"long\", download_mode=\"force_redownload\")", "Fixed by:\r\n- #3787 \r\n- #3843" ]

[ "Hi ! It could an issue with torchaudio, which version of torchaudio are you using ? Can you also try updating `datasets` to 2.0.0 and see if it works ?", "I _might_ have a similar issue. I'm trying to use the librispeech_asr dataset and read it with soundfile.\r\n\r\n```python\r\nfrom datasets import load_dataset, load_metric\r\nfrom transformers import Speech2TextForConditionalGeneration, Speech2TextProcessor\r\nimport soundfile as sf\r\n\r\nlibrispeech_eval = load_dataset(\"librispeech_asr\", \"clean\", split=\"test\") # change to \"other\" for other test dataset\r\nwer = load_metric(\"wer\")\r\n\r\nmodel = Speech2TextForConditionalGeneration.from_pretrained(\"facebook/s2t-small-librispeech-asr\").to(\"cuda\")\r\nprocessor = Speech2TextProcessor.from_pretrained(\"facebook/s2t-small-librispeech-asr\", do_upper_case=True)\r\n\r\ndef map_to_array(batch):\r\n speech, _ = sf.read(batch[\"file\"])\r\n batch[\"speech\"] = speech\r\n return batch\r\n\r\nlibrispeech_eval = librispeech_eval.map(map_to_array)\r\n\r\ndef map_to_pred(batch):\r\n features = processor(batch[\"speech\"], sampling_rate=16000, padding=True, return_tensors=\"pt\")\r\n input_features = features.input_features.to(\"cuda\")\r\n attention_mask = features.attention_mask.to(\"cuda\")\r\n\r\n gen_tokens = model.generate(input_ids=input_features, attention_mask=attention_mask)\r\n batch[\"transcription\"] = processor.batch_decode(gen_tokens, skip_special_tokens=True)\r\n return batch\r\n\r\nresult = librispeech_eval.map(map_to_pred, batched=True, batch_size=8, remove_columns=[\"speech\"])\r\n\r\nprint(\"WER:\", wer(predictions=result[\"transcription\"], references=result[\"text\"]))\r\n```\r\n\r\nThe code is taken directly from \"https://huggingface.co/facebook/s2t-small-librispeech-asr\".\r\n\r\nThe short error code is \"RuntimeError: Error opening '6930-75918-0000.flac': System error.\" (it can't find the first file), and I agree, I can't find the file either. The dataset has downloaded correctly (it says), but on the location, there are only \".arrow\" files, no \".flac\" files.\r\n\r\n**Error message:**\r\n\r\n```python\r\nRuntimeError Traceback (most recent call last)\r\nInput In [15], in <cell line: 16>()\r\n 13 batch[\"speech\"] = speech\r\n 14 return batch\r\n---> 16 librispeech_eval = librispeech_eval.map(map_to_array)\r\n 18 def map_to_pred(batch):\r\n 19 features = processor(batch[\"speech\"], sampling_rate=16000, padding=True, return_tensors=\"pt\")\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\datasets\\arrow_dataset.py:1953, in Dataset.map(self, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint, desc)\r\n 1950 disable_tqdm = not logging.is_progress_bar_enabled()\r\n 1952 if num_proc is None or num_proc == 1:\r\n-> 1953 return self._map_single(\r\n 1954 function=function,\r\n 1955 with_indices=with_indices,\r\n 1956 with_rank=with_rank,\r\n 1957 input_columns=input_columns,\r\n 1958 batched=batched,\r\n 1959 batch_size=batch_size,\r\n 1960 drop_last_batch=drop_last_batch,\r\n 1961 remove_columns=remove_columns,\r\n 1962 keep_in_memory=keep_in_memory,\r\n 1963 load_from_cache_file=load_from_cache_file,\r\n 1964 cache_file_name=cache_file_name,\r\n 1965 writer_batch_size=writer_batch_size,\r\n 1966 features=features,\r\n 1967 disable_nullable=disable_nullable,\r\n 1968 fn_kwargs=fn_kwargs,\r\n 1969 new_fingerprint=new_fingerprint,\r\n 1970 disable_tqdm=disable_tqdm,\r\n 1971 desc=desc,\r\n 1972 )\r\n 1973 else:\r\n 1975 def format_cache_file_name(cache_file_name, rank):\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\datasets\\arrow_dataset.py:519, in transmit_tasks.<locals>.wrapper(*args, **kwargs)\r\n 517 self: \"Dataset\" = kwargs.pop(\"self\")\r\n 518 # apply actual function\r\n--> 519 out: Union[\"Dataset\", \"DatasetDict\"] = func(self, *args, **kwargs)\r\n 520 datasets: List[\"Dataset\"] = list(out.values()) if isinstance(out, dict) else [out]\r\n 521 for dataset in datasets:\r\n 522 # Remove task templates if a column mapping of the template is no longer valid\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\datasets\\arrow_dataset.py:486, in transmit_format.<locals>.wrapper(*args, **kwargs)\r\n 479 self_format = {\r\n 480 \"type\": self._format_type,\r\n 481 \"format_kwargs\": self._format_kwargs,\r\n 482 \"columns\": self._format_columns,\r\n 483 \"output_all_columns\": self._output_all_columns,\r\n 484 }\r\n 485 # apply actual function\r\n--> 486 out: Union[\"Dataset\", \"DatasetDict\"] = func(self, *args, **kwargs)\r\n 487 datasets: List[\"Dataset\"] = list(out.values()) if isinstance(out, dict) else [out]\r\n 488 # re-apply format to the output\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\datasets\\fingerprint.py:458, in fingerprint_transform.<locals>._fingerprint.<locals>.wrapper(*args, **kwargs)\r\n 452 kwargs[fingerprint_name] = update_fingerprint(\r\n 453 self._fingerprint, transform, kwargs_for_fingerprint\r\n 454 )\r\n 456 # Call actual function\r\n--> 458 out = func(self, *args, **kwargs)\r\n 460 # Update fingerprint of in-place transforms + update in-place history of transforms\r\n 462 if inplace: # update after calling func so that the fingerprint doesn't change if the function fails\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\datasets\\arrow_dataset.py:2318, in Dataset._map_single(self, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, new_fingerprint, rank, offset, disable_tqdm, desc, cache_only)\r\n 2316 if not batched:\r\n 2317 for i, example in enumerate(pbar):\r\n-> 2318 example = apply_function_on_filtered_inputs(example, i, offset=offset)\r\n 2319 if update_data:\r\n 2320 if i == 0:\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\datasets\\arrow_dataset.py:2218, in Dataset._map_single.<locals>.apply_function_on_filtered_inputs(inputs, indices, check_same_num_examples, offset)\r\n 2216 if with_rank:\r\n 2217 additional_args += (rank,)\r\n-> 2218 processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)\r\n 2219 if update_data is None:\r\n 2220 # Check if the function returns updated examples\r\n 2221 update_data = isinstance(processed_inputs, (Mapping, pa.Table))\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\datasets\\arrow_dataset.py:1913, in Dataset.map.<locals>.decorate.<locals>.decorated(item, *args, **kwargs)\r\n 1909 decorated_item = (\r\n 1910 Example(item, features=self.features) if not batched else Batch(item, features=self.features)\r\n 1911 )\r\n 1912 # Use the LazyDict internally, while mapping the function\r\n-> 1913 result = f(decorated_item, *args, **kwargs)\r\n 1914 # Return a standard dict\r\n 1915 return result.data if isinstance(result, LazyDict) else result\r\n\r\nInput In [15], in map_to_array(batch)\r\n 11 def map_to_array(batch):\r\n---> 12 speech, _ = sf.read(batch[\"file\"])\r\n 13 batch[\"speech\"] = speech\r\n 14 return batch\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\soundfile.py:256, in read(file, frames, start, stop, dtype, always_2d, fill_value, out, samplerate, channels, format, subtype, endian, closefd)\r\n 170 def read(file, frames=-1, start=0, stop=None, dtype='float64', always_2d=False,\r\n 171 fill_value=None, out=None, samplerate=None, channels=None,\r\n 172 format=None, subtype=None, endian=None, closefd=True):\r\n 173 \"\"\"Provide audio data from a sound file as NumPy array.\r\n 174 \r\n 175 By default, the whole file is read from the beginning, but the\r\n (...)\r\n 254 \r\n 255 \"\"\"\r\n--> 256 with SoundFile(file, 'r', samplerate, channels,\r\n 257 subtype, endian, format, closefd) as f:\r\n 258 frames = f._prepare_read(start, stop, frames)\r\n 259 data = f.read(frames, dtype, always_2d, fill_value, out)\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\soundfile.py:629, in SoundFile.__init__(self, file, mode, samplerate, channels, subtype, endian, format, closefd)\r\n 626 self._mode = mode\r\n 627 self._info = _create_info_struct(file, mode, samplerate, channels,\r\n 628 format, subtype, endian)\r\n--> 629 self._file = self._open(file, mode_int, closefd)\r\n 630 if set(mode).issuperset('r+') and self.seekable():\r\n 631 # Move write position to 0 (like in Python file objects)\r\n 632 self.seek(0)\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\soundfile.py:1183, in SoundFile._open(self, file, mode_int, closefd)\r\n 1181 else:\r\n 1182 raise TypeError(\"Invalid file: {0!r}\".format(self.name))\r\n-> 1183 _error_check(_snd.sf_error(file_ptr),\r\n 1184 \"Error opening {0!r}: \".format(self.name))\r\n 1185 if mode_int == _snd.SFM_WRITE:\r\n 1186 # Due to a bug in libsndfile version <= 1.0.25, frames != 0\r\n 1187 # when opening a named pipe in SFM_WRITE mode.\r\n 1188 # See http://github.com/erikd/libsndfile/issues/77.\r\n 1189 self._info.frames = 0\r\n\r\nFile C:\\ProgramData\\Miniconda3\\envs\\noise_cancel\\lib\\site-packages\\soundfile.py:1357, in _error_check(err, prefix)\r\n 1355 if err != 0:\r\n 1356 err_str = _snd.sf_error_number(err)\r\n-> 1357 raise RuntimeError(prefix + _ffi.string(err_str).decode('utf-8', 'replace'))\r\n\r\nRuntimeError: Error opening '6930-75918-0000.flac': System error.\r\n```\r\n\r\n**Package versions:**\r\n```python\r\npython: 3.9\r\ntransformers: 4.17.0\r\ndatasets: 2.0.0\r\nSoundFile: 0.10.3.post1\r\n```\r\n", "Hi ! In `datasets` 2.0 can access the audio array with `librispeech_eval[0][\"audio\"][\"array\"]` already, no need to use `map_to_array`. See our documentation on [how to process audio data](https://huggingface.co/docs/datasets/audio_process) :)\r\n\r\ncc @patrickvonplaten we will need to update the readme at [facebook/s2t-small-librispeech-asr](https://huggingface.co/facebook/s2t-small-librispeech-asr) as well as https://huggingface.co/docs/transformers/model_doc/speech_to_text", "Thanks!\r\n\r\nAnd sorry for posting this problem in what turned on to be an unrelated thread.\r\n\r\nI rewrote the code, and the model works. The WER is 0.137 however, so I'm not sure if I have missed a step. I will look further into that at a later point. The transcriptions look good through manual inspection.\r\n\r\nThe rewritten code:\r\n```python\r\nfrom datasets import load_dataset, load_metric\r\nfrom transformers import Speech2TextForConditionalGeneration, Speech2TextProcessor, Wav2Vec2Processor\r\n\r\nlibrispeech_eval = load_dataset(\"librispeech_asr\", \"clean\", split=\"test\") # change to \"other\" for other test dataset\r\nwer = load_metric(\"wer\")\r\n\r\nmodel = Speech2TextForConditionalGeneration.from_pretrained(\"facebook/s2t-small-librispeech-asr\").to(\"cuda\")\r\nprocessor = Speech2TextProcessor.from_pretrained(\"facebook/s2t-small-librispeech-asr\", do_upper_case=True)\r\n\r\ndef map_to_pred(batch):\r\n audio = batch[\"audio\"]\r\n features = processor(audio[\"array\"], sampling_rate=audio[\"sampling_rate\"], padding=True, return_tensors=\"pt\")\r\n input_features = features.input_features.to(\"cuda\")\r\n attention_mask = features.attention_mask.to(\"cuda\")\r\n\r\n gen_tokens = model.generate(input_features=input_features, attention_mask=attention_mask)\r\n batch[\"transcription\"] = processor.batch_decode(gen_tokens, skip_special_tokens=True)\r\n return batch\r\n\r\nresult = librispeech_eval.map(map_to_pred)#, batched=True, batch_size=8)\r\n\r\nprint(\"WER:\", wer.compute(predictions=result[\"transcription\"], references=result[\"text\"]))\r\n```", "I think the issue comes from the fact that you set `batched=False` while `map_to_pred` still returns a list of strings for \"transcription\". You can fix it by adding `[0]` at the end of this line to get the string:\r\n```python\r\nbatch[\"transcription\"] = processor.batch_decode(gen_tokens, skip_special_tokens=True)[0]\r\n```", "Updating as many model cards now as I can find", "https://github.com/huggingface/transformers/pull/16611", "We no longer use `torchaudio` for decoding MP3 files, and the problem with model cards has been addressed, so I'm closing this issue." ]

