Datasets:

safik

/

github-issues

like 0

https://api.github.com/repos/huggingface/datasets/issues/2805

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2805/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2805/comments

https://api.github.com/repos/huggingface/datasets/issues/2805/events

https://github.com/huggingface/datasets/pull/2805

MDExOlB1bGxSZXF1ZXN0NzEzMTc3MTI4

Fix streaming zip files from canonical datasets

closed

MEMBER

Previous PR #2798 fixed streaming remote zip files when passing the parameter `data_files`. However, that broke streaming zip files used in canonical `datasets` scripts, which normally have a subsequent `join()` (patched with `xjoin()`) after the `StreamingDownloadManager.download_and_extract()` is called. This PR fixes this issue and allows streaming zip files both from: - canonical datasets scripts and - data files.

https://api.github.com/repos/huggingface/datasets/issues/2805/timeline

https://api.github.com/repos/huggingface/datasets/issues/2804

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2804/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2804/comments

https://api.github.com/repos/huggingface/datasets/issues/2804/events

https://github.com/huggingface/datasets/pull/2804

MDExOlB1bGxSZXF1ZXN0NzEzMTA2NTMw

Add Food-101

closed

CONTRIBUTOR

Adds image classification dataset [Food-101](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/).

https://api.github.com/repos/huggingface/datasets/issues/2804/timeline

https://api.github.com/repos/huggingface/datasets/issues/2803

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2803/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2803/comments

https://api.github.com/repos/huggingface/datasets/issues/2803/events

https://github.com/huggingface/datasets/pull/2803

MDExOlB1bGxSZXF1ZXN0NzEyNzQxODMz

add stack exchange

closed

CONTRIBUTOR

stack exchange is part of EleutherAI/The Pile, but AFAIK, The Pile dataset blend all sub datasets together thus we are not able to use just one of its sub dataset from The Pile data. So I create an independent dataset using The Pile preliminary components. I also change default `timeout` to 100 seconds instead of 10 seconds, otherwise I keep getting read time out when downloading source data of stack exchange and cc100 dataset. When I was creating dataset card. I found there is room for creating / editing dataset card. I've made it an issue. #2797 Also I am wondering whether the import of The Pile dataset is actively undertaken (because I may need it recently)? #1675

https://api.github.com/repos/huggingface/datasets/issues/2803/timeline

https://api.github.com/repos/huggingface/datasets/issues/2802

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2802/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2802/comments

https://api.github.com/repos/huggingface/datasets/issues/2802/events

https://github.com/huggingface/datasets/pull/2802

MDExOlB1bGxSZXF1ZXN0NzEyNzM0MTc3

add openwebtext2

closed

CONTRIBUTOR

openwebtext2 is part of EleutherAI/The Pile, but AFAIK, The Pile dataset blend all sub datasets together thus we are not able to use just one of its sub dataset from The Pile data. So I create an independent dataset using The Pile preliminary components. When I was creating dataset card. I found there is room for creating / editing dataset card. I've made it an issue. #2797 Also I am wondering whether the import of The Pile dataset is actively undertaken (because I may need it recently)? #1675

https://api.github.com/repos/huggingface/datasets/issues/2802/timeline

https://api.github.com/repos/huggingface/datasets/issues/2801

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2801/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2801/comments

https://api.github.com/repos/huggingface/datasets/issues/2801/events

https://github.com/huggingface/datasets/pull/2801

MDExOlB1bGxSZXF1ZXN0NzEyNzMwODEz

add books3

closed

CONTRIBUTOR

books3 is part of EleutherAI/The Pile, but AFAIK, The Pile dataset blend all sub datasets together thus we are not able to use just one of its sub dataset from The Pile data. So I create an independent dataset using The Pile preliminary components. When I was creating dataset card. I found there is room for creating / editing dataset card. I've made it an issue. #2797 Also I am wondering whether the import of The Pile dataset is actively undertaken (because I may need it recently)? #1675

https://api.github.com/repos/huggingface/datasets/issues/2801/timeline

https://api.github.com/repos/huggingface/datasets/issues/2800

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2800/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2800/comments

https://api.github.com/repos/huggingface/datasets/issues/2800/events

https://github.com/huggingface/datasets/pull/2800

MDExOlB1bGxSZXF1ZXN0NzEyNzExNTcx

Support streaming tar files

closed

MEMBER

This PR adds support to stream tar files by using the `fsspec` tar protocol. It also uses the custom `readline` implemented in PR #2786. The corresponding test is implemented in PR #2786.

https://api.github.com/repos/huggingface/datasets/issues/2800/timeline

https://api.github.com/repos/huggingface/datasets/issues/2799

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2799/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2799/comments

https://api.github.com/repos/huggingface/datasets/issues/2799/events

https://github.com/huggingface/datasets/issues/2799

MDU6SXNzdWU5NzA1MDczNTE=

Loading JSON throws ArrowNotImplementedError

closed

MEMBER

## Describe the bug I have created a [dataset](https://huggingface.co/datasets/lewtun/github-issues-test) of GitHub issues in line-separated JSON format and am finding that I cannot load it with the `json` loading script (see stack trace below). Curiously, there is no problem loading the dataset with `pandas` which suggests some incorrect type inference is being made on the `datasets` side. For example, the stack trace indicates that some URL fields are being parsed as timestamps. You can find a Colab notebook which reproduces the error [here](https://colab.research.google.com/drive/1YUCM0j1vx5ZrouQbYSzal6RwB4-Aoh4o?usp=sharing). **Edit:** If one repeatedly tries to load the dataset, it _eventually_ works but I think it would still be good to understand why it fails in the first place :) ## Steps to reproduce the bug ```python from datasets import load_dataset from huggingface_hub import hf_hub_url import pandas as pd # returns https://huggingface.co/datasets/lewtun/github-issues-test/resolve/main/issues-datasets.jsonl data_files = hf_hub_url(repo_id="lewtun/github-issues-test", filename="issues-datasets.jsonl", repo_type="dataset") # throws ArrowNotImplementedError dset = load_dataset("json", data_files=data_files, split="test") # no problem with pandas ... df = pd.read_json(data_files, orient="records", lines=True) df.head() ``` ## Expected results I can load any line-separated JSON file, similar to `pandas`. ## Actual results ``` --------------------------------------------------------------------------- ArrowNotImplementedError Traceback (most recent call last) <ipython-input-7-5b8e82b6c3a2> in <module>() ----> 1 dset = load_dataset("json", data_files=data_files, split="test") 9 frames /usr/local/lib/python3.7/dist-packages/pyarrow/error.pxi in pyarrow.lib.check_status() ArrowNotImplementedError: JSON conversion to struct<url: timestamp[s], html_url: timestamp[s], labels_url: timestamp[s], id: int64, node_id: timestamp[s], number: int64, title: timestamp[s], description: timestamp[s], creator: struct<login: timestamp[s], id: int64, node_id: timestamp[s], avatar_url: timestamp[s], gravatar_id: timestamp[s], url: timestamp[s], html_url: timestamp[s], followers_url: timestamp[s], following_url: timestamp[s], gists_url: timestamp[s], starred_url: timestamp[s], subscriptions_url: timestamp[s], organizations_url: timestamp[s], repos_url: timestamp[s], events_url: timestamp[s], received_events_url: timestamp[s], type: timestamp[s], site_admin: bool>, open_issues: int64, closed_issues: int64, state: timestamp[s], created_at: timestamp[s], updated_at: timestamp[s], due_on: timestamp[s], closed_at: timestamp[s]> is not supported ``` ## Environment info <!-- You can run the command `datasets-cli env` and copy-and-paste its output below. --> - `datasets` version: 1.11.0 - Platform: Linux-5.4.104+-x86_64-with-Ubuntu-18.04-bionic - Python version: 3.7.11 - PyArrow version: 3.0.0

https://api.github.com/repos/huggingface/datasets/issues/2799/timeline

https://api.github.com/repos/huggingface/datasets/issues/2798

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2798/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2798/comments

https://api.github.com/repos/huggingface/datasets/issues/2798/events

https://github.com/huggingface/datasets/pull/2798

MDExOlB1bGxSZXF1ZXN0NzEyNDM3ODc2

Fix streaming zip files

closed

MEMBER

Currently, streaming remote zip data files gives `FileNotFoundError` message: ```python data_files = f"https://huggingface.co/datasets/albertvillanova/datasets-tests-compression/resolve/main/sample.zip" ds = load_dataset("json", split="train", data_files=data_files, streaming=True) next(iter(ds)) ``` This PR fixes it by adding a glob string. The corresponding test is implemented in PR #2786.

https://api.github.com/repos/huggingface/datasets/issues/2798/timeline

https://api.github.com/repos/huggingface/datasets/issues/2797

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2797/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2797/comments

https://api.github.com/repos/huggingface/datasets/issues/2797/events

https://github.com/huggingface/datasets/issues/2797

MDU6SXNzdWU5NzAzMzE2MzQ=

Make creating/editing dataset cards easier, by editing on site and dumping info from test command.

open

CONTRIBUTOR

**Is your feature request related to a problem? Please describe.** Creating and editing dataset cards should be but not that easy - If other else know Some information I don't know (bias of dataset, dataset curation, supported dataset, ...), he/she should know the description on hf.co comes from README.md under github huggingface/datasets/datasets/the dataset, and willing to make a pr to add or fix information. - Many information is also saved in `dataset_info.json` (citaion, description), but still need to write it down to README.md again. - Contributor need to pip install and start a local server just for tagging the dataset's size. And contributor may be creating the dataset on lab's server, which can't open a browser. - if any one proposes a new tag, it doesn't show in the list that another creator see. (a stackoverflow way may be ideal) - dataset card generator web app doesn't generate the necessary subsecion `Contributions` for us. **Describe the solution you'd like** - Everyone (or at least the author/contributor) can edit the description, information, tags of the dataset, on hf.co website. Just like wikipedia+stackoverflow - We can infer the actual data size, citation, data instance, ... from `dataset_info.json` and `dataset.arrow` via `dataset-cli test`

https://api.github.com/repos/huggingface/datasets/issues/2797/timeline

https://api.github.com/repos/huggingface/datasets/issues/2796

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2796/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2796/comments

https://api.github.com/repos/huggingface/datasets/issues/2796/events

https://github.com/huggingface/datasets/pull/2796

MDExOlB1bGxSZXF1ZXN0NzEyMjE1ODM2

add cedr dataset

closed

CONTRIBUTOR

https://api.github.com/repos/huggingface/datasets/issues/2796/timeline

https://api.github.com/repos/huggingface/datasets/issues/2794

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2794/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2794/comments

https://api.github.com/repos/huggingface/datasets/issues/2794/events

https://github.com/huggingface/datasets/issues/2794

MDU6SXNzdWU5Njk3Mjg1NDU=

Warnings and documentation about pickling incorrect

open

NONE

## Describe the bug I have a docs bug and a closely related docs enhancement suggestion! ### Bug The warning and documentation say "either `dill` or `pickle`" for fingerprinting. But it seems that `dill`, which is installed by `datasets` by default, _must_ work, or else the fingerprinting fails. Warning: https://github.com/huggingface/datasets/blob/450b9174765374111e5c6daab0ed294bc3d9b639/src/datasets/fingerprint.py#L262 Docs: > For a transform to be hashable, it needs to be pickleable using dill or pickle. > – [docs](https://huggingface.co/docs/datasets/processing.html#fingerprinting) For my code, `pickle` works, but `dill` fails. The `dill` failure has already been reported in https://github.com/huggingface/datasets/issues/2643. However, the `dill` failure causes a hashing failure in the datasets library, without any backing off to `pickle`. This implies that it's not the case that either `dill` **or** `pickle` can work, but that `dill` must work if it is installed. I think this is more accurate wording, since it is installed and used by default: https://github.com/huggingface/datasets/blob/c93525dc291346e54212567fa72d7d607befe937/setup.py#L83 ... and the hashing will fail if it fails. ### Enhancement I think it'd be very helpful to add to the documentation how to debug hashing failures. It took me a while to figure out how to diagnose this. There is a very nice two-liner by @lhoestq in https://github.com/huggingface/datasets/issues/2516#issuecomment-865173139: ```python from datasets.fingerprint import Hasher Hasher.hash(my_object) ``` I think add this to the docs will help future users quickly debug any hashing troubles of their own :-) ## Steps to reproduce the bug `dill` but not `pickle` hashing failure in https://github.com/huggingface/datasets/issues/2643 ## Expected results If either `dill` or `pickle` can successfully hash, the hashing will succeed. ## Actual results If `dill` or `pickle` cannot hash, the hashing fails. ## Environment info - `datasets` version: 1.9.0 - Platform: Linux-5.8.0-1038-gcp-x86_64-with-glibc2.31 - Python version: 3.9.6 - PyArrow version: 4.0.1

https://api.github.com/repos/huggingface/datasets/issues/2794/timeline

https://api.github.com/repos/huggingface/datasets/issues/2793

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2793/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2793/comments

https://api.github.com/repos/huggingface/datasets/issues/2793/events

https://github.com/huggingface/datasets/pull/2793

MDExOlB1bGxSZXF1ZXN0NzExMDQ4NDY2

Fix type hint for data_files

closed

MEMBER

Fix type hint for `data_files` in signatures and docstrings.

https://api.github.com/repos/huggingface/datasets/issues/2793/timeline

https://api.github.com/repos/huggingface/datasets/issues/2792

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2792/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2792/comments

https://api.github.com/repos/huggingface/datasets/issues/2792/events

https://github.com/huggingface/datasets/pull/2792

MDExOlB1bGxSZXF1ZXN0NzEwNzUyMjc0

Update: GooAQ - add train/val/test splits

closed

CONTRIBUTOR

[GooAQ](https://github.com/allenai/gooaq) dataset was recently updated after splits were added for the same. This PR contains new updated GooAQ with train/val/test splits and updated README as well.

https://api.github.com/repos/huggingface/datasets/issues/2792/timeline

https://api.github.com/repos/huggingface/datasets/issues/2791

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2791/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2791/comments

https://api.github.com/repos/huggingface/datasets/issues/2791/events

https://github.com/huggingface/datasets/pull/2791

MDExOlB1bGxSZXF1ZXN0NzEwNDgxNDAy

Fix typo in cnn_dailymail

closed

CONTRIBUTOR

https://api.github.com/repos/huggingface/datasets/issues/2791/timeline

https://api.github.com/repos/huggingface/datasets/issues/2790

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2790/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2790/comments

https://api.github.com/repos/huggingface/datasets/issues/2790/events

https://github.com/huggingface/datasets/pull/2790

MDExOlB1bGxSZXF1ZXN0NzA5OTI3NjM2

Fix typo in test_dataset_common

closed

CONTRIBUTOR

https://api.github.com/repos/huggingface/datasets/issues/2790/timeline

https://api.github.com/repos/huggingface/datasets/issues/2789

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2789/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2789/comments

https://api.github.com/repos/huggingface/datasets/issues/2789/events

https://github.com/huggingface/datasets/pull/2789

MDExOlB1bGxSZXF1ZXN0NzA5NTQwMzY5

Updated dataset description of DaNE

closed

CONTRIBUTOR

https://api.github.com/repos/huggingface/datasets/issues/2789/timeline

https://api.github.com/repos/huggingface/datasets/issues/2788

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2788/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2788/comments

https://api.github.com/repos/huggingface/datasets/issues/2788/events

https://github.com/huggingface/datasets/issues/2788

MDU6SXNzdWU5NjcxNDkzODk=

How to sample every file in a list of files making up a split in a dataset when loading?

open

NONE

I am loading a dataset with multiple train, test, and validation files like this: ``` data_files_dict = { "train": [train_file1, train_file2], "test": [test_file1, test_file2], "val": [val_file1, val_file2] } dataset = datasets.load_dataset( "csv", data_files=data_files_dict, split=['train[:8]', 'test[:8]', 'val[:8]'] ) ``` However, this only selects the first 8 rows from train_file1, test_file1, val_file1, since they are the first files in the lists. I'm trying to formulate a split argument that can sample from each file specified in my list of files that make up each split. Is this type of splitting supported? If so, how can I do it?

https://api.github.com/repos/huggingface/datasets/issues/2788/timeline

https://api.github.com/repos/huggingface/datasets/issues/2787

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2787/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2787/comments

https://api.github.com/repos/huggingface/datasets/issues/2787/events

https://github.com/huggingface/datasets/issues/2787

MDU6SXNzdWU5NjcwMTg0MDY=

ConnectionError: Couldn't reach https://raw.githubusercontent.com

closed

NONE

Hello, I am trying to run run_glue.py and it gives me this error - Traceback (most recent call last): File "E:/BERT/pytorch_hugging/transformers/examples/pytorch/text-classification/run_glue.py", line 546, in <module> main() File "E:/BERT/pytorch_hugging/transformers/examples/pytorch/text-classification/run_glue.py", line 250, in main datasets = load_dataset("glue", data_args.task_name, cache_dir=model_args.cache_dir) File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\load.py", line 718, in load_dataset use_auth_token=use_auth_token, File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\load.py", line 320, in prepare_module local_path = cached_path(file_path, download_config=download_config) File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\utils\file_utils.py", line 291, in cached_path use_auth_token=download_config.use_auth_token, File "C:\install\Anaconda3\envs\huggingface\lib\site-packages\datasets\utils\file_utils.py", line 623, in get_from_cache raise ConnectionError("Couldn't reach {}".format(url)) ConnectionError: Couldn't reach https://raw.githubusercontent.com/huggingface/datasets/1.7.0/datasets/glue/glue.py Trying to do python run_glue.py --model_name_or_path bert-base-cased --task_name mrpc --do_train --do_eval --max_seq_length 128 --per_device_train_batch_size 32 --learning_rate 2e-5 --num_train_epochs 3 --output_dir ./tmp/mrpc/ Is this something on my end? From what I can tell, this was re-fixeded by @fullyz a few months ago. Thank you!

https://api.github.com/repos/huggingface/datasets/issues/2787/timeline

https://api.github.com/repos/huggingface/datasets/issues/2786

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2786/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2786/comments

https://api.github.com/repos/huggingface/datasets/issues/2786/events

https://github.com/huggingface/datasets/pull/2786

MDExOlB1bGxSZXF1ZXN0NzA4NTQwMzU0

Support streaming compressed files

closed

MEMBER

Add support to stream compressed files (current options in fsspec): - bz2 - lz4 - xz - zstd cc: @lewtun

https://api.github.com/repos/huggingface/datasets/issues/2786/timeline

https://api.github.com/repos/huggingface/datasets/issues/2783

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2783/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2783/comments

https://api.github.com/repos/huggingface/datasets/issues/2783/events

https://github.com/huggingface/datasets/pull/2783

MDExOlB1bGxSZXF1ZXN0NzA3NzcxOTM3

Add KS task to SUPERB

closed

MEMBER

Add the KS (keyword spotting) task as described in the [SUPERB paper](https://arxiv.org/abs/2105.01051). - [s3prl instructions](https://github.com/s3prl/s3prl/blob/master/s3prl/downstream/README.md#ks-keyword-spotting) - [s3prl implementation](https://github.com/s3prl/s3prl/blob/master/s3prl/downstream/speech_commands/dataset.py) - [TFDS implementation](https://github.com/tensorflow/datasets/blob/master/tensorflow_datasets/audio/speech_commands.py) Some notable quirks: - The dataset is originally single-archive (train+val+test all in one), but the test set has a "canonical" distribution in a separate archive, which is also used here (see `_split_ks_files()`). - The `_background_noise_`/`_silence_` audio files are much longer than others, so they require some sort of slicing for downstream training. I decided to leave the implementation of that up to the users, since TFDS and s3prl take different approaches (either slicing wavs deterministically, or subsampling randomly at runtime) Related to #2619.

https://api.github.com/repos/huggingface/datasets/issues/2783/timeline

https://api.github.com/repos/huggingface/datasets/issues/2782

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2782/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2782/comments

https://api.github.com/repos/huggingface/datasets/issues/2782/events

https://github.com/huggingface/datasets/pull/2782

MDExOlB1bGxSZXF1ZXN0NzA3MjQ5NDE5

Fix renaming of corpus_bleu args

closed

MEMBER

Last `sacrebleu` release (v2.0.0) has renamed `sacrebleu.corpus_bleu` args from `(sys_stream, ref_streams)` to `(hipotheses, references)`: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15 This PR passes the args without parameter names, so that it is valid for all versions of `sacrebleu`. This is a partial hotfix of #2781. Close #2781.

https://api.github.com/repos/huggingface/datasets/issues/2782/timeline

https://api.github.com/repos/huggingface/datasets/issues/2781

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2781/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2781/comments

https://api.github.com/repos/huggingface/datasets/issues/2781/events

https://github.com/huggingface/datasets/issues/2781

MDU6SXNzdWU5NjQ4MDUzNTE=

Latest v2.0.0 release of sacrebleu has broken some metrics

closed

MEMBER

## Describe the bug After `sacrebleu` v2.0.0 release (see changes here: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15), some of `datasets` metrics are broken: - Default tokenizer `sacrebleu.DEFAULT_TOKENIZER` no longer exists: - #2739 - #2778 - Bleu tokenizers are no longer accessible with `sacrebleu.TOKENIZERS`: - #2779 - `corpus_bleu` args have been renamed from `(sys_stream, ref_streams)` to `(hipotheses, references)`: - #2782

https://api.github.com/repos/huggingface/datasets/issues/2781/timeline

https://api.github.com/repos/huggingface/datasets/issues/2780

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2780/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2780/comments

https://api.github.com/repos/huggingface/datasets/issues/2780/events

https://github.com/huggingface/datasets/pull/2780

MDExOlB1bGxSZXF1ZXN0NzA3MTk2NjA3

VIVOS dataset for Vietnamese ASR

closed

CONTRIBUTOR

https://api.github.com/repos/huggingface/datasets/issues/2780/timeline

https://api.github.com/repos/huggingface/datasets/issues/2779

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2779/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2779/comments

https://api.github.com/repos/huggingface/datasets/issues/2779/events

https://github.com/huggingface/datasets/pull/2779

MDExOlB1bGxSZXF1ZXN0NzA3MTgwNTgw

Fix sacrebleu tokenizers

closed

MEMBER

Last `sacrebleu` release (v2.0.0) has removed `sacrebleu.TOKENIZERS`: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15 This PR makes a hot fix of the bug by using a private function in `sacrebleu`: `sacrebleu.metrics.bleu._get_tokenizer()`. Eventually, this should be further fixed in order to use only public functions. This is a partial hotfix of #2781.