[ "Hi @kolk, thanks for reporting.\r\n\r\nIndeed, Google Drive service recently changed their service and we had to add a fix to our library to cope with that change:\r\n- #3787 \r\n\r\nWe just made patch release last week: 1.18.4 https://github.com/huggingface/datasets/releases/tag/1.18.4\r\n\r\nPlease, feel free to update your local `datasets` version, so that you get the fix:\r\n```shell\r\npip install -U datasets\r\n```" ]

[ "Thanks for reporting, @omarespejel.\r\n\r\nI'm sorry but I can't reproduce the issue: the loading of the dataset works perfecto for me and I can reach the data URL: https://data.deepai.org/conll2003.zip\r\n\r\nMight it be due to a temporary problem in the data owner site (https://data.deepai.org/) that is fixed now?\r\nCould you please try loading the dataset again and tell if the problem persists?", "@omarespejel I'm closing this issue. Feel free to reopen it if the problem persists.", "getting same issue. Can't find any solution.", "I am getting the same issue. I use google colab with CPU.\r\nThe code I used is exactly the same as described above.\r\n```\r\nfrom datasets import load_dataset\r\ndataset = load_dataset(\"conll2003\")\r\n```\r\n\r\nThe produced error:\r\n\r\n\r\nNote: This error is different from what was initially described in this thread. This is because I use CPU. When I use GPU I reproduce the same initial error of the thread.\r\n\r\nMoreover, I receive the following warning:\r\n```\r\nWARNING:urllib3.connection:Certificate did not match expected hostname: data.deepai.org. Certificate: {'subject': ((('commonName', '*.b-cdn.net'),),), 'issuer': ((('countryName', 'GB'),), (('stateOrProvinceName', 'Greater Manchester'),), (('localityName', 'Salford'),), (('organizationName', 'Sectigo Limited'),), (('commonName', 'Sectigo RSA Domain Validation Secure Server CA'),)), 'version': 3, 'serialNumber': 'DDED48B13E1EA03983E833AB2C35EF07', 'notBefore': 'Nov 7 00:00:00 2022 GMT', 'notAfter': 'Nov 11 23:59:59 2023 GMT', 'subjectAltName': (('DNS', '*.b-cdn.net'), ('DNS', 'b-cdn.net')), 'OCSP': ('http://ocsp.sectigo.com/',), 'caIssuers': ('http://crt.sectigo.com/SectigoRSADomainValidationSecureServerCA.crt',)}\r\nDownloading and preparing dataset conll2003/conll2003 to /root/.cache/huggingface/datasets/conll2003/conll2003/1.0.0/9a4d16a94f8674ba3466315300359b0acd891b68b6c8743ddf60b9c702adce98...\r\nWARNING:urllib3.connection:Certificate did not match expected hostname: data.deepai.org. Certificate: {'subject': ((('commonName', '*.b-cdn.net'),),), 'issuer': ((('countryName', 'GB'),), (('stateOrProvinceName', 'Greater Manchester'),), (('localityName', 'Salford'),), (('organizationName', 'Sectigo Limited'),), (('commonName', 'Sectigo RSA Domain Validation Secure Server CA'),)), 'version': 3, 'serialNumber': 'DDED48B13E1EA03983E833AB2C35EF07', 'notBefore': 'Nov 7 00:00:00 2022 GMT', 'notAfter': 'Nov 11 23:59:59 2023 GMT', 'subjectAltName': (('DNS', '*.b-cdn.net'), ('DNS', 'b-cdn.net')), 'OCSP': ('http://ocsp.sectigo.com/',), 'caIssuers': ('http://crt.sectigo.com/SectigoRSADomainValidationSecureServerCA.crt',)}\r\n```\r\n" ]

[ "Update: `\"python3 -c \"from from datasets import Dataset, DatasetDict\"` works, but not if I import without the `python3 -c`", "Hi @arunasank, thanks for reporting.\r\n\r\nIt seems that this can be caused because you are using an old version of `fsspec`: the reason why it works if you run `python3` seems to be that `python3` runs in a Python virtual env (with an updated version of `fsspec`); whereas the error arises when you run the import from other Python virtual env (with an old version of `fsspec`).\r\n\r\nIn order to fix this, you should update `fsspec` from within the \"problematic\" Python virtual env:\r\n```\r\npip install -U \"fsspec[http]>=2021.05.0\"", "I'm closing this issue, @arunasank.\r\n\r\nFeel free to re-open it if the problem persists. ", "from lightgbm import LGBMModel,LGBMClassifier, plot_importance\r\nafter importing lib getting (partially initialized module 'fsspec' has no attribute 'utils' (most likely due to a circular import) error, can help me", "@deepakmahtha I think you are not using `datasets`: this is the GitHub repository of Hugging Face Datasets.\r\n\r\nIf you are using `lightgbm`, you should report the issue to their repository instead.\r\n\r\nAnyway, we have proposed a possible fix just in a comment above: to update fsspec.\r\nhttps://github.com/huggingface/datasets/issues/3902#issuecomment-1066517824" ]

[ "It seems to have been fixed:\r\n\r\n<img width=\"1534\" alt=\"Capture d’écran 2022-04-12 à 14 10 07\" src=\"https://user-images.githubusercontent.com/1676121/162959599-6b7fef7c-8411-4e03-8f00-90040a658079.png\">\r\n" ]

[ "reported by @abidlabs ", "related to https://github.com/huggingface/datasets/pull/3843?", "`datasets` 1.18.4 fixes the issue when you load the dataset with `load_dataset`.\r\n\r\nWhen loading in streaming mode, the fix is indeed on https://github.com/huggingface/datasets/pull/3843 which will be merged soon :)", "That is. The PR #3843 was just opened a bit later we had made our 1.18.4 patch release...\r\nOnce merged, that will fix this issue. ", "OK. Should fix the viewer for 50 datasets\r\n\r\n<img width=\"148\" alt=\"Capture d’écran 2022-03-14 à 11 51 02\" src=\"https://user-images.githubusercontent.com/1676121/158157853-6c544a47-2d6d-4ac4-964a-6f10951ec36b.png\">\r\n" ]

[ "Hi ! A pull request is open to fix the dataset, we'll release a patch soon with a new release of `datasets` :)" ]

[ "#self-assign", "@alvarobartt I've assigned you the issue since I'm not actively working on it.", "Cool thanks @mariosasko I'll try to fix it in the upcoming days, thanks!", "@lhoestq so in order to address what’s not completed in this issue, do you think it makes sense to add a param `features` to `IterableDataset.map` so that the output features right after the `map` are defined there? ", "Yes that would be ideal IMO, thanks again for the help :)", "@lhoestq cool then if you agree I can work on that! I’ll also update the docs accordingly once done, thanks!", "I've already started with a PR as a draft @lhoestq, should we also try to look for a way to explicitly request pre-fetching right after a map operation is applied, so that the features are inferred if the user says explicitly so? Thanks!", "> should we also try to look for a way to explicitly request pre-fetching right after a map operation is applied, so that the features are inferred if the user says explicitly so?\r\n\r\nRight now one can use `ds = ds._resolve_features()` do to so. It can be used after `map` or `load_dataset` if the features are not known. Maybe we can make this method public ?" ]

[ "Hi @zhaowei-wang98, thanks for reporting.\r\n\r\nWe are fixing it... " ]

[ "Even this from docs throw same error\r\n```\r\ndataset = load_dataset(\"imagefolder\", data_files=\"https://download.microsoft.com/download/3/E/1/3E1C3F21-ECDB-4869-8368-6DEBA77B919F/kagglecatsanddogs_3367a.zip\", split=\"train\")\r\n\r\n```", "Hi @INF800,\r\n\r\nPlease note that the `imagefolder` feature enhancement was just recently merged to our master branch (https://github.com/huggingface/datasets/commit/207be676bffe9d164740a41a883af6125edef135), but has not yet been released.\r\n\r\nWe are planning to make the 2.0 release of our library in the coming days and then that feature will be available by updating your `datasets` library from PyPI.\r\n\r\nIn the meantime, you can incorporate that feature if you install our library from our GitHub master branch:\r\n```shell\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\n\r\nThen:\r\n```python\r\nIn [1]: from datasets import load_dataset\r\n ds = load_dataset(\"imagefolder\", data_files=\"https://download.microsoft.com/download/3/E/1/3E1C3F21-ECDB-4869-8368-6DEBA77B919F/kagglecatsanddogs_3367a.zip\", split=\"train\")\r\nUsing custom data configuration default-7eb4e80d960deb18\r\nDownloading and preparing dataset image_folder/default to .../.cache/huggingface/datasets/image_folder/default-7eb4e80d960deb18/0.0.0/8de8dc6d68ce3c81cc102b93cc82ede27162b5d30cd003094f935942c8294f60...\r\nDownloading data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 690.19it/s]\r\nExtracting data files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 852.85it/s]\r\nDataset image_folder downloaded and prepared to .../.cache/huggingface/datasets/image_folder/default-7eb4e80d960deb18/0.0.0/8de8dc6d68ce3c81cc102b93cc82ede27162b5d30cd003094f935942c8294f60. Subsequent calls will reuse this data.\r\n\r\nIn [2]: ds\r\nOut[2]: \r\nDataset({\r\n features: ['image', 'label'],\r\n num_rows: 25000\r\n})\r\n```", "Hey @albertvillanova. Does this load entire dataset in memory? Because I am facing huge trouble with loading very big datasets (OOM errors)", "Can you provide the error stack trace? The loader only stores the `data_files` dict, which can get big after globbing. Then, the OOM error would mean you don't have enough memory to keep all the paths to the image files. You can circumvent this by generating an archive and loading the dataset from there. Maybe we can optimize the globbing part in our data files resolution at some point, cc @lhoestq for visibility.", "Hey, memory error is resolved. It was fluke.\r\n\r\nBut there is another issue. Currently `load_dataset(\"imagefolder\", data_dir=\"./path/to/train\",)` takes only `train` as arg to `split` parameter.\r\n\r\nI am creating vaildation dataset using\r\n\r\n```\r\nds_valid = datasets.DatasetDict(valid=load_dataset(\"imagefolder\", data_dir=\"./path/to/valid\",)['train'])\r\n```", "`data_dir=\"path/to/folder\"` is a shorthand syntax fox `data_files={\"train\": \"path/to/folder/**\"}`, so use `data_files` in that case instead:\r\n```python\r\nds = load_dataset(\"imagefolder\", data_files={\"train\": \"path/to/train/**\", \"test\": \"path/to/test/**\", \"valid\": \"path/to/valid/**\"})\r\n```", "And there was another issue. I loaded black and white images (jpeg file). Using load dataset. It reads it as PIL jpeg data format. But instead of converting it into 3 channel tensor, input to collator function is coming as a single channel tensor.", "We don't apply any additional preprocessing on top of `PIL.Image.open(image_file)`, so you need to do the conversion yourself:\r\n\r\n```python\r\ndef to_rgb(batch):\r\n batch[\"image\"] = [img.convert(\"RGB\") for img in batch[\"image\"]]\r\n return batch\r\n\r\nds_rgb = ds.map(to_rgb, batched=True)\r\n```\r\n\r\nPlease use our Forum for questions of this kind in the future." ]