https://api.github.com/repos/huggingface/datasets/issues/2779/timeline

https://api.github.com/repos/huggingface/datasets/issues/2778

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2778/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2778/comments

https://api.github.com/repos/huggingface/datasets/issues/2778/events

https://github.com/huggingface/datasets/pull/2778

MDExOlB1bGxSZXF1ZXN0NzA3MTQ5MTk2

Do not pass tokenize to sacrebleu

closed

MEMBER

Last `sacrebleu` release (v2.0.0) has removed `sacrebleu.DEFAULT_TOKENIZER`: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15 This PR does not pass `tokenize` to `sacrebleu` (note that the user cannot pass it anyway) and `sacrebleu` will use its default, no matter where it is and how it is called. Related to #2739. This is a partial hotfix of #2781.

https://api.github.com/repos/huggingface/datasets/issues/2778/timeline

https://api.github.com/repos/huggingface/datasets/issues/2777

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2777/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2777/comments

https://api.github.com/repos/huggingface/datasets/issues/2777/events

https://github.com/huggingface/datasets/pull/2777

MDExOlB1bGxSZXF1ZXN0NzA3MTEzNzg3

Use packaging to handle versions

closed

MEMBER

Use packaging module to handle/validate/check versions of Python packages. Related to #2769.

https://api.github.com/repos/huggingface/datasets/issues/2777/timeline

https://api.github.com/repos/huggingface/datasets/issues/2776

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2776/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2776/comments

https://api.github.com/repos/huggingface/datasets/issues/2776/events

https://github.com/huggingface/datasets/issues/2776

MDU6SXNzdWU5NjQ0MDA1OTY=

document `config.HF_DATASETS_OFFLINE` and precedence

open

MEMBER

https://github.com/huggingface/datasets/pull/1976 implemented `HF_DATASETS_OFFLINE`, but: 1. `config.HF_DATASETS_OFFLINE` is not documented 2. the precedence is not documented (env, config) I'm thinking it probably should be similar to what it says https://huggingface.co/docs/datasets/loading_datasets.html#from-the-huggingface-hub about `datasets.config.IN_MEMORY_MAX_SIZE`: Quote: > The default in 🤗 Datasets is to memory-map the dataset on disk unless you set datasets.config.IN_MEMORY_MAX_SIZE different from 0 bytes (default). In that case, the dataset will be copied in-memory if its size is smaller than datasets.config.IN_MEMORY_MAX_SIZE bytes, and memory-mapped otherwise. This behavior can be enabled by setting either the configuration option datasets.config.IN_MEMORY_MAX_SIZE (higher precedence) or the environment variable HF_DATASETS_IN_MEMORY_MAX_SIZE (lower precedence) to nonzero. Context: trying to use `config.HF_DATASETS_OFFLINE` here: https://github.com/bigscience-workshop/Megatron-DeepSpeed/pull/48 but are uncertain if it's safe, since it's not documented as a public API. Thank you! @lhoestq, @albertvillanova

https://api.github.com/repos/huggingface/datasets/issues/2776/timeline

https://api.github.com/repos/huggingface/datasets/issues/2775

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2775/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2775/comments

https://api.github.com/repos/huggingface/datasets/issues/2775/events

https://github.com/huggingface/datasets/issues/2775

MDU6SXNzdWU5NjQzMDM2MjY=

`generate_random_fingerprint()` deterministic with 🤗Transformers' `set_seed()`

open

NONE

## Describe the bug **Update:** I dug into this to try to reproduce the underlying issue, and I believe it's that `set_seed()` from the `transformers` library makes the "random" fingerprint identical each time. I believe this is still a bug, because `datasets` is used exactly this way in `transformers` after `set_seed()` has been called, and I think that using `set_seed()` is a standard procedure to aid reproducibility. I've added more details to reproduce this below. Hi there! I'm using my own local dataset and custom preprocessing function. My preprocessing function seems to be unpickle-able, perhaps because it is from a closure (will debug this separately). I get this warning, which is expected: https://github.com/huggingface/datasets/blob/450b9174765374111e5c6daab0ed294bc3d9b639/src/datasets/fingerprint.py#L260-L265 However, what's not expected is that the `datasets` actually _does_ seem to cache and reuse this dataset between runs! After that line, the next thing that's logged looks like: ```text Loading cached processed dataset at /home/xxx/.cache/huggingface/datasets/csv/default-xxx/0.0.0/xxx/cache-xxx.arrow ``` The path is exactly the same each run (e.g., last 26 runs). This becomes a problem because I'll pass in the `--max_eval_samples` flag to the HuggingFace example script I'm running off of ([run_swag.py](https://github.com/huggingface/transformers/blob/master/examples/pytorch/multiple-choice/run_swag.py)). The fact that the cached dataset is reused means this flag gets ignored. I'll try to load 100 examples, and it will load the full cached 1,000,000. I think that https://github.com/huggingface/datasets/blob/450b9174765374111e5c6daab0ed294bc3d9b639/src/datasets/fingerprint.py#L248 ... is actually consistent because randomness is being controlled in HuggingFace/Transformers for reproducibility. I've added a demo of this below. ## Steps to reproduce the bug ```python # Contents of print_fingerprint.py from transformers import set_seed from datasets.fingerprint import generate_random_fingerprint set_seed(42) print(generate_random_fingerprint()) ``` ```bash for i in {0..10}; do python print_fingerprint.py done 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d 1c80317fa3b1799d ``` ## Expected results After the "random hash" warning is emitted, a random hash is generated, and no outdated cached datasets are reused. ## Actual results After the "random hash" warning is emitted, an identical hash is generated each time, and an outdated cached dataset is reused each run. ## Environment info <!-- You can run the command `datasets-cli env` and copy-and-paste its output below. --> - `datasets` version: 1.9.0 - Platform: Linux-5.8.0-1038-gcp-x86_64-with-glibc2.31 - Python version: 3.9.6 - PyArrow version: 4.0.1

https://api.github.com/repos/huggingface/datasets/issues/2775/timeline

https://api.github.com/repos/huggingface/datasets/issues/2774

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2774/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2774/comments

https://api.github.com/repos/huggingface/datasets/issues/2774/events

https://github.com/huggingface/datasets/pull/2774

MDExOlB1bGxSZXF1ZXN0NzA2NDY2MDc0

Prevent .map from using multiprocessing when loading from cache

closed

MEMBER

## Context On our setup, we use different setup to train vs proprocessing datasets. Usually we are able to obtain a high number of cpus to preprocess, which allows us to use `num_proc` however we can't use as many during training phase. Currently if we use `num_proc={whatever the preprocessing value was}` we load from cache, but we get: ``` Traceback (most recent call last): File "lib/python3.8/site-packages/multiprocess/pool.py", line 131, in worker put((job, i, result)) File "lib/python3.8/site-packages/multiprocess/queues.py", line 371, in put self._writer.send_bytes(obj) File "lib/python3.8/site-packages/multiprocess/connection.py", line 203, in send_bytes self._send_bytes(m[offset:offset + size]) File "lib/python3.8/site-packages/multiprocess/connection.py", line 414, in _send_bytes self._send(header + buf) File "lib/python3.8/site-packages/multiprocess/connection.py", line 371, in _send n = write(self._handle, buf) BrokenPipeError: [Errno 32] Broken pipe ``` Our current guess, is that we're spawning too many processes compared to the number of cpus available, and it's running OOM. Also we're loading this in DDP setting which means that for each gpu, I need to spawn a high number of processes to match the preprocessing fingerprint. Instead what we suggest: - Allow loading shard sequentially, sharing the same fingerprint as the multiprocessed one, in order to leverage multiprocessing when we actually generate the cache, and remove it when loading from cache. ## Current issues ~I'm having a hard time making fingerprints match. For some reason, the multiprocessing and the sequential version generate two different hash.~ **EDIT**: Turns out multiprocessing and sequential have different `transform` value for fingerprinting (check `fingerprint_transform`) when running `_map_single`: - sequential : `datasets.arrow_dataset.Dataset._map_single` - multiprocessing: `datasets.arrow_dataset._map_single` This discrepancy is caused by multiprocessing pickling the transformer function, it doesn't seem to keep the `Dataset` hierarchy. I'm still unclear on why `func.__qual_name__` isn't handled correctly in multiprocessing. But replacing `__qualname__` by `__name__` fixes the issue. ## What was done ~We try to prevent the usage of multiprocessing when loading a dataset. Instead we load all cached shards sequentially.~ I couldn't find a nice way to obtain the cached_file_name and check they all exist before deciding to use the multiprocessing flow or not. Instead I expose an optional boolean `sequential` in `map` method. ## TODO - [x] Check that the multiprocessed version and the sequential version output the same output - [x] Check that sequential can load multiprocessed - [x] Check that multiprocessed can load sequential ## Test ```python from datasets import load_dataset from multiprocessing import Pool import random def process(batch, rng): length = len(batch["text"]) return {**batch, "processed_text": [f"PROCESSED {rng.random()}" for _ in range(length)]} dataset = load_dataset("stas/openwebtext-10k", split="train") print(dataset.column_names) print(type(dataset)) rng = random.Random(42) dataset1 = dataset.map(process, batched=True, batch_size=50, num_proc=4, fn_kwargs={"rng": rng}) # This one should be loaded from cache rng = random.Random(42) dataset2 = dataset.map(process, batched=True, batch_size=50, num_proc=4, fn_kwargs={"rng": rng}, sequential=True) # Just to check that the random generator was correct print(dataset1[-1]["processed_text"]) print(dataset2[-1]["processed_text"]) ``` ## Other solutions I chose to load everything sequentially, but we can probably find a way to load shards in parallel using another number of workers (essentially this would be an argument not used for fingerprinting, allowing to allow `m` shards using `n` processes, which would be very useful when same dataset have to be loaded on two different setup, and we still want to leverage cache). Also we can use a env variable similarly to `TOKENIZERS_PARALLELISM` as this seems generally setup related (though this changes slightly if we use multiprocessing). cc @lhoestq (since I had asked you previously on `num_proc` being used for fingerprinting). Don't know if this is acceptable.