[]

[ "is pushing directly with git (and git-lfs) an option for you?", "I have installed git-lfs and doing this push with that\r\n", "yes but is there any way you could try pushing with `git` command line directly instead of `push_to_hub`?", "Okay. I didnt saved the dataset to my local machine. So, I processed the dataset and pushed it directly to the hub. I think I should try saving those dataset to my local machine by `save_to_disk` and then push it with git command line", "cc @lhoestq @albertvillanova @LysandreJik because maybe I'm giving dumb advice here 😅 ", "`push_to_hub` is the preferred way of uploading a dataset to the Hub, which can then be reloaded with `load_dataset`. Feel free to try again and see if the server is working as expected now. Maybe we can add a retry mechanism in the meantime to workaround 504 errors.\r\n\r\nRegarding `save_to_disk`, this must only be used for local serialization (because it's uncompressed and compatible with memory-mapping). If you upload a dataset saved with `save_to_disk` to the Hub, then to reload it you will have to download/clone the repository locally by yourself and use `load_from_disk`." ]

[ "Hi @omarespejel! I think the philosophy for `datasets` issues is to create concrete issues with proposals to add a specific, individual dataset rather than umbrella issues for things such as datasets for a language, since we could end up with hundreds of issues (one per language). I see NILC - USP has many datasets, I would suggest to either create an issue for their datasets, or even better, we are trying to push to upload datasets as community datasets instead of adding them to the core library as guided in https://huggingface.co/docs/datasets/share. That would have the additional benefit that the dataset would live under the NILC organization.\r\n\r\n@lhoestq correct me if I'm wrong please 😄 " ]

[ "To follow up on this: the cased and uncased versions actually contain different content, and the cased one is easier since it contains a Summary of the Invention in the input.\r\n\r\nSee the paper describing the issue here:\r\nhttps://aclanthology.org/2022.gem-1.34/", "Thanks for proposing the addition of the cased version of this dataset and for pinging again recently.\r\n\r\nI have just merged a PR that adds the cased version: https://huggingface.co/datasets/big_patent/discussions/3\r\n\r\nThe cased version (2.1.2) is the default one:\r\n```python\r\nds = load_dataset(\"big_patent\", \"all\")\r\n```\r\n\r\nTo use the 1.0.0 version (lower cased tokenized words), pass both parameters `codes` and `version`:\r\n```python\r\nds = load_dataset(\"big_patent\", codes=\"all\", version=\"1.0.0\")\r\n```\r\n\r\nClosed by: https://huggingface.co/datasets/big_patent/discussions/3" ]

[ "Hi @slvcsl, thanks for reporting.\r\n\r\nYesterday we just made a patch release of our `datasets` library that fixes this issue: version 1.18.4.\r\nhttps://pypi.org/project/datasets/#history\r\n\r\nPlease, feel free to update `datasets` library to the latest version: \r\n```shell\r\npip install -U datasets\r\n```\r\nAnd then you should force redownload of the data file to update your local cache: \r\n```python\r\nds = load_dataset(\"big_patent\", \"g\", split=\"validation\", download_mode=\"force_redownload\")\r\n```\r\n- Note that before the fix, you just downloaded and cached the Google Drive virus scan warning page, instead of the data file\r\n\r\nThis issue was already reported \r\n- #3784\r\n\r\nand its root cause is a change in the Google Drive service. See:\r\n- #3786 \r\n\r\nWe already fixed it. See:\r\n- #3787 \r\n" ]

[ "I agree using `glob.glob` alone is bad practice because it's not deterministic. Using `sorted` is a nice solution.\r\n\r\nNote that the `xglob` function you are referring to in the `streaming_download_manager.py` code just extends `glob.glob` for URLs - we don't change its behavior. That's why it has no `sorted()`" ]

[ "We raise the error “ FileNotFoundError: can’t find the dataset” mainly to follow best practice in security (otherwise users could be able to guess what private repositories users/orgs may have)\r\n\r\nWe can indeed reformulate this and add the \"If this is a private repository,...\" part !", "Resolved via https://github.com/huggingface/datasets/pull/4536" ]

[ "Hi @amanjaiswal777,\r\n\r\nFirst note that the dataset you are trying to load is deprecated: it was the Common Voice dataset release as of Dec 2020.\r\n\r\nCurrently, Common Voice dataset releases are directly hosted on the Hub, under the Mozilla Foundation organization: https://huggingface.co/mozilla-foundation\r\n\r\nFor example, to get their latest Common Voice relase (8.0):\r\n- Go to the dataset page and request access permission (Mozilla Foundation requires this for people willing to use their datasets): https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0\r\n- Looking at the dataset card, you can check that data instances have, among other fields, the ones you are interested in: \"accent\", \"age\",... \r\n- Then you can load their \"en\" language dataset as usual, besides passing your authentication token (more info on auth token here: https://huggingface.co/docs/hub/security)\r\n ```python\r\n from datasets import load_dataset\r\n ds_en = load_dataset(\"mozilla-foundation/common_voice_8_0\", \"en\", use_auth_token=True)\r\n ```\r\n- Finally, you can filter only the data instances you are interested in (more info on `filter` here: https://huggingface.co/docs/datasets/process#select-and-filter):\r\n ```python\r\n ds_england_en = ds_en.filter(lambda item: item[\"accent\"] == \"England English\")\r\n ```\r\n\r\nFeel free to reopen this issue if you need further assistance.", "Hey @albertvillanova trying the same approach as you with the common_voice_16_1 dataset. What I'm trying to do is to filter the valencian accent in the catalan subset. Gave me this error and I have everything it asks for decoding mp3:\r\n\r\n\r\n\r\n\r\n" ]

[ "Hi @lemoner20, thanks for reporting.\r\n\r\nI'm sorry but I cannot reproduce your problem:\r\n```python\r\nIn [1]: from datasets import load_dataset, load_metric, Audio\r\n ...: raw_datasets = load_dataset(\"superb\", \"ks\", split=\"train\")\r\n ...: print(raw_datasets[0][\"audio\"])\r\nDownloading builder script: 30.2kB [00:00, 13.0MB/s] \r\nDownloading metadata: 38.0kB [00:00, 16.6MB/s] \r\nDownloading and preparing dataset superb/ks (download: 1.45 GiB, generated: 9.64 MiB, post-processed: Unknown size, total: 1.46 GiB) to .../.cache/huggingface/datasets/superb/ks/1.9.0/fc1f59e1fa54262dfb42de99c326a806ef7de1263ece177b59359a1a3354a9c9...\r\nDownloading data: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.49G/1.49G [00:37<00:00, 39.3MB/s]\r\nDownloading data: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 71.3M/71.3M [00:01<00:00, 36.1MB/s]\r\nDownloading data files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:41<00:00, 20.67s/it]\r\nExtracting data files: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:28<00:00, 14.24s/it]\r\nDataset superb downloaded and prepared to .../.cache/huggingface/datasets/superb/ks/1.9.0/fc1f59e1fa54262dfb42de99c326a806ef7de1263ece177b59359a1a3354a9c9. Subsequent calls will reuse this data.\r\n{'path': '.../.cache/huggingface/datasets/downloads/extracted/8571921d3088b48f58f75b2e514815033e1ffbd06aa63fd4603691ac9f1c119f/_background_noise_/doing_the_dishes.wav', 'array': array([ 0. , 0. , 0. , ..., -0.00592041,\r\n -0.00405884, -0.00253296], dtype=float32), 'sampling_rate': 16000}\r\n``` \r\n\r\nWhich version of `datasets` are you using? Could you please fill in the environment info requested in the bug report template? You can run the command `datasets-cli env` and copy-and-paste its output below\r\n## Environment info\r\n<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->\r\n- `datasets` version:\r\n- Platform:\r\n- Python version:\r\n- PyArrow version:", "@albertvillanova Thanks for your reply. The environment info below\r\n\r\n## Environment info\r\n- `datasets` version: 1.18.3\r\n- Platform: Linux-4.19.91-007.ali4000.alios7.x86_64-x86_64-with-debian-buster-sid\r\n- Python version: 3.6.12\r\n- PyArrow version: 6.0.1", "Thanks @lemoner20,\r\n\r\nI cannot reproduce your issue in datasets version 1.18.3 either.\r\n\r\nMaybe redownloading the data file may work if you had already cached this dataset previously. Could you please try passing \"force_redownload\"?\r\n```python\r\nraw_datasets = load_dataset(\"superb\", \"ks\", split=\"train\", download_mode=\"force_redownload\")", "Thanks, @albertvillanova,\r\n\r\nI install the python package of **librosa=0.9.1** again, it works now!\r\n\r\n\r\n", "Cool!", "@albertvillanova, you can actually reproduce the error if you reach the cell `common_voice_train[0][\"path\"]` of this [notebook](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/Fine_Tune_XLSR_Wav2Vec2_on_Turkish_ASR_with_%F0%9F%A4%97_Transformers.ipynb#scrollTo=_0kRndSvqaKk). Error gets solved after updating the versions of the libraries used in there.", "@jvel07, thanks for reporting and finding a solution.\r\n\r\nMaybe we could tell @patrickvonplaten about the version pinning issue in his notebook.", "Should I update the version of datasets @albertvillanova ? " ]

[ "Hi @jxmorris12, thanks for reporting.\r\n\r\nThe objective of `verify_checksums` is to check that both checksums are equal. Therefore if one is None and the other is non-None, they are not equal, and the function accordingly raises a NonMatchingChecksumError. That behavior is expected.\r\n\r\nThe question is: how did you generate the expected checksum? Normally, it should not be None. To properly generate it (it is contained in the `dataset_infos.json` file), you should have runned: https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md\r\n```shell\r\ndatasets-cli test <your-dataset-folder> --save_infos --all_configs\r\n```\r\n\r\nOn the other hand, you should take into account that the generation of this file is NOT mandatory for personal/community datasets (we only require it for \"canonical\" datasets, i.e., datasets added to our library GitHub repository: https://github.com/huggingface/datasets/tree/master/datasets). Therefore, other option would be just to delete the `dataset_infos.json` file. If that file is not present, the function `verify_checksums` is not executed.\r\n\r\nFinally, you can circumvent the `verify_checksums` function by passing `ignore_verifications=True` to `load_dataset`:\r\n```python\r\nload_dataset(..., ignore_verifications=True)\r\n``` ", "Thanks @albertvillanova!\r\n\r\nThat's fine. I did run that command when I was adding a new dataset. Maybe because the command crashed in the middle, the checksum wasn't stored properly. I don't know where the bug is happening. But either (i) `verify_checksums` should properly handle this edge case, where the passed checksum is None or (ii) the `datasets-cli test` shouldn't generate a corrupted dataset_infos.json file.\r\n\r\nJust a more high-level thing, I was trying to follow the instructions for adding a dataset in the CONTRIBUTING.md, so if running that command isn't even necessary, that should probably be mentioned in the document, right? But that's somewhat of a moot point, since something isn't working quite right internally if I was able to get into this corrupted state in the first place, just by following those instructions.", "Hi @jxmorris12,\r\n\r\nDefinitely, your `dataset_infos.json` was corrupted (and wrongly contains expected None checksum). \r\n\r\nWhile we further investigate how this can happen and fix it, feel free to delete your `dataset_infos.json` file and recreate it with:\r\n```shell\r\ndatasets-cli test <your-dataset-folder> --save_infos --all_configs\r\n```\r\n\r\nAlso note that `verify_checksum` is working as expected: if it receives a None and and a non-None checksums as input pair, it must raise an exception: they are not equal. That is not a bug.", "At a higher level, also note that we are preparing the release of `datasets` version 2.0, and some docs are being updated...\r\n\r\nIn order to add a dataset, I think the most updated instructions are in our official documentation pages: https://huggingface.co/docs/datasets/share", "Thanks for the info. Maybe you can update the contributing.md if it's not up-to-date.", "Hi @jxmorris12, we have discovered the bug why `None` checksums wrongly appeared when generating the `dataset_infos.json` file:\r\n- #3892\r\n\r\nThe fix will be accessible once this PR merged. And we are planning to do our 2.0 release today.\r\n\r\nWe are also working on updating all our docs for our release today.", "Thanks @albertvillanova - congrats on the release!" ]

[ "<s>I think this is because the tokenizer is stateful and because the order in which the splits are processed is not deterministic. Because of that, the hash of the tokenizer may change for certain splits, which causes issues with caching.\r\n\r\nTo fix this we can try making the order of the splits deterministic for map.</s>", "Actually this is not because of the order of the splits, but most likely because the tokenizer used to process the second split is in a state that has been modified by the first split.\r\n\r\nTherefore after reloading the first split from the cache, then the second split can't be reloaded since the tokenizer hasn't seen the first split (and therefore is considered a different tokenizer).\r\n\r\nThis is a bit trickier to fix, we can explore fixing this next week maybe", "Sorry didn't have the bandwidth to take care of this yet - will re-assign when I'm diving into it again !", "I had this issue with `run_speech_recognition_ctc.py` for wa2vec2.0 fine-tuning. I made a small change and the hash for the function (which includes tokenisation) is now the same before and after pre-porocessing. With the hash being the same, the caching works as intended.\r\n\r\nBefore:\r\n```\r\n def prepare_dataset(batch):\r\n # load audio\r\n sample = batch[audio_column_name]\r\n\r\n inputs = feature_extractor(sample[\"array\"], sampling_rate=sample[\"sampling_rate\"])\r\n batch[\"input_values\"] = inputs.input_values[0]\r\n batch[\"input_length\"] = len(batch[\"input_values\"])\r\n\r\n # encode targets\r\n additional_kwargs = {}\r\n if phoneme_language is not None:\r\n additional_kwargs[\"phonemizer_lang\"] = phoneme_language\r\n\r\n batch[\"labels\"] = tokenizer(batch[\"target_text\"], **additional_kwargs).input_ids\r\n\r\n return batch\r\n\r\n with training_args.main_process_first(desc=\"dataset map preprocessing\"):\r\n vectorized_datasets = raw_datasets.map(\r\n prepare_dataset,\r\n remove_columns=next(iter(raw_datasets.values())).column_names,\r\n num_proc=num_workers,\r\n desc=\"preprocess datasets\",\r\n )\r\n```\r\nAfter:\r\n```\r\n def prepare_dataset(batch, feature_extractor, tokenizer):\r\n # load audio\r\n sample = batch[audio_column_name]\r\n\r\n inputs = feature_extractor(sample[\"array\"], sampling_rate=sample[\"sampling_rate\"])\r\n batch[\"input_values\"] = inputs.input_values[0]\r\n batch[\"input_length\"] = len(batch[\"input_values\"])\r\n\r\n # encode targets\r\n additional_kwargs = {}\r\n if phoneme_language is not None:\r\n additional_kwargs[\"phonemizer_lang\"] = phoneme_language\r\n\r\n batch[\"labels\"] = tokenizer(batch[\"target_text\"], **additional_kwargs).input_ids\r\n\r\n return batch\r\n\r\n pd = lambda batch: prepare_dataset(batch, feature_extractor, tokenizer)\r\n\r\n with training_args.main_process_first(desc=\"dataset map preprocessing\"):\r\n vectorized_datasets = raw_datasets.map(\r\n pd,\r\n remove_columns=next(iter(raw_datasets.values())).column_names,\r\n num_proc=num_workers,\r\n desc=\"preprocess datasets\",\r\n )\r\n```", "Not sure why the second one would work and not the first one - they're basically the same with respect to hashing. In both cases the function is hashed recursively, and therefore the feature_extractor and the tokenizer are hashed the same way.\r\n\r\nWith which tokenizer or feature extractor are you experiencing this behavior ?\r\n\r\nDo you also experience this ?\r\n> Tokenization for some subsets are repeated at the 2nd and 3rd run. Starting from the 4th run, everything are loaded from cache.", "Thanks ! Hopefully this can be useful to others, and also to better understand and improve hashing/caching ", "`tokenizer.save_pretrained(training_args.output_dir)` produces a different tokenizer hash when loaded on restart of the script. When I was debugging before I was terminating the script prior to this command, then rerunning. \r\n\r\nI compared the tokenizer items on the first and second runs, there are two different items:\r\n1st:\r\n```\r\n('_additional_special_tokens', [AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True)])\r\n\r\n...\r\n\r\n('tokens_trie', <transformers.tokenization_utils.Trie object at 0x7f4d6d0ddb38>)\r\n```\r\n\r\n2nd:\r\n```\r\n('_additional_special_tokens', [AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"<s>\", rstrip=False, lstrip=False, single_word=False, normalized=True), AddedToken(\"</s>\", rstrip=False, lstrip=False, single_word=False, normalized=True)])\r\n\r\n...\r\n\r\n('tokens_trie', <transformers.tokenization_utils.Trie object at 0x7efc23dcce80>)\r\n```\r\n\r\n On every run of this the special tokens are being added on, and the hash is different on the `tokens_trie`. The increase in the special tokens category could be cleaned, but not sure about the hash for the `tokens_trie`. What might work is that the call for the tokenizer encoding can be translated into a function that strips any unnecessary information out, but that's a guess.\r\n", "Thanks for investigating ! Does that mean that `save_pretrained`() produces non-deterministic tokenizers on disk ? Or is it `from_pretrained()` which is not deterministic given the same files on disk ?\r\n\r\nI think one way to fix this would be to make save/from_pretrained deterministic, or make the pickling of `transformers.tokenization_utils.Trie` objects deterministic (this could be implemented in `transformers`, but maybe let's discuss in an issue in `transformers` before opening a PR)", "Late to the party but everything should be deterministic (afaik at least).\r\n\r\nBut `Trie` is a simple class object, so afaik it's hash function is linked to its `id(self)` so basically where it's stored in memory, so super highly non deterministic. Could that be the issue ?", "> But Trie is a simple class object, so afaik it's hash function is linked to its id(self) so basically where it's stored in memory, so super highly non deterministic. Could that be the issue ?\r\n\r\nWe're computing the hash of the pickle dump of the class so it should be fine, as long as the pickle dump is deterministic", "I've ported wav2vec2.0 fine-tuning into Optimum-Graphcore which is where I found the issue. The majority of the script was copied from the Transformers version to keep it similar, [here is the tokenizer loading section from the source](https://github.com/huggingface/transformers/blob/f0982682bd6fd0b438dda79ec45f3a8fac83a985/examples/pytorch/speech-recognition/run_speech_recognition_ctc.py#L531).\r\n\r\nIn the last comment I have two loaded tokenizers, one from run 'N' of the script and one from 'N+1'. I think what's happening is that when you add special tokens (e.g. PAD and UNK) another AddedToken object is appended when tokenizer is saved regardless of whether special tokens are there already. \r\n\r\nIf there is a AddedTokens cleanup at load/save this could solve the issue, but then is Trie going to cause hash to be different? I'm not sure. ", "Which Python version are you using ?\r\n\r\nThe trie is basically a big dict of dics, so deterministic nature depends on python version:\r\nhttps://stackoverflow.com/questions/2053021/is-the-order-of-a-python-dictionary-guaranteed-over-iterations\r\n\r\nMaybe the investigation is actually not finding the right culprit though (the memory id is changed, but `datasets` is not using that to compare, so maybe we need to be looking within `datasets` so see where the comparison fails)", "Similar issue found on `BartTokenizer`. You can bypass the bug by loading a fresh new tokenizer everytime.\r\n\r\n```\r\n dataset = dataset.map(lambda x: tokenize_func(x, BartTokenizer.from_pretrained(xxx)),\r\n num_proc=num_proc, desc='Tokenize')\r\n```", "Linking in https://github.com/huggingface/datasets/issues/6179#issuecomment-1701244673 with an explanation.", "I got the same problem while using Wav2Vec2CTCTokenizer in a distributed experiment (many processes), and found that the problem was localized in the serialization (pickle dump) of the field `tokenizer.tokens_trie._tokens` (just a python set). I focussed into the set serialization and found it is not deterministic:\r\n\r\n```\r\nfrom datasets.fingerprint import Hasher\r\nfrom pickle import dumps,loads\r\n\r\n# used just once to get a serialized literal\r\n#print(dumps(set(\"abc\")))\r\nserialized = b'\\x80\\x04\\x95\\x11\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x8f\\x94(\\x8c\\x01a\\x94\\x8c\\x01c\\x94\\x8c\\x01b\\x94\\x90.'\r\n\r\nmyset = loads(serialized)\r\nprint(f'{myset=} {Hasher.hash(myset)}')\r\nprint(serialized == dumps(myset))\r\n```\r\n\r\nEvery time you run the python script (different processes) you get a random result. @lhoestq does it make any sense?", "OK, I assume python's set is just a hash table implementation that uses internally the hash() function. The problem is that python's hash() is not deterministic. I believe that setting the environment variable PYTHONHASHSEED to a fixed value, you can force it to be deterministic. I tried it (file `set_pickle_dump.py`):\r\n\r\n```\r\n#!/usr/bin/python3\r\n\r\nfrom datasets.fingerprint import Hasher\r\nfrom pickle import dumps,loads\r\n\r\n# used just once to get a serialized literal (with environment variable PYTHONHASHSEED set to 42)\r\n#print(dumps(set(\"abc\")))\r\nserialized = b'\\x80\\x04\\x95\\x11\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x8f\\x94(\\x8c\\x01b\\x94\\x8c\\x01c\\x94\\x8c\\x01a\\x94\\x90.'\r\n\r\nmyset = loads(serialized)\r\nprint(f'{myset=} {Hasher.hash(myset)}')\r\nprint(serialized == dumps(myset))\r\n```\r\n\r\nand now every run (`PYTHONHASHSEED=42 ./set_pickle_dump.py`) gets tthe same result. I tried then to test it with the tokenizer (file `test_tokenizer.py`):\r\n\r\n```\r\n#!/usr/bin/python3\r\nfrom transformers import AutoTokenizer\r\nfrom datasets.fingerprint import Hasher\r\n\r\ntokenizer = AutoTokenizer.from_pretrained('model')\r\nprint(f'{type(tokenizer)=}')\r\nprint(f'{Hasher.hash(tokenizer)=}')\r\n```\r\n\r\nexecuted as `PYTHONHASHSEED=42 ./test_tokenizer.py` and now the tokenizer fingerprint is allways the same!\r\n", "Thanks for reporting. I opened a PR here to propose a fix: https://github.com/huggingface/datasets/pull/6318 and doesn't require setting `PYTHONHASHSEED`\r\n\r\nCan you try to install `datasets` from this branch and tell me if it fixes the issue ?", "I patched (*) the file `datasets/utils/py_utils.py` and cache is working propperly now. Thanks!\r\n\r\n(*): I am running my experiments inside a docker container that depends on `huggingface/transformers-pytorch-gpu:latest`, so pattched the file instead of rebuilding the container from scratch", "Fixed by #6318.", "The OP issue hasn't been fixed, re-opening", "I think the Trie()._tokens of PreTrainedTokenizer need to be a sorted set So that the results of `hash_bytes(dumps(tokenizer))` are consistent every time", "I believe the issue may be linked to [tokenization_utils.py#L507](https://github.com/huggingface/transformers/blob/main/src/transformers/tokenization_utils.py#L507)，specifically in the line where self.tokens_trie.add(token.content) is called. The function _update_trie appears to modify an unordered set. Consequently, this line:\r\n`value = hash_bytes(dumps(tokenizer.tokens_trie._tokens))`\r\ncan lead to inconsistencies when rerunning the code.\r\n\r\nThis, in turn, results in inconsistent outputs for both `hash_bytes(dumps(function))` at [arrow_dataset.py#L3053](https://github.com/huggingface/datasets/blob/main/src/datasets/arrow_dataset.py#L3053) and\r\n`hasher.update(transform_args[key])` at [fingerprint.py#L323](https://github.com/huggingface/datasets/blob/main/src/datasets/fingerprint.py#L323)\r\n\r\n```\r\ndataset_kwargs = {\r\n \"shard\": raw_datasets,\r\n \"function\": tokenize_function,\r\n}\r\ntransform = format_transform_for_fingerprint(Dataset._map_single)\r\nkwargs_for_fingerprint = format_kwargs_for_fingerprint(Dataset._map_single, (), dataset_kwargs)\r\nkwargs_for_fingerprint[\"fingerprint_name\"] = \"new_fingerprint\"\r\nnew_fingerprint = update_fingerprint(raw_datasets._fingerprint, transform, kwargs_for_fingerprint)\r\n```\r\n", "Alternatively, does the \"dumps\" function require separate processing for the set?", "We did a fix that does sorting whenever we hash sets. The fix is available on `main` if you want to try it out. We'll do a new release soon :)", "Is there a documentation chapter that discusses in which cases you should expect your dataset preprocessing to be cached. Including do's and don'ts for the preprocessing functions? I think Datasets team does amazing job at tacking this issue on their side, but it would be great to have some guidelines on the user side as well.\r\n\r\nIn our current project we have two cases (text-to-text classification and summarization) and in one of them the cache is sometimes reused when it's not supposed to be reused while in the other it's never used at all 😅", "You can find some docs here :) \r\nhttps://huggingface.co/docs/datasets/about_cache" ]