https://api.github.com/repos/huggingface/datasets/issues/2774/timeline

https://api.github.com/repos/huggingface/datasets/issues/2773

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2773/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2773/comments

https://api.github.com/repos/huggingface/datasets/issues/2773/events

https://github.com/huggingface/datasets/issues/2773

MDU6SXNzdWU5NjM3MzA0OTc=

Remove dataset_infos.json

open

MEMBER

**Is your feature request related to a problem? Please describe.** As discussed, there are infos in the `dataset_infos.json` which are redundant and we could have them only in the README file. Others could be migrated to the README, like: "dataset_size", "size_in_bytes", "download_size", "splits.split_name.[num_bytes, num_examples]",... However, there are others that do not seem too meaningful in the README, like the checksums. **Describe the solution you'd like** Open a discussion to decide what to do with the `dataset_infos.json` files: which information to be migrated and/or which information to be kept. cc: @julien-c @lhoestq

https://api.github.com/repos/huggingface/datasets/issues/2773/timeline

https://api.github.com/repos/huggingface/datasets/issues/2772

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2772/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2772/comments

https://api.github.com/repos/huggingface/datasets/issues/2772/events

https://github.com/huggingface/datasets/issues/2772

MDU6SXNzdWU5NjMzNDg4MzQ=

Remove returned feature constrain

open

NONE

In the current version, the returned value of the map function has to be list or ndarray. However, this makes it unsuitable for many tasks. In NLP, many features are sparse like verb words, noun chunks, if we want to assign different values to different words, which will result in a large sparse matrix if we only score useful words like verb words. Mostly, when using it on large scale, saving it as a whole takes a lot of disk storage and making it hard to read, the normal method is saving it in sparse form. However, the NumPy does not support sparse, therefore I have to use PyTorch or scipy to transform a matrix into special sparse form, which is not a form that can be transformed into list or ndarry. This violates the feature constraints of the map function. I do appreciate the convenience of Datasets package, but I do not think the compulsory datatype constrain is necessary, in some cases, we just cannot transform it into a list or ndarray due to some reasons. Any way to fix this? Or what I can do to disable the compulsory datatype constrain?

https://api.github.com/repos/huggingface/datasets/issues/2772/timeline

https://api.github.com/repos/huggingface/datasets/issues/2771

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2771/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2771/comments

https://api.github.com/repos/huggingface/datasets/issues/2771/events

https://github.com/huggingface/datasets/pull/2771

MDExOlB1bGxSZXF1ZXN0NzA1OTExMDMw

[WIP][Common Voice 7] Add common voice 7.0

closed

MEMBER

This PR allows to load the new common voice dataset manually as explained when doing: ```python from datasets import load_dataset ds = load_dataset("./datasets/datasets/common_voice_7", "ab") ``` => ``` Please follow the manual download instructions: You need to manually the dataset from `https://commonvoice.mozilla.org/en/datasets`. Make sure you choose the version `Common Voice Corpus 7.0`. Choose a language of your choice and find the corresponding language-id, *e.g.*, `Abkhaz` with language-id `ab`. The following language-ids are available: ['ab', 'ar', 'as', 'az', 'ba', 'bas', 'be', 'bg', 'br', 'ca', 'cnh', 'cs', 'cv', 'cy', 'de', 'dv', 'el', 'en', 'eo', 'es', 'et', 'eu', 'fa', 'fi', 'fr', 'fy-NL', 'ga-IE', 'gl', 'gn', 'ha', 'hi', 'hsb', 'hu', 'hy-AM', 'ia', 'id', 'it', 'ja', 'ka', 'kab', 'kk', 'kmr', 'ky', 'lg', 'lt', 'lv', 'mn', 'mt', 'nl', 'or', 'pa-IN', 'pl', 'pt', 'rm-sursilv', 'rm-vallader', 'ro', 'ru', 'rw', 'sah', 'sk', 'sl', 'sr', 'sv-SE', 'ta', 'th', 'tr', 'tt', 'ug', 'uk', 'ur', 'uz', 'vi', 'vot', 'zh-CN', 'zh-HK', 'zh-TW'] Next, you will have to enter your email address to download the dataset in the `tar.gz` format. Save the file under <path-to-file>. The file should then be extracted with: ``tar -xvzf <path-to-file>`` which will extract a folder called ``cv-corpus-7.0-2021-07-21``. The dataset can then be loaded with `datasets.load_dataset("common_voice", <language-id>, data_dir="<path-to-'cv-corpus-7.0-2021-07-21'-folder>", ignore_verifications=True). ``` Having followed those instructions one can then download the data as follows: ```python from datasets import load_dataset ds = load_dataset("./datasets/datasets/common_voice_7", "ab", data_dir="./cv-corpus-7.0-2021-07-21/", ignore_verifications=True) ``` ## TODO - [ ] Discuss naming. Is the name ok here "common_voice_7"? The dataset script differs only really in one point from `common_voice.py` in that all the metadata is different (more hours etc...) and that it has to use manual data dir for now - [ ] Ideally we should get a bundled download link. For `common_voice.py` there is a bundled download link: `https://voice-prod-bundler-ee1969a6ce8178826482b88e843c335139bd3fb4.s3.amazonaws.com/cv-corpus-6.1-2020-12-11/{}.tar.gz` that allows one to directly download the data. However such a link is missing for Common Voice 7. I guess we should try to contact common voice about it and ask whether we could host the data or help otherwise somehow. See: https://github.com/common-voice/common-voice-bundler/issues/15 cc @yjernite - [ ] I did not compute the dataset.json and it would mean that I'd have to download 76 datasets totalling around 1TB manually before running the checksum command. This just takes too much time. For now the user will have to add a `ignore_verifications=True` to download the data. This step would also be much easier if we could get a bundled link - [ ] Add dummy data

https://api.github.com/repos/huggingface/datasets/issues/2771/timeline

https://api.github.com/repos/huggingface/datasets/issues/2770

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2770/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2770/comments

https://api.github.com/repos/huggingface/datasets/issues/2770/events

https://github.com/huggingface/datasets/pull/2770

MDExOlB1bGxSZXF1ZXN0NzA1OTAzMzIy

Add support for fast tokenizer in BertScore

closed

CONTRIBUTOR

This PR adds support for a fast tokenizer in BertScore, which has been added recently to the lib. Fixes #2765

https://api.github.com/repos/huggingface/datasets/issues/2770/timeline

https://api.github.com/repos/huggingface/datasets/issues/2769

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2769/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2769/comments

https://api.github.com/repos/huggingface/datasets/issues/2769/events

https://github.com/huggingface/datasets/pull/2769

MDExOlB1bGxSZXF1ZXN0NzA1ODk5MTYy

Allow PyArrow from source

closed

MEMBER

When installing pyarrow from source the version is: ```python >>> import pyarrow; pyarrow.__version__ '2.1.0.dev612' ``` -> however this breaks the install check at init of `datasets`. This PR makes sure that everything coming after the last `'.'` is removed.

https://api.github.com/repos/huggingface/datasets/issues/2769/timeline

https://api.github.com/repos/huggingface/datasets/issues/2768

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2768/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2768/comments

https://api.github.com/repos/huggingface/datasets/issues/2768/events

https://github.com/huggingface/datasets/issues/2768

MDU6SXNzdWU5NjMyMjkxNzM=

`ArrowInvalid: Added column's length must match table's length.` after using `select`

closed

CONTRIBUTOR

## Describe the bug I would like to add a column to a downsampled dataset. However I get an error message saying the length don't match with the length of the unsampled dataset indicated. I suspect that the dataset size is not updated when calling `select`. ## Steps to reproduce the bug ```python from datasets import load_dataset ds = load_dataset("tweets_hate_speech_detection")['train'].select(range(128)) ds = ds.add_column('ones', [1]*128) ``` ## Expected results I would expect a new column named `ones` filled with `1`. When I check the length of `ds` it says `128`. Interestingly, it works when calling `ds = ds.map(lambda x: x)` before adding the column. ## Actual results Specify the actual results or traceback. ```python --------------------------------------------------------------------------- ArrowInvalid Traceback (most recent call last) /var/folders/l4/2905jygx4tx5jv8_kn03vxsw0000gn/T/ipykernel_6301/868709636.py in <module> 1 from datasets import load_dataset 2 ds = load_dataset("tweets_hate_speech_detection")['train'].select(range(128)) ----> 3 ds = ds.add_column('ones', [0]*128) ~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs) 183 } 184 # apply actual function --> 185 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs) 186 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out] 187 # re-apply format to the output ~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/fingerprint.py in wrapper(*args, **kwargs) 395 # Call actual function 396 --> 397 out = func(self, *args, **kwargs) 398 399 # Update fingerprint of in-place transforms + update in-place history of transforms ~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/arrow_dataset.py in add_column(self, name, column, new_fingerprint) 2965 column_table = InMemoryTable.from_pydict({name: column}) 2966 # Concatenate tables horizontally -> 2967 table = ConcatenationTable.from_tables([self._data, column_table], axis=1) 2968 # Update features 2969 info = self.info.copy() ~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in from_tables(cls, tables, axis) 715 table_blocks = to_blocks(table) 716 blocks = _extend_blocks(blocks, table_blocks, axis=axis) --> 717 return cls.from_blocks(blocks) 718 719 @property ~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in from_blocks(cls, blocks) 663 return cls(table, blocks) 664 else: --> 665 table = cls._concat_blocks_horizontally_and_vertically(blocks) 666 return cls(table, blocks) 667 ~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in _concat_blocks_horizontally_and_vertically(cls, blocks) 623 if not tables: 624 continue --> 625 pa_table_horizontally_concatenated = cls._concat_blocks(tables, axis=1) 626 pa_tables_to_concat_vertically.append(pa_table_horizontally_concatenated) 627 return cls._concat_blocks(pa_tables_to_concat_vertically, axis=0) ~/git/semantic-clustering/env/lib/python3.8/site-packages/datasets/table.py in _concat_blocks(blocks, axis) 612 else: 613 for name, col in zip(table.column_names, table.columns): --> 614 pa_table = pa_table.append_column(name, col) 615 return pa_table 616 else: ~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.append_column() ~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.add_column() ~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status() ~/git/semantic-clustering/env/lib/python3.8/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status() ArrowInvalid: Added column's length must match table's length. Expected length 31962 but got length 128 ``` ## Environment info - `datasets` version: 1.11.0 - Platform: macOS-10.16-x86_64-i386-64bit - Python version: 3.8.5 - PyArrow version: 5.0.0

https://api.github.com/repos/huggingface/datasets/issues/2768/timeline

https://api.github.com/repos/huggingface/datasets/issues/2767

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2767/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2767/comments

https://api.github.com/repos/huggingface/datasets/issues/2767/events

https://github.com/huggingface/datasets/issues/2767

MDU6SXNzdWU5NjMwMDIxMjA=

equal operation to perform unbatch for huggingface datasets

open

NONE

Hi I need to use "unbatch" operation in tensorflow on a huggingface dataset, I could not find this operation, could you kindly direct me how I can do it, here is the problem I am trying to solve: I am considering "record" dataset in SuperGlue and I need to replicate each entery of the dataset for each answer, to make it similar to what T5 originally did: https://github.com/google-research/text-to-text-transfer-transformer/blob/3c58859b8fe72c2dbca6a43bc775aa510ba7e706/t5/data/preprocessors.py#L925 Here please find an example: For example, a typical example from ReCoRD might look like { 'passsage': 'This is the passage.', 'query': 'A @placeholder is a bird.', 'entities': ['penguin', 'potato', 'pigeon'], 'answers': ['penguin', 'pigeon'], } and I need a prosessor which would turn this example into the following two examples: { 'inputs': 'record query: A @placeholder is a bird. entities: penguin, ' 'potato, pigeon passage: This is the passage.', 'targets': 'penguin', } and { 'inputs': 'record query: A @placeholder is a bird. entities: penguin, ' 'potato, pigeon passage: This is the passage.', 'targets': 'pigeon', } For doing this, one need unbatch, as each entry can map to multiple samples depending on the number of answers, I am not sure how to perform this operation with huggingface datasets library and greatly appreciate your help @lhoestq Thank you very much.