[ "Hi! \r\n\r\nThis issue stems from `datasets` having `py.typed` defined (see https://github.com/microsoft/pyright/discussions/3764#discussioncomment-3282142) - to avoid it, we would either have to remove `py.typed` (added to be compliant with PEP-561) or export the names with `__all__`/`from .submodule import name as name`.\r\n\r\nTransformers is fine as it no longer has `py.typed` (removed in https://github.com/huggingface/transformers/pull/18485)\r\n\r\nWDYT @lhoestq @albertvillanova @polinaeterna \r\n\r\n@sgugger's point makes sense - we should either be \"properly typed\" (have py.typed + mypy tests) or drop `py.typed` as Transformers did (I like this option better).\r\n\r\n(cc @Wauplin since `huggingface_hub` has the same issue.)", "I'm fine with dropping it, but autotrain people won't be happy @SBrandeis ", "> (cc @Wauplin since huggingface_hub has the same issue.)\r\n\r\nHmm maybe we have the same issue but I haven't been able to reproduce something similar to `\"load_dataset\" is not exported from module \"datasets\"` message (using VSCode+Pylance -that is powered by Pyright). `huggingface_hub` contains a `py.typed` file but the package itself is actually typed. We are running `mypy` in our CI tests since ~3 months and so far it seems to be ok. But happy to change if it causes some issues with linters.\r\n\r\nAlso the top-level [`__init__.py`](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/__init__.py) is quite different in `hfh` than `datasets` (at first glance). We have a section at the bottom to import all high level methods/classes in a `if TYPE_CHECKING` block.", "@Wauplin I only get the error if I use Pyright's CLI tool or the Pyright extension (not sure why, but Pylance also doesn't report this issue on my machine)\r\n\r\n> Also the top-level [`__init__.py`](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/__init__.py) is quite different in `hfh` than `datasets` (at first glance). We have a section at the bottom to import all high level methods/classes in a `if TYPE_CHECKING` block.\r\n\r\nI tried to fix the issue with `TYPE_CHECKING`, but it still fails if `py.typed` is present.", "@mariosasko thank for the tip. I have been able to reproduce the issue as well. I would be up for including a (huge) static `__all__` variable in the `__init__.py` (since the file is already generated automatically in `hfh`) but honestly I don't think it's worth the hassle. \r\n\r\nI'll delete the `py.typed` file in `huggingface_hub` to be consistent between HF libraries. I opened a PR here: https://github.com/huggingface/huggingface_hub/pull/1329", "I am getting this error in google colab today:\r\n\r\n\r\n\r\nThe code runs just fine too." ]

[]

[]

[]

[ "It will be indeed really great to add support to GNN datasets. Big :+1: for this initiative.", "@napoles-uach identifies the [TUDatasets](https://chrsmrrs.github.io/datasets/) (A collection of benchmark datasets for graph classification and regression). \r\n\r\nAdded to the Tasks in the initial issue.", "Thanks Omar, that is a great collection!", "Great initiative! Let's keep this issue for these 3 datasets, but moving forward maybe let's create a new issue per dataset :rocket: great work @napoles-uach and @omarespejel!" ]

[ "Maybe @Rocketknight1 can help here", "Hi @greenned, this is expected behaviour for `to_tf_dataset`. By default, we drop the smaller 'remainder' batch during training (i.e. when `shuffle=True`). If you really want to keep that batch, you can set `drop_remainder=False` when calling `to_tf_dataset()`.", "@Rocketknight1 Oh, thank you. I didn't get **drop_remainder** Have a nice day!", "No problem!\r\n" ]

[ "Was able to reproduce the issue on Colab; full logs below. \r\n\r\n```\r\n---------------------------------------------------------------------------\r\nNotADirectoryError Traceback (most recent call last)\r\n[<ipython-input-2-39967739ba7f>](https://localhost:8080/#) in <module>()\r\n 1 import datasets\r\n 2 \r\n----> 3 train_data = datasets.load_dataset(\"cnn_dailymail\", \"3.0.0\", split=\"train\")\r\n\r\n5 frames\r\n[/usr/local/lib/python3.7/dist-packages/datasets/load.py](https://localhost:8080/#) in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, keep_in_memory, save_infos, revision, use_auth_token, task, streaming, script_version, **config_kwargs)\r\n 1705 ignore_verifications=ignore_verifications,\r\n 1706 try_from_hf_gcs=try_from_hf_gcs,\r\n-> 1707 use_auth_token=use_auth_token,\r\n 1708 )\r\n 1709 \r\n\r\n[/usr/local/lib/python3.7/dist-packages/datasets/builder.py](https://localhost:8080/#) in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, base_path, use_auth_token, **download_and_prepare_kwargs)\r\n 593 if not downloaded_from_gcs:\r\n 594 self._download_and_prepare(\r\n--> 595 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs\r\n 596 )\r\n 597 # Sync info\r\n\r\n[/usr/local/lib/python3.7/dist-packages/datasets/builder.py](https://localhost:8080/#) in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)\r\n 659 split_dict = SplitDict(dataset_name=self.name)\r\n 660 split_generators_kwargs = self._make_split_generators_kwargs(prepare_split_kwargs)\r\n--> 661 split_generators = self._split_generators(dl_manager, **split_generators_kwargs)\r\n 662 \r\n 663 # Checksums verification\r\n\r\n[/root/.cache/huggingface/modules/datasets_modules/datasets/cnn_dailymail/3cb851bf7cf5826e45d49db2863f627cba583cbc32342df7349dfe6c38060234/cnn_dailymail.py](https://localhost:8080/#) in _split_generators(self, dl_manager)\r\n 253 def _split_generators(self, dl_manager):\r\n 254 dl_paths = dl_manager.download_and_extract(_DL_URLS)\r\n--> 255 train_files = _subset_filenames(dl_paths, datasets.Split.TRAIN)\r\n 256 # Generate shared vocabulary\r\n 257 \r\n\r\n[/root/.cache/huggingface/modules/datasets_modules/datasets/cnn_dailymail/3cb851bf7cf5826e45d49db2863f627cba583cbc32342df7349dfe6c38060234/cnn_dailymail.py](https://localhost:8080/#) in _subset_filenames(dl_paths, split)\r\n 154 else:\r\n 155 logger.fatal(\"Unsupported split: %s\", split)\r\n--> 156 cnn = _find_files(dl_paths, \"cnn\", urls)\r\n 157 dm = _find_files(dl_paths, \"dm\", urls)\r\n 158 return cnn + dm\r\n\r\n[/root/.cache/huggingface/modules/datasets_modules/datasets/cnn_dailymail/3cb851bf7cf5826e45d49db2863f627cba583cbc32342df7349dfe6c38060234/cnn_dailymail.py](https://localhost:8080/#) in _find_files(dl_paths, publisher, url_dict)\r\n 133 else:\r\n 134 logger.fatal(\"Unsupported publisher: %s\", publisher)\r\n--> 135 files = sorted(os.listdir(top_dir))\r\n 136 \r\n 137 ret_files = []\r\n\r\nNotADirectoryError: [Errno 20] Not a directory: '/root/.cache/huggingface/datasets/downloads/1bc05d24fa6dda2468e83a73cf6dc207226e01e3c48a507ea716dc0421da583b/cnn/stories'\r\n```", "Hi @jon-tow, thanks for reporting. And hi @dynamicwebpaige, thanks for your investigation. \r\n\r\nThis issue was already reported \r\n- #3784\r\n\r\nand its root cause is a change in the Google Drive service. See:\r\n- #3786 \r\n\r\nWe have already fixed it. See:\r\n- #3787 \r\n\r\nWe are planning to make a patch release today (indeed, we were planning to do it last Friday).\r\n\r\nIn the meantime, you can get this fix by installing our library from the GitHub master branch:\r\n```shell\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\nThen, if you had previously tried to load the data and got the checksum error, you should force the redownload of the data (before the fix, you just downloaded and cached the virus scan warning page, instead of the data file):\r\n```shell\r\nload_dataset(\"...\", download_mode=\"force_redownload\")\r\n```\r\n\r\nCC: @lhoestq " ]

[ "Hi ! Yes this is definitely something we'll explore, since optimizing processing pipelines can be challenging and because performance is key here: we want anyone to be able to play with large-scale datasets more easily.\r\n\r\nI think we'll start by documenting the performance of the dataset transforms we provide, and then we can have some tools to help debugging/optimizing them" ]

[ "Hi @dlwh, thanks for reporting.\r\n\r\nPlease note, that the source data files for \"all\" config are different from the other configurations.\r\n\r\nThe \"all\" config contains the official Pile data files, from https://mystic.the-eye.eu/public/AI/pile/\r\nAll data examples contain a \"meta\" dict with a single \"pile_set_name\" key:\r\n```python\r\nIn [1]: from datasets import load_dataset\r\n ds = load_dataset(\"the_pile\", \"all\", split=\"train\", streaming=True)\r\n item = next(iter(ds))\r\nDownloading builder script: 9.09kB [00:00, 4.42MB/s]\r\n\r\nIn [3]: item[\"meta\"]\r\nOut[3]: {'pile_set_name': 'Pile-CC'}\r\n```\r\n\r\nOn the other hand, all the other subset configs data files come from the Pile preliminary components directory: https://mystic.the-eye.eu/public/AI/pile_preliminary_components/\r\nFor theses components, the \"meta\" field may have different keys depending on the subset: \"id\", \"language\", \"pmid\",... Because of that, if we had kept the `dict` data format for the \"meta\" field, we would have an error when trying to concatenate different subsets, whose \"meta\" keys are not identical. In order to avoid that, the \"meta\" field is cast to `str` in all these cases, so that there is no incompatibility in their \"meta\" data type when concatenating.\r\n\r\nYou can check, for example, that for \"pubmed_central\" the \"meta\" field is cast to `str`:\r\n```python\r\nIn [4]: from datasets import load_dataset\r\n ds = load_dataset(\"the_pile\", \"pubmed_central\", split=\"train\", streaming=True)\r\n item = next(iter(ds))\r\n\r\nIn [5]: item[\"meta\"]\r\nOut[5]: \"{'id': 'PMC6071596'}\"\r\n```\r\n\r\nFeel free to reopen this issue if you have further questions. " ]

[ "Hi @jacobbieker, thanks for reporting!\r\n\r\nI have transferred this issue to our Hub team and they are investigating it. I keep you informed. ", "Hi @jacobbieker, we are investigating this issue on our side and we'll see if we can fix it, but please note that your repo is considered problematic for git. Here are the results of running https://github.com/github/git-sizer on it:\r\n\r\n```\r\nProcessing blobs: 147448 \r\nProcessing trees: 27 \r\nProcessing commits: 4 \r\nMatching commits to trees: 4 \r\nProcessing annotated tags: 0 \r\nProcessing references: 3 \r\n| Name | Value | Level of concern |\r\n| ---------------------------- | --------- | ------------------------------ |\r\n| Biggest objects | | |\r\n| * Trees | | |\r\n| * Maximum entries [1] | 167 k | !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! |\r\n| | | |\r\n| Biggest checkouts | | |\r\n| * Number of files [2] | 189 k | *** |\r\n\r\n[1] aa057d2667c34c70c6146efc631f5c9917ff326e (refs/heads/main:2016.zarr/unknown)\r\n[2] 6897b7bf6440fdd16b2c39d08085a669e7eaa59d (refs/heads/main^{tree})\r\n```\r\n\r\nYou can check https://github.com/github/git-sizer for more information on how to avoid such pathological structures.", "Hi, thanks for getting back to me so quick! And yeah, I figured that was probably the problem. I was going to try to delete the repo, but couldn't through the website, so if that's the easiest way to solve it, I can regenerate the dataset in a different format with less tiny files, and you guys can delete the repo as it is. Zarr just saves everything as lots of small files to make chunks easy to load, which is why I was preferring that format, but maybne that just doesn't work well for HF datasets.", "Hi @jacobbieker,\r\n\r\nFor future use cases, our Hub team is still pondering whether to limit the maximum number of files per repo to avoid technical issues...\r\n\r\nOn the meantime, they have made a fix and your dataset is working: https://huggingface.co/datasets/openclimatefix/mrms" ]

[ "Official dataset location : https://icu.ee.ethz.ch/research/datsets.html\r\nIn the \"Biwi Kinect Head Pose Database\" section, I do not find any information regarding \"Downloading the dataset.\" . Do we mail the authors regarding this ?\r\n\r\nI found the dataset on Kaggle : [Link](https://www.kaggle.com/kmader/biwi-kinect-head-pose-database) , but since 🤗 does not host any of the datasets, this would require the user to provide their Kaggle username and API key to download. \r\n\r\nAny inputs on how we could proceed ? Thank you.\r\n[ Need your inputs here, @lhoestq or @mariosasko ]", "Hi @dnaveenr! Thanks for tackling this issue. This link should work: https://data.vision.ee.ethz.ch/cvl/gfanelli/kinect_head_pose_db.tgz", "#self-assign", "Added in https://github.com/huggingface/datasets/pull/3903, thanks @dnaveenr !" ]

[ "Hi @jon-tow, thanks for reporting.\r\n\r\nThis issue was already reported and its root cause is a change in the Google Drive service. See:\r\n- #3786 \r\n\r\nWe have already fixed it. See:\r\n- #3787 \r\n\r\nWe are planning to make a patch release today.\r\n\r\nIn the meantime, you can get this fix by installing our library from the GitHub master branch:\r\n```shell\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\nThen, if you had previously tried to load the data and got the checksum error, you should force the redownload of the data (before the fix, you just downloaded and cached the virus scan warning page, instead of the data file):\r\n```shell\r\nload_dataset(\"...\", download_mode=\"force_redownload\")\r\n```" ]

[ "Hi, thanks for reporting! We can add a `sources: datasets.Value(\"string\")` feature to the `Features` dict in the `SARI` script to fix this. Would you be interested in submitting a PR?", "Hi Mario,\r\n\r\nThanks for your message. I did try to add `sources` into the `Features` dict using a script for the metric:\r\n```\r\n features=datasets.Features(\r\n {\r\n \"sources\": datasets.Value(\"string\", id=\"sequence\"),\r\n \"predictions\": datasets.Value(\"string\", id=\"sequence\"),\r\n \"references\": datasets.Sequence(datasets.Value(\"string\", id=\"sequence\"), id=\"references\"),\r\n }\r\n ),\r\n```\r\n\r\nBut that only avoids a failure in `encode_batch` in the `add_batch` method:\r\n```\r\n batch = {\"predictions\": predictions, \"references\": references}\r\n batch = self.info.features.encode_batch(batch)\r\n```\r\n\r\nThe real problem is that `add_batch()`, `add()` and `compute()` does not receive a `sources` param:\r\n```\r\ndef add_batch(self, *, predictions=None, references=None):\r\ndef add(self, *, prediction=None, reference=None):\r\ndef compute(self, *, predictions=None, references=None, **kwargs)\r\n```\r\n\r\nAnd then, it fails:\r\n`TypeError: add_batch() got an unexpected keyword argument sources`\r\n\r\nI need this for adding any metric based on SARI or alike, not only for sari.py :)\r\n\r\nLet me know if I understood correctly the proposed solution.\r\n", "The `Metric` class has been modified recently to support this use-case, but the `add_batch` + `compute` pattern still doesn't work correctly. I'll open a PR." ]

[ "I would like to take this up and give it a shot. Any image specific - dataset guidelines to keep in mind ? Thank you.", "#self-assign", "I've started working on adding this dataset. I require some inputs on the following : \r\n\r\nRef for the initial draft [here](https://github.com/dnaveenr/datasets/blob/add_metashift_dataset/datasets/metashift/metashift.py)\r\n1. The dataset does not have a typical - train/test/val split. What do we do for the _split_generators() function ? How do we go about this ?\r\n2. This dataset builds on the Visual Genome dataset, using a metadata file. The dataset is generated using generate_full_MetaShift.py script. By default, the authors choose to generate the dataset only for a SELECTED_CLASSES. The following script is used : \r\nCode : https://github.com/Weixin-Liang/MetaShift/blob/main/dataset/generate_full_MetaShift.py \r\nInfo : https://metashift.readthedocs.io/en/latest/sub_pages/download_MetaShift.html#generate-the-full-metashift-dataset\r\nCan I just copy over the required functions into the metashift.py to generate the dataset ?\r\n3. How do we complete the _generate_examples for this dataset ?\r\n\r\nThe user has the ability to use default selected classes, get the complete dataset or add more specific additional classes. I think config would be a good option here.\r\n\r\nInputs, suggestions would be helpful. Thank you.", "I think @mariosasko and @lhoestq should be able to help here 😄 ", "Hi ! Thanks for adding this dataset :) Let me answer your questions:\r\n\r\n1. in this case you can put everything in the \"train\" split\r\n2. Yes you can copy the script (provided you also include the MIT license of the code in the file header for example). Though we ideally try to not create new directories nor files when generating dataset, so if possible this script should be adapted to not create the file structure they mentioned, but instead yield the images one by one in `_generate_examples`. Let me know if you think this is feasible\r\n3. see point 2 haha\r\n\r\n> The user has the ability to use default selected classes, get the complete dataset or add more specific additional classes. I think config would be a good option here.\r\n\r\nYup ! We can also define a `selected_classes` parameter such that users can do\r\n```python\r\nload_dataset(\"metashift\", selected_classes=[\"cat\", \"dog\", ...])\r\n```", "Great. This is helpful. Thanks @lhoestq .\r\nRegarding Point 2, I'll try using yield instead of creating the directories and see if its feasible. selected_classes config sounds good.", "Closed via #3900 " ]

[ "Hi @muelletm, thanks for reporting.\r\n\r\nThis issue was already reported and its root cause is a change in the Google Drive service. See:\r\n- #3786 \r\n\r\nWe have already fixed it. See:\r\n- #3787 \r\n\r\nUntil our next `datasets` library release, you can get this fix by installing our library from the GitHub master branch:\r\n```shell\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\nThen, if you had previously tried to load the data and got the checksum error, you should force the redownload of the data (before the fix, you just downloaded and cached the virus scan warning page, instead of the data file):\r\n```shell\r\nload_dataset(\"...\", download_mode=\"force_redownload\")\r\n```" ]

[ "Ok - this is still an issue but I believe the root cause is different than I originally thought. I'm now able to get caching to work consistently with the above example as long as I fix the python hash seed `export PYTHONHASHSEED=1234`", "Hi! \r\n\r\nYes, our hasher should work with decorators. For instance, this dummy example:\r\n```python\r\ndef f(arg):\r\n def f1(ex):\r\n return {\"a\": ex[\"col1\"] + arg}\r\n return f1\r\n```\r\ngives the same hash across different Python sessions (`datasets.fingerprint.Hasher.hash(f(\"string1\")` returns `\"408c9059f89dbd6c\"` on my machine).\r\n\r\nCould you please make the example self-contained? This way, we can reproduce the bug. Additionally, you can try to find the problematic object yourself by testing their hash with `datasets.fingerprint.Hasher.hash(obj)`\r\n\r\nThis could be related to https://github.com/huggingface/datasets/issues/3638.", "#3638 was indeed my issue. Thanks!" ]