https://api.github.com/repos/huggingface/datasets/issues/2767/timeline

https://api.github.com/repos/huggingface/datasets/issues/2766

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2766/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2766/comments

https://api.github.com/repos/huggingface/datasets/issues/2766/events

https://github.com/huggingface/datasets/pull/2766

MDExOlB1bGxSZXF1ZXN0NzA1NzAyNjM5

fix typo (ShuffingConfig -> ShufflingConfig)

closed

CONTRIBUTOR

pretty straightforward, it should be Shuffling instead of Shuffing

https://api.github.com/repos/huggingface/datasets/issues/2766/timeline

https://api.github.com/repos/huggingface/datasets/issues/2765

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2765/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2765/comments

https://api.github.com/repos/huggingface/datasets/issues/2765/events

https://github.com/huggingface/datasets/issues/2765

MDU6SXNzdWU5NjI4NjEzOTU=

BERTScore Error

closed

NONE

## Describe the bug A clear and concise description of what the bug is. ## Steps to reproduce the bug ```python predictions = ["hello there", "general kenobi"] references = ["hello there", "general kenobi"] bert = load_metric('bertscore') bert.compute(predictions=predictions, references=references,lang='en') ``` # Bug `TypeError: get_hash() missing 1 required positional argument: 'use_fast_tokenizer'` ## Environment info <!-- You can run the command `datasets-cli env` and copy-and-paste its output below. --> - `datasets` version: - Platform: Colab - Python version: - PyArrow version:

https://api.github.com/repos/huggingface/datasets/issues/2765/timeline

https://api.github.com/repos/huggingface/datasets/issues/2764

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2764/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2764/comments

https://api.github.com/repos/huggingface/datasets/issues/2764/events

https://github.com/huggingface/datasets/pull/2764

MDExOlB1bGxSZXF1ZXN0NzA1MzI3MDQ5

Add DER metric for SUPERB speaker diarization task

open

MEMBER

https://api.github.com/repos/huggingface/datasets/issues/2764/timeline

https://api.github.com/repos/huggingface/datasets/issues/2763

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2763/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2763/comments

https://api.github.com/repos/huggingface/datasets/issues/2763/events

https://github.com/huggingface/datasets/issues/2763

MDU6SXNzdWU5NjE4OTU1MjM=

English wikipedia datasets is not clean

open

CONTRIBUTOR

## Describe the bug Wikipedia english dumps contain many wikipedia paragraphs like "References", "Category:" and "See Also" that should not be used for training. ## Steps to reproduce the bug ```python # Sample code to reproduce the bug from datasets import load_dataset w = load_dataset('wikipedia', '20200501.en') print(w['train'][0]['text']) ``` > 'Yangliuqing () is a market town in Xiqing District, in the western suburbs of Tianjin, People\'s Republic of China. Despite its relatively small size, it has been named since 2006 in the "famous historical and cultural market towns in China".\n\nIt is best known in China for creating nianhua or Yangliuqing nianhua. For more than 400 years, Yangliuqing has in effect specialised in the creation of these woodcuts for the New Year. wood block prints using vivid colourschemes to portray traditional scenes of children\'s games often interwoven with auspiciouse objects.\n\n, it had 27 residential communities () and 25 villages under its administration.\n\nShi Family Grand Courtyard\n\nShi Family Grand Courtyard (Tiānjīn Shí Jiā Dà Yuàn, 天津石家大院) is situated in Yangliuqing Town of Xiqing District, which is the former residence of wealthy merchant Shi Yuanshi - the 4th son of Shi Wancheng, one of the eight great masters in Tianjin. First built in 1875, it covers over 6,000 square meters, including large and small yards and over 200 folk houses, a theater and over 275 rooms that served as apartments and places of business and worship for this powerful family. Shifu Garden, which finished its expansion in October 2003, covers 1,200 square meters, incorporates the elegance of imperial garden and delicacy of south garden. Now the courtyard of Shi family covers about 10,000 square meters, which is called the first mansion in North China. Now it serves as the folk custom museum in Yangliuqing, which has a large collection of folk custom museum in Yanliuqing, which has a large collection of folk art pieces like Yanliuqing New Year pictures, brick sculpture.\n\nShi\'s ancestor came from Dong\'e County in Shandong Province, engaged in water transport of grain. As the wealth gradually accumulated, the Shi Family moved to Yangliuqing and bought large tracts of land and set up their residence. Shi Yuanshi came from the fourth generation of the family, who was a successful businessman and a good household manager, and the residence was thus enlarged for several times until it acquired the present scale. It is believed to be the first mansion in the west of Tianjin.\n\nThe residence is symmetric based on the axis formed by a passageway in the middle, on which there are four archways. On the east side of the courtyard, there are traditional single-story houses with rows of rooms around the four sides, which was once the living area for the Shi Family. The rooms on north side were the accountants\' office. On the west are the major constructions including the family hall for worshipping Buddha, theater and the south reception room. On both sides of the residence are side yard rooms for maids and servants.\n\nToday, the Shi mansion, located in the township of Yangliuqing to the west of central Tianjin, stands as a surprisingly well-preserved monument to China\'s pre-revolution mercantile spirit. It also serves as an on-location shoot for many of China\'s popular historical dramas. Many of the rooms feature period furniture, paintings and calligraphy, and the extensive Shifu Garden.\n\nPart of the complex has been turned into the Yangliuqing Museum, which includes displays focused on symbolic aspects of the courtyards\' construction, local folk art and customs, and traditional period furnishings and crafts.\n\n**See also \n\nList of township-level divisions of Tianjin\n\nReferences \n\n http://arts.cultural-china.com/en/65Arts4795.html\n\nCategory:Towns in Tianjin'** ## Expected results I expect no junk in the data. ## Actual results Specify the actual results or traceback. ## Environment info - `datasets` version: 1.10.2 - Platform: macOS-10.15.7-x86_64-i386-64bit - Python version: 3.8.5 - PyArrow version: 3.0.0

https://api.github.com/repos/huggingface/datasets/issues/2763/timeline

https://api.github.com/repos/huggingface/datasets/issues/2762

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2762/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2762/comments

https://api.github.com/repos/huggingface/datasets/issues/2762/events

https://github.com/huggingface/datasets/issues/2762

MDU6SXNzdWU5NjE2NTIwNDY=

Add RVL-CDIP dataset

open

NONE

## Adding a Dataset - **Name:** RVL-CDIP - **Description:** The RVL-CDIP (Ryerson Vision Lab Complex Document Information Processing) dataset consists of 400,000 grayscale images in 16 classes, with 25,000 images per class. There are 320,000 training images, 40,000 validation images, and 40,000 test images. The images are sized so their largest dimension does not exceed 1000 pixels. - **Paper:** https://www.cs.cmu.edu/~aharley/icdar15/ - **Data:** https://www.cs.cmu.edu/~aharley/rvl-cdip/ - **Motivation:** I'm currently adding LayoutLMv2 and LayoutXLM to HuggingFace Transformers. LayoutLM (v1) already exists in the library. This dataset has a large value for document image classification (i.e. classifying scanned documents). LayoutLM models obtain SOTA on this dataset, so would be great to directly use it in notebooks. Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).

https://api.github.com/repos/huggingface/datasets/issues/2762/timeline

https://api.github.com/repos/huggingface/datasets/issues/2761

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2761/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2761/comments

https://api.github.com/repos/huggingface/datasets/issues/2761/events

https://github.com/huggingface/datasets/issues/2761

MDU6SXNzdWU5NjE1NjgyODc=

Error loading C4 realnewslike dataset

closed

NONE

## Describe the bug Error loading C4 realnewslike dataset. Validation part mismatch ## Steps to reproduce the bug ```python raw_datasets = load_dataset('c4', 'realnewslike', cache_dir=model_args.cache_dir) ## Expected results success on data loading ## Actual results Downloading: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 15.3M/15.3M [00:00<00:00, 28.1MB/s]Traceback (most recent call last): File "run_mlm_tf.py", line 794, in <module> main() File "run_mlm_tf.py", line 425, in main raw_datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name, cache_dir=model_args.cache_dir) File "/home/dshirron/.local/lib/python3.8/site-packages/datasets/load.py", line 843, in load_dataset builder_instance.download_and_prepare( File "/home/dshirron/.local/lib/python3.8/site-packages/datasets/builder.py", line 608, in download_and_prepare self._download_and_prepare( File "/home/dshirron/.local/lib/python3.8/site-packages/datasets/builder.py", line 698, in _download_and_prepare verify_splits(self.info.splits, split_dict) File "/home/dshirron/.local/lib/python3.8/site-packages/datasets/utils/info_utils.py", line 74, in verify_splits raise NonMatchingSplitsSizesError(str(bad_splits)) datasets.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='validation', num_bytes=38165657946, num_examples=13799838, dataset_name='c4'), 'recorded': SplitInfo(name='validation', num_bytes=37875873, num_examples=13863, dataset_name='c4')}] ## Environment info - `datasets` version: 1.10.2 - Platform: Linux-5.4.0-58-generic-x86_64-with-glibc2.29 - Python version: 3.8.10 - PyArrow version: 4.0.1

https://api.github.com/repos/huggingface/datasets/issues/2761/timeline

https://api.github.com/repos/huggingface/datasets/issues/2760

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2760/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2760/comments

https://api.github.com/repos/huggingface/datasets/issues/2760/events

https://github.com/huggingface/datasets/issues/2760

MDU6SXNzdWU5NjEzNzI2Njc=

Add Nuswide dataset

open

NONE

## Adding a Dataset - **Name:** *NUSWIDE* - **Description:** *[A Real-World Web Image Dataset from National University of Singapore](https://lms.comp.nus.edu.sg/wp-content/uploads/2019/research/nuswide/NUS-WIDE.html)* - **Paper:** *[here](https://lms.comp.nus.edu.sg/wp-content/uploads/2019/research/nuswide/nuswide-civr2009.pdf)* - **Data:** *[here](https://github.com/wenting-zhao/nuswide)* - **Motivation:** *This dataset is a benchmark in the Text Retrieval task.* Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).