[ "@albertvillanova here's a separate issue for a bug similar to #3792", "Hi @afcruzs-ms, thanks for opening this separate issue for your problem.\r\n\r\nThe root problem in the other issue (#3792) was a change in the service of Google Drive.\r\n\r\nBut in your case, the `xcopa` dataset is not hosted on Google Drive. Therefore, the root cause should be a different one.\r\n\r\nLet me look at it... ", "@afcruzs-ms, I'm not able to reproduce the issue you reported:\r\n```python\r\nIn [1]: from datasets import load_dataset\r\n ...: dataset = load_dataset(\"xcopa\", \"it\")\r\nDownloading builder script: 5.21kB [00:00, 2.75MB/s] \r\nDownloading metadata: 28.6kB [00:00, 14.5MB/s] \r\nDownloading and preparing dataset xcopa/it (download: 627.09 KiB, generated: 76.43 KiB, post-processed: Unknown size, total: 703.52 KiB) to .../.cache/huggingface/datasets/xcopa/it/1.0.0/e1fab65f984b24c8b66bcf7ac27a26a1182f84adfb2e74035861be65e214b9e6...\r\nDownloading data: 642kB [00:00, 5.42MB/s]\r\nDataset xcopa downloaded and prepared to .../.cache/huggingface/datasets/xcopa/it/1.0.0/e1fab65f984b24c8b66bcf7ac27a26a1182f84adfb2e74035861be65e214b9e6. Subsequent calls will reuse this data. \r\n100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 733.27it/s]\r\n\r\nIn [2]: dataset\r\nOut[2]: \r\nDatasetDict({\r\n test: Dataset({\r\n features: ['premise', 'choice1', 'choice2', 'question', 'label', 'idx', 'changed'],\r\n num_rows: 500\r\n })\r\n validation: Dataset({\r\n features: ['premise', 'choice1', 'choice2', 'question', 'label', 'idx', 'changed'],\r\n num_rows: 100\r\n })\r\n})\r\n```\r\n\r\nMaybe you have some issue with your cached data... Could you please try to force the redownload of the data?\r\n```python\r\ndataset = load_dataset(\"xcopa\", \"it\", download_mode=\"force_redownload\")\r\n```", "It works indeed, thanks! ", "unfortunately, i am having a similar problem with the irc_disentaglement dataset :/\r\nmy code:\r\n```\r\nfrom datasets import load_dataset\r\n\r\ndataset = load_dataset(\"irc_disentangle\", download_mode=\"force_redownload\")\r\n```\r\n\r\nhowever, it produces the same error as @afcruzs-ms \r\n```\r\n[38](file:///Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/datasets/utils/info_utils.py?line=37) if len(bad_urls) > 0:\r\n [39](file:///Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/datasets/utils/info_utils.py?line=38) error_msg = \"Checksums didn't match\" + for_verification_name + \":\\n\"\r\n---> [40](file:///Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/datasets/utils/info_utils.py?line=39) raise NonMatchingChecksumError(error_msg + str(bad_urls))\r\n [41](file:///Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/datasets/utils/info_utils.py?line=40) logger.info(\"All the checksums matched successfully\" + for_verification_name)\r\n\r\nNonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://github.com/jkkummerfeld/irc-disentanglement/tarball/master']\r\n```\r\n\r\nI attempted to use the `ignore_verifications' as such:\r\n```\r\nds = datasets.load_dataset('irc_disentangle', download_mode=\"force_redownload\", ignore_verifications=True)\r\n\r\n```\r\n```\r\nDownloading builder script: 12.0kB [00:00, 5.92MB/s] \r\nDownloading metadata: 7.58kB [00:00, 3.48MB/s] \r\nNo config specified, defaulting to: irc_disentangle/ubuntu\r\nDownloading and preparing dataset irc_disentangle/ubuntu (download: 112.98 MiB, generated: 60.05 MiB, post-processed: Unknown size, total: 173.03 MiB) to /Users/laylabouzoubaa/.cache/huggingface/datasets/irc_disentangle/ubuntu/1.0.0/0f24ab262a21d8c1d989fa53ed20caa928f5880be26c162bfbc02445dbade7e5...\r\nDownloading data: 118MB [00:09, 12.1MB/s] \r\n \r\nDataset irc_disentangle downloaded and prepared to /Users/laylabouzoubaa/.cache/huggingface/datasets/irc_disentangle/ubuntu/1.0.0/0f24ab262a21d8c1d989fa53ed20caa928f5880be26c162bfbc02445dbade7e5. Subsequent calls will reuse this data.\r\n100%|██████████| 3/3 [00:00<00:00, 675.38it/s]\r\n```\r\nbut, this returns an empty set?\r\n\r\n```\r\nDatasetDict({\r\n train: Dataset({\r\n features: ['id', 'raw', 'ascii', 'tokenized', 'date', 'connections'],\r\n num_rows: 0\r\n })\r\n test: Dataset({\r\n features: ['id', 'raw', 'ascii', 'tokenized', 'date', 'connections'],\r\n num_rows: 0\r\n })\r\n validation: Dataset({\r\n features: ['id', 'raw', 'ascii', 'tokenized', 'date', 'connections'],\r\n num_rows: 0\r\n })\r\n})\r\n```\r\n\r\nnot sure what else to try at this point?\r\nThanks in advanced🤗", "Thanks @labouz for reporting: yes, better opening a new GitHub issue as you did. I'm addressing it:\r\n- #4376" ]

[ "Gently pinging @lhoestq", "Hi ! Interresting :)\r\n\r\nCould you give more details on what kind of separators you would like to use instead ?", "In my case, I just want to use `\\n` but not `U+2028`.", "Ok I see, maybe there can be a `sep` parameter to allow users to specify what line/paragraph separator they'd like to use", "Related to:\r\n- #3729 \r\n- #3910", "Thanks for requesting this enhancement. We have recently found a somehow related issue with another dataset:\r\n- #3704\r\n\r\nLet me make a PR proposal." ]

[ "same issue. downgrade it to a lower version.", "Thanks for reporting, I'll take a look tomorrow :)" ]

[ "Same issue with `dataset = load_dataset(\"dbpedia_14\")`\r\n```\r\nNonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://drive.google.com/uc?export=download&id=0Bz8a_Dbh9QhbQ2Vic1kxMmZZQ1k']", "I think this is a side-effect of #3787. The checksums won't match because the URLs have changed. @rafikg @Y0mingZhang, while this is fixed, maybe you can load the datasets as such:\r\n\r\n`data = datasets.load_dataset(\"wiki_lingua\", name=language, split=\"train[:2000]\", ignore_verifications=True)`\r\n`dataset = load_dataset(\"dbpedia_14\", ignore_verifications=True)`\r\n\r\nThis will, most probably, skip the verifications and integrity checks listed [here](https://huggingface.co/docs/datasets/loading_datasets.html#integrity-verifications)", "Hi! Installing the `datasets` package from master (`pip install git+https://github.com/huggingface/datasets.git`) and then redownloading the datasets with `download_mode` set to `force_redownload` (e.g. `dataset = load_dataset(\"dbpedia_14\", download_mode=\"force_redownload\")`) should fix the issue.", "Hi @rafikg and @Y0mingZhang, thanks for reporting.\r\n\r\nIndeed it seems that Google Drive changed their way to access their data files. We have recently handled that change:\r\n- #3787\r\n\r\nbut it will be accessible to users only in our next release of the `datasets` version.\r\n- Note that our latest release (version 1.18.3) was made before this fix: https://github.com/huggingface/datasets/releases/tag/1.18.3\r\n\r\nIn the meantime, as @mariosasko explained, you can incorporate this \"fix\" by installing our library from the GitHub master branch:\r\n```shell\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\nThen, you should force the redownload of the data (before the fix, you are just downloading/caching the virus scan warning page, instead of the data file):\r\n```shell\r\ndata = datasets.load_dataset(\"wiki_lingua\", name=language, split=\"train[:2000]\", download_mode=\"force_redownload\")", "@albertvillanova by running:\r\n```\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\ndata = datasets.load_dataset(\"wiki_lingua\", name=language, split=\"train[:2000]\", download_mode=\"force_redownload\", ignore_verifications=True)\r\n```\r\n\r\nI had a pickle error **UnpicklingError: invalid load key, '<'** in this part of code both `locally and on google colab`:\r\n\r\n```\r\n\"\"\"Yields examples.\"\"\"\r\nwith open(filepath, \"rb\") as f:\r\n data = pickle.load(f)\r\nfor id_, row in enumerate(data.items()):\r\n yield id_, {\"url\": row[0], \"article\": self._process_article(row[1])}\r\n```\r\n", "This issue impacts many more datasets than the ones mention in this thread. Can we post # of downloads for each dataset by day (by successes and failures)? If so, it should be obvious which ones are failing.", "I can see this problem too in xcopa, unfortunately installing the latest master (1.18.4.dev0) doesn't work, @albertvillanova .\r\n\r\n```\r\nfrom datasets import load_dataset\r\ndataset = load_dataset(\"xcopa\", \"it\")\r\n```\r\n\r\nThrows\r\n\r\n```\r\nin verify_checksums(expected_checksums, recorded_checksums, verification_name)\r\n 38 if len(bad_urls) > 0:\r\n 39 error_msg = \"Checksums didn't match\" + for_verification_name + \":\\n\"\r\n---> 40 raise NonMatchingChecksumError(error_msg + str(bad_urls))\r\n 41 logger.info(\"All the checksums matched successfully\" + for_verification_name)\r\n 42 \r\n\r\nNonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://github.com/cambridgeltl/xcopa/archive/master.zip']\r\n```", "Hi @rafikg, I think that is another different issue. Let me check it... \r\n\r\nI guess maybe you are using a different Python version that the one the dataset owner used to create the pickle file...", "@kwchurch the datasets impacted for this specific issue are the ones which are hosted at Google Drive.", "@afcruzs-ms I think your issue is a different one, because that dataset is not hosted at Google Drive. Would you mind open another issue for that other problem, please? Thanks! :)", "@albertvillanova just to let you know that I tried it locally and on colab and it is the same error", "There are many many datasets on HugggingFace that are receiving this checksum error. Some of these datasets are very popular. There must be a way to track these errors, or to do regression testing. We don't want to catch each of these errors on each dataset, one at a time.", "@rafikg I am sorry, but I can't reproduce your issue. For me it works OK for all languages. See: https://colab.research.google.com/drive/1yIcLw1it118-TYE3ZlFmV7gJcsF6UCsH?usp=sharing", "@kwchurch the PR #3787 fixes this issue (generated by a change in Google Drive service) for ALL datasets with this issue. Once we make our next library release (in a couple of days), the fix will be accessible to all users that update our library from PyPI.", "By the way, @rafikg, I discovered the URL for Spanish was wrong. I've created a PR to fix it:\r\n- #3806 ", "I have the same problem with \"wider_face\" dataset. It seems that \"load_dataset\" function can not download the dataset from google drive.\r\n", "still getting this issue with datasets==2.2.2 for \r\ndataset_fever_original_dev = load_dataset('fever', \"v1.0\", split=\"labelled_dev\")\r\n(this one seems to be hosted by aws though)\r\n\r\nupdate: also tried to install from source to get the latest 2.2.3.dev0, but still get the error below (and also force-redownloaded)\r\n\r\nupdate2: Seems like this issues is linked to a change in the links in the specific fever datasets: https://fever.ai/\r\n\"28/04/2022\r\nDataset download URLs have changed\r\nDownload URLs for shared task data for FEVER, FEVER2.0 and FEVEROUS have been updated. New URLS begin with https://fever.ai/download/[task name]/[filename]. All resource pages have been updated with the new URLs. Previous dataset URLs may not work and should be updated if you require these in your scripts. \"\r\n\r\n=> I don't know how to update the links for HF datasets - would be great if someone could update them :) \r\n\r\n```\r\n\r\nDownloading and preparing dataset fever/v1.0 (download: 42.78 MiB, generated: 38.39 MiB, post-processed: Unknown size, total: 81.17 MiB) to /root/.cache/huggingface/datasets/fever/v1.0/1.0.0/956b0a9c4b05e126fd956be73e09da5710992b5c85c30f0e5e1c500bc6051d0a...\r\n\r\nDownloading data files: 100%\r\n6/6 [00:07<00:00, 1.21s/it]\r\nDownloading data:\r\n278/? [00:00<00:00, 2.34kB/s]\r\nDownloading data:\r\n278/? [00:00<00:00, 1.53kB/s]\r\nDownloading data:\r\n278/? [00:00<00:00, 7.43kB/s]\r\nDownloading data:\r\n278/? [00:00<00:00, 5.54kB/s]\r\nDownloading data:\r\n278/? [00:00<00:00, 6.19kB/s]\r\nDownloading data:\r\n278/? [00:00<00:00, 7.51kB/s]\r\nExtracting data files: 100%\r\n6/6 [00:00<00:00, 108.05it/s]\r\n\r\n---------------------------------------------------------------------------\r\n\r\nNonMatchingChecksumError Traceback (most recent call last)\r\n\r\n[<ipython-input-20-92ec5c728ecf>](https://localhost:8080/#) in <module>()\r\n 27 # get labels for fever-nli-dev from original fever - only works for dev\r\n 28 # \"(The labels for both dev and test are hidden but you can retrieve the label for dev using the cid and the original FEVER data.)\"\" https://github.com/easonnie/combine-FEVER-NSMN/blob/master/other_resources/nli_fever.md\r\n---> 29 dataset_fever_original_dev = load_dataset('fever', \"v1.0\", split=\"labelled_dev\")\r\n 30 df_fever_original_dev = pd.DataFrame(data={\"id\": dataset_fever_original_dev[\"id\"], \"label\": dataset_fever_original_dev[\"label\"], \"claim\": dataset_fever_original_dev[\"claim\"], \"evidence_id\": dataset_fever_original_dev[\"evidence_id\"]})\r\n 31 df_fever_dev = pd.merge(df_fever_dev, df_fever_original_dev, how=\"left\", left_on=\"cid\", right_on=\"id\")\r\n\r\n4 frames\r\n\r\n[/usr/local/lib/python3.7/dist-packages/datasets/utils/info_utils.py](https://localhost:8080/#) in verify_checksums(expected_checksums, recorded_checksums, verification_name)\r\n 38 if len(bad_urls) > 0:\r\n 39 error_msg = \"Checksums didn't match\" + for_verification_name + \":\\n\"\r\n---> 40 raise NonMatchingChecksumError(error_msg + str(bad_urls))\r\n 41 logger.info(\"All the checksums matched successfully\" + for_verification_name)\r\n 42 \r\n\r\nNonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://s3-eu-west-1.amazonaws.com/fever.public/train.jsonl', 'https://s3-eu-west-1.amazonaws.com/fever.public/shared_task_dev.jsonl', 'https://s3-eu-west-1.amazonaws.com/fever.public/shared_task_dev_public.jsonl', 'https://s3-eu-west-1.amazonaws.com/fever.public/shared_task_test.jsonl', 'https://s3-eu-west-1.amazonaws.com/fever.public/paper_dev.jsonl', 'https://s3-eu-west-1.amazonaws.com/fever.public/paper_test.jsonl']\r\n```\r\n", "I think this has to be fixed on the google drive side, but you also have to delete the bad stuff from your local cache. This is not a great design, but it is what it is.", "We have fixed the issues with the datasets:\r\n- wider_face: by hosting their data files on the HuggingFace Hub (CC: @HosseynGT)\r\n- fever: by updating to their new data URLs (CC: @MoritzLaurer)", "The yelp_review_full datasets has this problem as well and can't be fixed with the suggestion.", "This is a super-common failure mode. We really need to find a better workaround. My solution was to wait until the owner of the dataset in question did the right thing, and then I had to delete my cached versions of the datasets with the bad checksums. I don't understand why this happens. Would it be possible to maintain a copy of the most recent version that was known to work, and roll back to that automatically if the checksums fail? And if the checksums fail, couldn't the system automatically flush the cached versions with the bad checksums? It feels like we are blaming the provider of the dataset, when in fact, there are things that the system could do to ease the pain. Let's take these error messages seriously. There are too many of them involving too many different datasets.", "the [exams](https://huggingface.co/datasets/exams) dataset also has this issue and the provided fix above doesn't work", "Same for [DART dataset](https://huggingface.co/datasets/dart):\r\n```\r\nNonMatchingChecksumError: Checksums didn't match for dataset source files:\r\n['https://raw.githubusercontent.com/Yale-LILY/dart/master/data/v1.1.1/dart-v1.1.1-full-train.json', 'https://raw.githubusercontent.com/Yale-LILY/dart/master/data/v1.1.1/dart-v1.1.1-full-dev.json', 'https://raw.githubusercontent.com/Yale-LILY/dart/master/data/v1.1.1/dart-v1.1.1-full-test.json']\r\n```", "same for multi_news dataset", "- @thesofakillers the issue with `exams` was fixed on 16 Aug by this PR:\r\n - #4853\r\n- @Aktsvigun the issue with `dart` has been transferred to the Hub: https://huggingface.co/datasets/dart/discussions/1\r\n - and fixed by PR: https://huggingface.co/datasets/dart/discussions/2\r\n- @Carol-gutianle the issue with `multi_news` have been transferred to the Hub as well: https://huggingface.co/datasets/multi_news/discussions/1\r\n - not reproducible: maybe you should try to update `datasets`\r\n\r\nFor information to everybody, we are removing the checksum verifications (that were creating a bad user experience). This will be in place in the following weeks.", "auto_gptq is required for real quantization\r\n['/home/sam/Doctorproject/OmniQuant-main/main.py', '--model', '/home/sam/Doctorproject/OmniQuant-main/PATH/TO/LLaMA/llama-7b/', '--epochs', '20', '--output_dir', '/home/sam/Doctorproject/OmniQuant-main/outdir/llama-7b-w3a16/', '--eval_ppl', '--wbits', '3', '--abits', '16', '--lwc', '--net', 'llama-7b', '--aug_loss']\r\n[2024-03-13 17:58:48 root](main.py 262): INFO Namespace(model='/home/sam/Doctorproject/OmniQuant-main/PATH/TO/LLaMA/llama-7b/', cache_dir='./cache', output_dir='/home/sam/Doctorproject/OmniQuant-main/outdir/llama-7b-w3a16/', save_dir=None, resume=None, real_quant=False, calib_dataset='wikitext2', nsamples=128, batch_size=1, seed=2, tasks='', eval_ppl=True, num_fewshot=0, wbits=3, abits=16, group_size=None, alpha=0.5, let_lr=0.005, lwc_lr=0.01, wd=0, epochs=20, let=False, lwc=True, aug_loss=True, symmetric=False, disable_zero_point=False, a_dynamic_method='per_token', w_dynamic_method='per_channel', limit=-1, multigpu=False, deactive_amp=False, attn_implementation='eager', net='llama-7b', act_scales=None, act_shifts=None)\r\nLoading checkpoint shards: 0%| | 0/33 [00:00<?, ?it/s]/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/torch/_utils.py:776: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly. To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()\r\n return self.fget.__get__(instance, owner)()\r\nLoading checkpoint shards: 100%|██████████| 33/33 [00:11<00:00, 2.98it/s]\r\nvocab size: 32000\r\n[2024-03-13 17:58:59 root](main.py 331): INFO === start quantization ===\r\nget_wikitext2\r\n[2024-03-13 18:02:20 datasets.load](load.py 1586): WARNING Using the latest cached version of the module from /home/sam/.cache/huggingface/modules/datasets_modules/datasets/wikitext/a241db52902eaf2c6aa732210bead40c090019a499ceb13bcbfa3f8ab646a126 (last modified on Wed Mar 13 16:54:26 2024) since it couldn't be found locally at wikitext, or remotely on the Hugging Face Hub.\r\nUsing the latest cached version of the module from /home/sam/.cache/huggingface/modules/datasets_modules/datasets/wikitext/a241db52902eaf2c6aa732210bead40c090019a499ceb13bcbfa3f8ab646a126 (last modified on Wed Mar 13 16:54:26 2024) since it couldn't be found locally at wikitext, or remotely on the Hugging Face Hub.\r\nDownloading data: 243B [00:00, 877kB/s]\r\nGenerating test split: 0%| | 0/4358 [00:00<?, ? examples/s]\r\nTraceback (most recent call last):\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/builder.py\", line 1742, in _prepare_split_single\r\n for key, record in generator:\r\n File \"/home/sam/.cache/huggingface/modules/datasets_modules/datasets/wikitext/a241db52902eaf2c6aa732210bead40c090019a499ceb13bcbfa3f8ab646a126/wikitext.py\", line 187, in _generate_examples\r\n with open(data_file, encoding=\"utf-8\") as f:\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/streaming.py\", line 75, in wrapper\r\n return function(*args, download_config=download_config, **kwargs)\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/download/streaming_download_manager.py\", line 507, in xopen\r\n return open(main_hop, mode, *args, **kwargs)\r\nNotADirectoryError: [Errno 20] Not a directory: '/home/sam/.cache/huggingface/datasets/downloads/94be2a7b3fff32ae7379658c8d3821035b666baddad3a06d29b55ab3a4ab3115/wikitext-2-raw/wiki.test.raw'\r\n\r\nThe above exception was the direct cause of the following exception:\r\n\r\nTraceback (most recent call last):\r\n File \"/home/sam/Doctorproject/OmniQuant-main/main.py\", line 382, in <module>\r\n main()\r\n File \"/home/sam/Doctorproject/OmniQuant-main/main.py\", line 339, in main\r\n dataloader, _ = get_loaders(\r\n File \"/home/sam/Doctorproject/OmniQuant-main/datautils.py\", line 178, in get_loaders\r\n return get_wikitext2(nsamples, seed, seqlen, model)\r\n File \"/home/sam/Doctorproject/OmniQuant-main/datautils.py\", line 37, in get_wikitext2\r\n traindata = load_dataset(path='wikitext', name='wikitext-2-raw-v1', split='train', download_mode=\"force_redownload\")\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/load.py\", line 2598, in load_dataset\r\n builder_instance.download_and_prepare(\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/builder.py\", line 1021, in download_and_prepare\r\n self._download_and_prepare(\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/builder.py\", line 1783, in _download_and_prepare\r\n super()._download_and_prepare(\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/builder.py\", line 1116, in _download_and_prepare\r\n self._prepare_split(split_generator, **prepare_split_kwargs)\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/builder.py\", line 1621, in _prepare_split\r\n for job_id, done, content in self._prepare_split_single(\r\n File \"/home/sam/anaconda3/envs/omniquant/lib/python3.10/site-packages/datasets/builder.py\", line 1778, in _prepare_split_single\r\n raise DatasetGenerationError(\"An error occurred while generating the dataset\") from e\r\ndatasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset\r\n\r\n\r\n@albertvillanova @Y0mingZhang @kwchurch @HosseynGT @rafikg I tried the solutions you provided above, but none of them worked. Could you please give me some guidance\r\n" ]

[ "I see two options:\r\n1. drop the \"dev\" keyword since it can be considered too generic\r\n2. improve the pattern to something more reasonable, e.g. asking for a separator before and after \"dev\"\r\n```python\r\n[\"*[ ._-]dev[ ._-]*\", \"dev[ ._-]*\"]\r\n```\r\n\r\nI think 2. is nice. If we agree on this one we can even decide to require the separation for the other split keywords \"train\", \"test\" etc.", "Yes, I had something like that on mind: \"dev\" not being part of a word.\r\n```\r\n\"[^a-zA-Z]dev[^a-zA-Z]\"", "Is there a reason why we want that regex? It feels like something that'll still be an issue for some weird case. \"my_dataset_dev\" doesn't match your regex, \"my_dataset_validation\" doesn't either ... Why not always \"train\" unless specified?", "The regex is needed as part of our effort to make datasets configurable without code. In particular we define some generic dataset repository structures that users can follow\r\n\r\n> ```\r\n> \"[^a-zA-Z]*dev[^a-zA-Z]*\"\r\n> ```\r\n\r\nunfortunately our glob doesn't support \"^\": \r\n\r\nhttps://github.com/fsspec/filesystem_spec/blob/3e739db7e53f5b408319dcc9d11e92bc1f938902/fsspec/spec.py#L465-L479", "> \"my_dataset_dev\" doesn't match your regex, \"my_dataset_validation\" doesn't either ... Why not always \"train\" unless specified?\r\n\r\nAnd `my_dataset_dev.foo` would match the pattern, and we also have the same pattern but for the \"validation\" keyword so `my_dataset_validation.foo` would work too", "> The regex is needed as part of our effort to make datasets configurable without code\r\n\r\nThis feels like coding with the filename ^^'", "This is still much easier than having to write a full dataset script right ? :p" ]

[ "Once the PR merged into master and until our next `datasets` library release, you can get this fix by installing our library from the GitHub master branch:\r\n```shell\r\npip install git+https://github.com/huggingface/datasets#egg=datasets\r\n```\r\nThen, if you had previously tried to load the data and got the checksum error, you should force the redownload of the data (before the fix, you just downloaded and cached the virus scan warning page, instead of the data file):\r\n```shell\r\nload_dataset(\"...\", download_mode=\"force_redownload\")\r\n```" ]

[ "#self-assign", "@AngadSethi thanks for reporting and thanks for your PR!", "Glad to help @albertvillanova! Just fine-tuning the PR, will comment once I am able to get it up and running 😀", "Fixed by:\r\n- #3787" ]