https://api.github.com/repos/huggingface/datasets/issues/2760/timeline

https://api.github.com/repos/huggingface/datasets/issues/2759

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2759/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2759/comments

https://api.github.com/repos/huggingface/datasets/issues/2759/events

https://github.com/huggingface/datasets/issues/2759

MDU6SXNzdWU5NjA2MzY1NzI=

the meteor metric seems not consist with the official version

open

NONE

## Describe the bug The computed meteor score seems strange because the value is very different from the scores computed by other tools. For example, I use the meteor score computed by [NLGeval](https://github.com/Maluuba/nlg-eval) as the reference (which reuses the official jar file for the computation) ## Steps to reproduce the bug ```python from datasets import load_metric from nlgeval import NLGEval, compute_individual_metrics meteor = load_metric('meteor') predictions = ["It is a guide to action which ensures that the military always obeys the commands of the party"] references = ["It is a guide to action that ensures that the military will forever heed Party commands"] results = meteor.compute(predictions=predictions, references=references) # print the actual result print(round(results["meteor"], 4)) metrics_dict = compute_individual_metrics(references, predictions[0]) # print the expected result print(round(metrics_dict["METEOR"], 4)) ``` By the way, you need to install the `nlg-eval` library first. Please check the installation guide [here](https://github.com/Maluuba/nlg-eval#setup), thanks! ## Expected results `0.4474` ## Actual results `0.7398` ## Environment info - `datasets` version: 1.10.2 - Platform: macOS-10.16-x86_64-i386-64bit - Python version: 3.8.5 - PyArrow version: 4.0.1

https://api.github.com/repos/huggingface/datasets/issues/2759/timeline

https://api.github.com/repos/huggingface/datasets/issues/2758

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2758/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2758/comments

https://api.github.com/repos/huggingface/datasets/issues/2758/events

https://github.com/huggingface/datasets/pull/2758

MDExOlB1bGxSZXF1ZXN0NzAzMjQ5Nzky

Raise ManualDownloadError when loading a dataset that requires previous manual download

closed

MEMBER

This PR implements the raising of a `ManualDownloadError` when loading a dataset that requires previous manual download, and this is missing. The `ManualDownloadError` is raised whether the dataset is loaded in normal or streaming mode. Close #2749. cc: @severo

https://api.github.com/repos/huggingface/datasets/issues/2758/timeline

https://api.github.com/repos/huggingface/datasets/issues/2757

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2757/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2757/comments

https://api.github.com/repos/huggingface/datasets/issues/2757/events

https://github.com/huggingface/datasets/issues/2757

MDU6SXNzdWU5NTk5ODQwODE=

Unexpected type after `concatenate_datasets`

closed

NONE

## Describe the bug I am trying to concatenate two `Dataset` using `concatenate_datasets` but it turns out that after concatenation the features are casted from `torch.Tensor` to `list`. It then leads to a weird tensors when trying to convert it to a `DataLoader`. However, if I use each `Dataset` separately everything behave as expected. ## Steps to reproduce the bug ```python >>> featurized_teacher Dataset({ features: ['t_labels', 't_input_ids', 't_token_type_ids', 't_attention_mask'], num_rows: 502 }) >>> for f in featurized_teacher.features: print(featurized_teacher[f].shape) torch.Size([502]) torch.Size([502, 300]) torch.Size([502, 300]) torch.Size([502, 300]) >>> featurized_student Dataset({ features: ['s_features', 's_labels'], num_rows: 502 }) >>> for f in featurized_student.features: print(featurized_student[f].shape) torch.Size([502, 64]) torch.Size([502]) ``` The shapes seem alright to me. Then the results after concatenation are as follow: ```python >>> concat_dataset = datasets.concatenate_datasets([featurized_student, featurized_teacher], axis=1) >>> type(concat_dataset["t_labels"]) <class 'list'> ``` One would expect to obtain the same type as the one before concatenation. Am I doing something wrong here? Any idea on how to fix this unexpected behavior? ## Environment info - `datasets` version: 1.9.0 - Platform: macOS-10.14.6-x86_64-i386-64bit - Python version: 3.9.5 - PyArrow version: 3.0.0

https://api.github.com/repos/huggingface/datasets/issues/2757/timeline

https://api.github.com/repos/huggingface/datasets/issues/2756

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2756/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2756/comments

https://api.github.com/repos/huggingface/datasets/issues/2756/events

https://github.com/huggingface/datasets/pull/2756

MDExOlB1bGxSZXF1ZXN0NzAyMzk4Mjk1

Fix metadata JSON for ubuntu_dialogs_corpus dataset

closed

MEMBER

Related to #2743.

https://api.github.com/repos/huggingface/datasets/issues/2756/timeline

https://api.github.com/repos/huggingface/datasets/issues/2755

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2755/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2755/comments

https://api.github.com/repos/huggingface/datasets/issues/2755/events

https://github.com/huggingface/datasets/pull/2755

MDExOlB1bGxSZXF1ZXN0NzAyMjgwMjI4

Fix metadata JSON for turkish_movie_sentiment dataset

closed

MEMBER

Related to #2743.

https://api.github.com/repos/huggingface/datasets/issues/2755/timeline

https://api.github.com/repos/huggingface/datasets/issues/2754

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2754/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2754/comments

https://api.github.com/repos/huggingface/datasets/issues/2754/events

https://github.com/huggingface/datasets/pull/2754

MDExOlB1bGxSZXF1ZXN0NzAyMjcxMjM4

Generate metadata JSON for telugu_books dataset

closed

MEMBER

Related to #2743.

https://api.github.com/repos/huggingface/datasets/issues/2754/timeline

https://api.github.com/repos/huggingface/datasets/issues/2753

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2753/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2753/comments

https://api.github.com/repos/huggingface/datasets/issues/2753/events

https://github.com/huggingface/datasets/pull/2753

MDExOlB1bGxSZXF1ZXN0NzAyMjEyMjMz

Generate metadata JSON for reclor dataset

closed

MEMBER

Related to #2743.

https://api.github.com/repos/huggingface/datasets/issues/2753/timeline

https://api.github.com/repos/huggingface/datasets/issues/2752

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2752/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2752/comments

https://api.github.com/repos/huggingface/datasets/issues/2752/events

https://github.com/huggingface/datasets/pull/2752

MDExOlB1bGxSZXF1ZXN0NzAyMjAxMjAy

Generate metadata JSON for lm1b dataset

closed

MEMBER

Related to #2743.

https://api.github.com/repos/huggingface/datasets/issues/2752/timeline

https://api.github.com/repos/huggingface/datasets/issues/2751

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2751/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2751/comments

https://api.github.com/repos/huggingface/datasets/issues/2751/events

https://github.com/huggingface/datasets/pull/2751

MDExOlB1bGxSZXF1ZXN0NzAyMTk5MjA5

Update metadata for wikihow dataset

closed

MEMBER

Update metadata for wikihow dataset: - Remove leading new line character in description and citation - Update metadata JSON - Remove no longer necessary `urls_checksums/checksums.txt` file Related to #2748.

https://api.github.com/repos/huggingface/datasets/issues/2751/timeline

https://api.github.com/repos/huggingface/datasets/issues/2750

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2750/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2750/comments

https://api.github.com/repos/huggingface/datasets/issues/2750/events

https://github.com/huggingface/datasets/issues/2750

MDU6SXNzdWU5NTg5ODQ3MzA=

Second concatenation of datasets produces errors

closed

NONE

Hi, I am need to concatenate my dataset with others several times, and after I concatenate it for the second time, the features of features (e.g. tags names) are collapsed. This hinders, for instance, the usage of tokenize function with `data.map`. ``` from datasets import load_dataset, concatenate_datasets data = load_dataset('trec')['train'] concatenated = concatenate_datasets([data, data]) concatenated_2 = concatenate_datasets([concatenated, concatenated]) print('True features of features:', concatenated.features) print('\nProduced features of features:', concatenated_2.features) ``` outputs ``` True features of features: {'label-coarse': ClassLabel(num_classes=6, names=['DESC', 'ENTY', 'ABBR', 'HUM', 'NUM', 'LOC'], names_file=None, id=None), 'label-fine': ClassLabel(num_classes=47, names=['manner', 'cremat', 'animal', 'exp', 'ind', 'gr', 'title', 'def', 'date', 'reason', 'event', 'state', 'desc', 'count', 'other', 'letter', 'religion', 'food', 'country', 'color', 'termeq', 'city', 'body', 'dismed', 'mount', 'money', 'product', 'period', 'substance', 'sport', 'plant', 'techmeth', 'volsize', 'instru', 'abb', 'speed', 'word', 'lang', 'perc', 'code', 'dist', 'temp', 'symbol', 'ord', 'veh', 'weight', 'currency'], names_file=None, id=None), 'text': Value(dtype='string', id=None)} Produced features of features: {'label-coarse': Value(dtype='int64', id=None), 'label-fine': Value(dtype='int64', id=None), 'text': Value(dtype='string', id=None)} ``` I am using `datasets` v.1.11.0

https://api.github.com/repos/huggingface/datasets/issues/2750/timeline

https://api.github.com/repos/huggingface/datasets/issues/2749

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2749/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2749/comments

https://api.github.com/repos/huggingface/datasets/issues/2749/events

https://github.com/huggingface/datasets/issues/2749

MDU6SXNzdWU5NTg5Njg3NDg=

Raise a proper exception when trying to stream a dataset that requires to manually download files

closed

CONTRIBUTOR

## Describe the bug At least for 'reclor', 'telugu_books', 'turkish_movie_sentiment', 'ubuntu_dialogs_corpus', 'wikihow', trying to `load_dataset` in streaming mode raises a `TypeError` without any detail about why it fails. ## Steps to reproduce the bug ```python from datasets import load_dataset dataset = load_dataset("reclor", streaming=True) ``` ## Expected results Ideally: raise a specific exception, something like `ManualDownloadError`. Or at least give the reason in the message, as when we load in normal mode: ```python from datasets import load_dataset dataset = load_dataset("reclor") ``` ``` AssertionError: The dataset reclor with config default requires manual data. Please follow the manual download instructions: to use ReClor you need to download it manually. Please go to its homepage (http://whyu.me/reclor/) fill the google form and you will receive a download link and a password to extract it.Please extract all files in one folder and use the path folder in datasets.load_dataset('reclor', data_dir='path/to/folder/folder_name') . Manual data can be loaded with `datasets.load_dataset(reclor, data_dir='<path/to/manual/data>') ``` ## Actual results ``` TypeError: expected str, bytes or os.PathLike object, not NoneType ``` ## Environment info - `datasets` version: 1.11.0 - Platform: macOS-11.5-x86_64-i386-64bit - Python version: 3.8.11 - PyArrow version: 4.0.1

https://api.github.com/repos/huggingface/datasets/issues/2749/timeline

https://api.github.com/repos/huggingface/datasets/issues/2748

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2748/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2748/comments

https://api.github.com/repos/huggingface/datasets/issues/2748/events

https://github.com/huggingface/datasets/pull/2748

MDExOlB1bGxSZXF1ZXN0NzAyMDg4NTk4

Generate metadata JSON for wikihow dataset

closed

MEMBER

Related to #2743.

https://api.github.com/repos/huggingface/datasets/issues/2748/timeline

https://api.github.com/repos/huggingface/datasets/issues/2747

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2747/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2747/comments

https://api.github.com/repos/huggingface/datasets/issues/2747/events

https://github.com/huggingface/datasets/pull/2747

MDExOlB1bGxSZXF1ZXN0NzAyMDcwOTgy

add multi-proc in `to_json`

closed

CONTRIBUTOR

Closes #2663. I've tried adding multiprocessing in `to_json`. Here's some benchmarking I did to compare the timings of current version (say v1) and multi-proc version (say v2). I did this with `cpu_count` 4 (2015 Macbook Air) 1. Dataset name: `ascent_kb` - 8.9M samples (all samples were used, reporting this for a single run) v1- ~225 seconds for converting whole dataset to json v2- ~200 seconds for converting whole dataset to json 2. Dataset name: `lama` - 1.3M samples (all samples were used, reporting this for 2 runs) v1- ~26 seconds for converting whole dataset to json v2- ~23.6 seconds for converting whole dataset to json I think it's safe to say that v2 is 10% faster as compared to v1. Timings may improve further with better configuration. The only bottleneck I feel is writing to file from the output list. If we can improve that aspect then timings may improve further. Let me know if any changes/improvements can be done in this @stas00, @lhoestq, @albertvillanova. @lhoestq even suggested to extend this work with other export methods as well like `csv` or `parquet`.

https://api.github.com/repos/huggingface/datasets/issues/2747/timeline

https://api.github.com/repos/huggingface/datasets/issues/2746

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2746/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2746/comments

https://api.github.com/repos/huggingface/datasets/issues/2746/events

https://github.com/huggingface/datasets/issues/2746

MDU6SXNzdWU5NTg1NTE2MTk=

Cannot load `few-nerd` dataset

closed

NONE

## Describe the bug Cannot load `few-nerd` dataset. ## Steps to reproduce the bug ```python from datasets import load_dataset load_dataset('few-nerd', 'supervised') ``` ## Actual results Executing above code will give the following error: ``` Using the latest cached version of the module from /Users/Mehrad/.cache/huggingface/modules/datasets_modules/datasets/few-nerd/62464ace912a40a0f33a11a8310f9041c9dc3590ff2b3c77c14d83ca53cfec53 (last modified on Wed Jun 2 11:34:25 2021) since it couldn't be found locally at /Users/Mehrad/Documents/GitHub/genienlp/few-nerd/few-nerd.py, or remotely (FileNotFoundError). Downloading and preparing dataset few_nerd/supervised (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /Users/Mehrad/.cache/huggingface/datasets/few_nerd/supervised/0.0.0/62464ace912a40a0f33a11a8310f9041c9dc3590ff2b3c77c14d83ca53cfec53... Traceback (most recent call last): File "/Users/Mehrad/opt/anaconda3/lib/python3.7/site-packages/datasets/builder.py", line 693, in _download_and_prepare self._prepare_split(split_generator, **prepare_split_kwargs) File "/Users/Mehrad/opt/anaconda3/lib/python3.7/site-packages/datasets/builder.py", line 1107, in _prepare_split disable=bool(logging.get_verbosity() == logging.NOTSET), File "/Users/Mehrad/opt/anaconda3/lib/python3.7/site-packages/tqdm/std.py", line 1133, in __iter__ for obj in iterable: File "/Users/Mehrad/.cache/huggingface/modules/datasets_modules/datasets/few-nerd/62464ace912a40a0f33a11a8310f9041c9dc3590ff2b3c77c14d83ca53cfec53/few-nerd.py", line 196, in _generate_examples with open(filepath, encoding="utf-8") as f: FileNotFoundError: [Errno 2] No such file or directory: '/Users/Mehrad/.cache/huggingface/datasets/downloads/supervised/train.json' ``` The bug is probably in identifying and downloading the dataset. If I download the json splits directly from [link](https://github.com/nbroad1881/few-nerd/tree/main/uncompressed) and put them under the downloads directory, they will be processed into arrow format correctly. ## Environment info <!-- You can run the command `datasets-cli env` and copy-and-paste its output below. --> - `datasets` version: 1.11.0 - Python version: 3.8 - PyArrow version: 1.0.1

https://api.github.com/repos/huggingface/datasets/issues/2746/timeline

https://api.github.com/repos/huggingface/datasets/issues/2745

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2745/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2745/comments

https://api.github.com/repos/huggingface/datasets/issues/2745/events

https://github.com/huggingface/datasets/pull/2745

MDExOlB1bGxSZXF1ZXN0NzAxNTc0Mjcz

added semeval18_emotion_classification dataset

closed

CONTRIBUTOR

I added the data set of SemEval 2018 Task 1 (Subtask 5) for emotion detection in three languages. ``` datasets-cli test datasets/semeval18_emotion_classification/ --save_infos --all_configs RUN_SLOW=1 pytest tests/test_dataset_common.py::LocalDatasetTest::test_load_real_dataset_semeval18_emotion_classification ``` Both commands ran successfully. I couldn't create the dummy data (the files are tsvs but have .txt ending, maybe that's the problem?) and therefore the test on the dummy data fails, maybe someone can help here. I also formatted the code: ``` black --line-length 119 --target-version py36 datasets/semeval18_emotion_classification/ isort datasets/semeval18_emotion_classification/ flake8 datasets/semeval18_emotion_classification/ ``` That's the publication for reference: Mohammad, S., Bravo-Marquez, F., Salameh, M., & Kiritchenko, S. (2018). SemEval-2018 task 1: Affect in tweets. Proceedings of the 12th International Workshop on Semantic Evaluation, 1–17. https://doi.org/10.18653/v1/S18-1001

https://api.github.com/repos/huggingface/datasets/issues/2745/timeline

https://api.github.com/repos/huggingface/datasets/issues/2744

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2744/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2744/comments

https://api.github.com/repos/huggingface/datasets/issues/2744/events

https://github.com/huggingface/datasets/pull/2744

MDExOlB1bGxSZXF1ZXN0NzAxNDY4NDcz

Fix key by recreating metadata JSON for journalists_questions dataset

closed

MEMBER

Close #2743.

https://api.github.com/repos/huggingface/datasets/issues/2744/timeline

https://api.github.com/repos/huggingface/datasets/issues/2743

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2743/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2743/comments

https://api.github.com/repos/huggingface/datasets/issues/2743/events

https://github.com/huggingface/datasets/issues/2743

MDU6SXNzdWU5NTgxMTkyNTE=

Dataset JSON is incorrect

closed

CONTRIBUTOR

## Describe the bug The JSON file generated for https://github.com/huggingface/datasets/blob/573f3d35081cee239d1b962878206e9abe6cde91/datasets/journalists_questions/journalists_questions.py is https://github.com/huggingface/datasets/blob/573f3d35081cee239d1b962878206e9abe6cde91/datasets/journalists_questions/dataset_infos.json. The only config should be `plain_text`, but the first key in the JSON is `journalists_questions` (the dataset id) instead. ```json { "journalists_questions": { "description": "The journalists_questions corpus (version 1.0) is a collection of 10K human-written Arabic\ntweets manually labeled for question identification over Arabic tweets posted by journalists.\n", ... ``` ## Steps to reproduce the bug Look at the files. ## Expected results The first key should be `plain_text`: ```json { "plain_text": { "description": "The journalists_questions corpus (version 1.0) is a collection of 10K human-written Arabic\ntweets manually labeled for question identification over Arabic tweets posted by journalists.\n", ... ``` ## Actual results ```json { "journalists_questions": { "description": "The journalists_questions corpus (version 1.0) is a collection of 10K human-written Arabic\ntweets manually labeled for question identification over Arabic tweets posted by journalists.\n", ... ```

https://api.github.com/repos/huggingface/datasets/issues/2743/timeline

https://api.github.com/repos/huggingface/datasets/issues/2742

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2742/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2742/comments

https://api.github.com/repos/huggingface/datasets/issues/2742/events

https://github.com/huggingface/datasets/issues/2742

MDU6SXNzdWU5NTgxMTQwNjQ=

Improve detection of streamable file types

closed

CONTRIBUTOR

**Is your feature request related to a problem? Please describe.** ```python from datasets import load_dataset_builder from datasets.utils.streaming_download_manager import StreamingDownloadManager builder = load_dataset_builder("journalists_questions", name="plain_text") builder._split_generators(StreamingDownloadManager(base_path=builder.base_path)) ``` raises ``` NotImplementedError: Extraction protocol for file at https://drive.google.com/uc?export=download&id=1CBrh-9OrSpKmPQBxTK_ji6mq6WTN_U9U is not implemented yet ``` But the file at https://drive.google.com/uc?export=download&id=1CBrh-9OrSpKmPQBxTK_ji6mq6WTN_U9U is a text file and it can be streamed: ```bash curl --header "Range: bytes=0-100" -L https://drive.google.com/uc\?export\=download\&id\=1CBrh-9OrSpKmPQBxTK_ji6mq6WTN_U9U 506938088174940160 yes 1 302221719412830209 yes 1 289761704907268096 yes 1 513820885032378369 yes % ``` Yet, it's wrongly categorized as a file type that cannot be streamed because the test is currently based on 1. the presence of a file extension at the end of the URL (here: no extension), and 2. the inclusion of this extension in a list of supported formats. **Describe the solution you'd like** In the case of an URL (instead of a local path), ask for the MIME type, and decide on that value? Note that it would not work in that case, because the value of `content_type` is `text/html; charset=UTF-8`. **Describe alternatives you've considered** Add a variable in the dataset script to set the data format by hand.

https://api.github.com/repos/huggingface/datasets/issues/2742/timeline

https://api.github.com/repos/huggingface/datasets/issues/2741

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2741/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2741/comments

https://api.github.com/repos/huggingface/datasets/issues/2741/events

https://github.com/huggingface/datasets/issues/2741

MDU6SXNzdWU5NTc5Nzk1NTk=

Add Hypersim dataset

open

MEMBER

## Adding a Dataset - **Name:** Hypersim - **Description:** photorealistic synthetic dataset for holistic indoor scene understanding - **Paper:** *link to the dataset paper if available* - **Data:** https://github.com/apple/ml-hypersim Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).

https://api.github.com/repos/huggingface/datasets/issues/2741/timeline

https://api.github.com/repos/huggingface/datasets/issues/2740

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2740/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2740/comments

https://api.github.com/repos/huggingface/datasets/issues/2740/events

https://github.com/huggingface/datasets/pull/2740

MDExOlB1bGxSZXF1ZXN0NzAxMjY0NTI3

Update release instructions

closed

MEMBER

Update release instructions.

https://api.github.com/repos/huggingface/datasets/issues/2740/timeline

https://api.github.com/repos/huggingface/datasets/issues/2739

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2739/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2739/comments

https://api.github.com/repos/huggingface/datasets/issues/2739/events

https://github.com/huggingface/datasets/pull/2739

MDExOlB1bGxSZXF1ZXN0NzAxMTI0ODQ3

Pass tokenize to sacrebleu only if explicitly passed by user

closed

MEMBER

Next `sacrebleu` release (v2.0.0) will remove `sacrebleu.DEFAULT_TOKENIZER`: https://github.com/mjpost/sacrebleu/pull/152/files#diff-2553a315bb1f7e68c9c1b00d56eaeb74f5205aeb3a189bc3e527b122c6078795L17-R15 This PR passes `tokenize` to `sacrebleu` only if explicitly passed by the user, otherwise it will not pass it (and `sacrebleu` will use its default, no matter where it is and how it is called). Close: #2737.

https://api.github.com/repos/huggingface/datasets/issues/2739/timeline

https://api.github.com/repos/huggingface/datasets/issues/2738

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2738/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2738/comments

https://api.github.com/repos/huggingface/datasets/issues/2738/events

https://github.com/huggingface/datasets/pull/2738

MDExOlB1bGxSZXF1ZXN0NzAwOTI5NzA4

Sunbird AI Ugandan low resource language dataset

open

NONE

Multi-way parallel text corpus of 5 key Ugandan languages for the task of machine translation.

https://api.github.com/repos/huggingface/datasets/issues/2738/timeline

https://api.github.com/repos/huggingface/datasets/issues/2737

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2737/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2737/comments

https://api.github.com/repos/huggingface/datasets/issues/2737/events

https://github.com/huggingface/datasets/issues/2737

MDU6SXNzdWU5NTcxMjQ4ODE=

SacreBLEU update

closed

NONE

With the latest release of [sacrebleu](https://github.com/mjpost/sacrebleu), `datasets.metrics.sacrebleu` is broken, and getting error. AttributeError: module 'sacrebleu' has no attribute 'DEFAULT_TOKENIZER' this happens since in new version of sacrebleu there is no `DEFAULT_TOKENIZER`, but sacrebleu.py tries to import it anyways. This can be fixed currently with fixing `sacrebleu==1.5.0` ## Steps to reproduce the bug ```python sacrebleu= datasets.load_metric('sacrebleu') predictions = ["It is a guide to action which ensures that the military always obeys the commands of the party"] references = ["It is a guide to action that ensures that the military will forever heed Party commands"] results = sacrebleu.compute(predictions=predictions, references=references) print(results) ``` ## Environment info <!-- You can run the command `datasets-cli env` and copy-and-paste its output below. --> - `datasets` version: 1.11.0 - Platform: Windows-10-10.0.19041-SP0 - Python version: Python 3.8.0 - PyArrow version: 5.0.0

https://api.github.com/repos/huggingface/datasets/issues/2737/timeline

https://api.github.com/repos/huggingface/datasets/issues/2736

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2736/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2736/comments

https://api.github.com/repos/huggingface/datasets/issues/2736/events

https://github.com/huggingface/datasets/issues/2736

MDU6SXNzdWU5NTY4OTUxOTk=

Add Microsoft Building Footprints dataset

open

MEMBER

## Adding a Dataset - **Name:** Microsoft Building Footprints - **Description:** With the goal to increase the coverage of building footprint data available as open data for OpenStreetMap and humanitarian efforts, we have released millions of building footprints as open data available to download free of charge. - **Paper:** *link to the dataset paper if available* - **Data:** https://www.microsoft.com/en-us/maps/building-footprints - **Motivation:** this can be a useful dataset for researchers working on climate change adaptation, urban studies, geography, etc. Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md). Reported by: @sashavor

https://api.github.com/repos/huggingface/datasets/issues/2736/timeline

https://api.github.com/repos/huggingface/datasets/issues/2735

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2735/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2735/comments

https://api.github.com/repos/huggingface/datasets/issues/2735/events

https://github.com/huggingface/datasets/issues/2735

MDU6SXNzdWU5NTY4ODkzNjU=

Add Open Buildings dataset

open

MEMBER

## Adding a Dataset - **Name:** Open Buildings - **Description:** A dataset of building footprints to support social good applications. Building footprints are useful for a range of important applications, from population estimation, urban planning and humanitarian response, to environmental and climate science. This large-scale open dataset contains the outlines of buildings derived from high-resolution satellite imagery in order to support these types of uses. The project being based in Ghana, the current focus is on the continent of Africa. See: "Mapping Africa's Buildings with Satellite Imagery" https://ai.googleblog.com/2021/07/mapping-africas-buildings-with.html - **Paper:** https://arxiv.org/abs/2107.12283 - **Data:** https://sites.research.google/open-buildings/ - **Motivation:** *what are some good reasons to have this dataset* Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md). Reported by: @osanseviero

https://api.github.com/repos/huggingface/datasets/issues/2735/timeline

https://api.github.com/repos/huggingface/datasets/issues/2734

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2734/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2734/comments

https://api.github.com/repos/huggingface/datasets/issues/2734/events

https://github.com/huggingface/datasets/pull/2734

MDExOlB1bGxSZXF1ZXN0NzAwMzc4NjI4

Update BibTeX entry

closed

MEMBER

Update BibTeX entry.

https://api.github.com/repos/huggingface/datasets/issues/2734/timeline

https://api.github.com/repos/huggingface/datasets/issues/2733

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2733/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2733/comments

https://api.github.com/repos/huggingface/datasets/issues/2733/events

https://github.com/huggingface/datasets/pull/2733

MDExOlB1bGxSZXF1ZXN0NzAwMjc1NDMy

Add missing parquet known extension

closed

MEMBER

This code was failing because the parquet extension wasn't recognized: ```python from datasets import load_dataset base_url = "https://storage.googleapis.com/huggingface-nlp/cache/datasets/wikipedia/20200501.en/1.0.0/" data_files = {"train": base_url + "wikipedia-train.parquet"} wiki = load_dataset("parquet", data_files=data_files, split="train", streaming=True) ``` It raises ```python NotImplementedError: Extraction protocol for file at https://storage.googleapis.com/huggingface-nlp/cache/datasets/wikipedia/20200501.en/1.0.0/wikipedia-train.parquet is not implemented yet ``` I added `parquet` to the list of known extensions EDIT: added pickle, conllu, xml extensions as well

https://api.github.com/repos/huggingface/datasets/issues/2733/timeline

https://api.github.com/repos/huggingface/datasets/issues/2732

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2732/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2732/comments

https://api.github.com/repos/huggingface/datasets/issues/2732/events

https://github.com/huggingface/datasets/pull/2732

MDExOlB1bGxSZXF1ZXN0NzAwMjMzMzQy

Updated TTC4900 Dataset

closed

CONTRIBUTOR

- The source address of the TTC4900 dataset of [@savasy](https://github.com/savasy) has been updated for direct download. - Updated readme.

https://api.github.com/repos/huggingface/datasets/issues/2732/timeline

https://api.github.com/repos/huggingface/datasets/issues/2731

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2731/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2731/comments

https://api.github.com/repos/huggingface/datasets/issues/2731/events

https://github.com/huggingface/datasets/pull/2731

MDExOlB1bGxSZXF1ZXN0Njk5NzQwMjg5

Adding to_tf_dataset method

closed

MEMBER

Oh my **god** do not merge this yet, it's just a draft. I've added a method (via a mixin) to the `arrow_dataset.Dataset` class that automatically converts our Dataset classes to TF Dataset classes ready for training. It hopefully has most of the features we want, including streaming from disk (no need to load the whole dataset in memory!), correct shuffling, variable-length batches to reduce compute, and correct support for unusual padding. It achieves that by calling the tokenizer `pad` method in the middle of a TF compute graph via a very hacky call to `tf.py_function`, which is heretical but seems to work. A number of issues need to be resolved before it's ready to merge, though: 1) Is a MixIn the right way to do this? Do other classes besides `arrow_dataset.Dataset` need this method too? 2) Needs an argument to support constant-length batches for TPU training - this is easy to add and I'll do it soon. 3) Needs the user to supply the list of columns to drop from the arrow `Dataset`. Is there some automatic way to get the columns we want, or see which columns were added by the tokenizer? 4) Assumes the label column is always present and always called "label" - this is probably not great, but I'm not sure what the 'correct' thing to do here is.

https://api.github.com/repos/huggingface/datasets/issues/2731/timeline

https://api.github.com/repos/huggingface/datasets/issues/2730

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2730/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2730/comments

https://api.github.com/repos/huggingface/datasets/issues/2730/events

https://github.com/huggingface/datasets/issues/2730

MDU6SXNzdWU5NTU5ODc4MzQ=

Update CommonVoice with new release

open

MEMBER

## Adding a Dataset - **Name:** CommonVoice mid-2021 release - **Description:** more data in CommonVoice: Languages that have increased the most by percentage are Thai (almost 20x growth, from 12 hours to 250 hours), Luganda (almost 9x growth, from 8 to 80), Esperanto (7x growth, from 100 to 840), and Tamil (almost 8x, from 24 to 220). - **Paper:** https://discourse.mozilla.org/t/common-voice-2021-mid-year-dataset-release/83812 - **Data:** https://commonvoice.mozilla.org/en/datasets - **Motivation:** More data and more varied. I think we just need to add configs in the existing dataset script. Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).

https://api.github.com/repos/huggingface/datasets/issues/2730/timeline

https://api.github.com/repos/huggingface/datasets/issues/2729

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2729/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2729/comments

https://api.github.com/repos/huggingface/datasets/issues/2729/events

https://github.com/huggingface/datasets/pull/2729

MDExOlB1bGxSZXF1ZXN0Njk5NTk5MjA4

Fix IndexError while loading Arabic Billion Words dataset

closed

MEMBER

Catch `IndexError` and ignore that record. Close #2727.

https://api.github.com/repos/huggingface/datasets/issues/2729/timeline

https://api.github.com/repos/huggingface/datasets/issues/2728

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2728/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2728/comments

https://api.github.com/repos/huggingface/datasets/issues/2728/events

https://github.com/huggingface/datasets/issues/2728

MDU6SXNzdWU5NTU4OTI5NzA=

Concurrent use of same dataset (already downloaded)

open

CONTRIBUTOR

## Describe the bug When launching several jobs at the same time loading the same dataset trigger some errors see (last comments). ## Steps to reproduce the bug export HF_DATASETS_CACHE=/gpfswork/rech/toto/datasets for MODEL in "bert-base-uncased" "roberta-base" "distilbert-base-cased"; do # "bert-base-uncased" "bert-large-cased" "roberta-large" "albert-base-v1" "albert-large-v1"; do for TASK_NAME in "mrpc" "rte" 'imdb' "paws" "mnli"; do export OUTPUT_DIR=${MODEL}_${TASK_NAME} sbatch --job-name=${OUTPUT_DIR} \ --gres=gpu:1 \ --no-requeue \ --cpus-per-task=10 \ --hint=nomultithread \ --time=1:00:00 \ --output=jobinfo/${OUTPUT_DIR}_%j.out \ --error=jobinfo/${OUTPUT_DIR}_%j.err \ --qos=qos_gpu-t4 \ --wrap="module purge; module load pytorch-gpu/py3/1.7.0 ; export HF_DATASETS_OFFLINE=1; export HF_DATASETS_CACHE=/gpfswork/rech/toto/datasets; python compute_measures.py --seed=$SEED --saving_path=results --batch_size=$BATCH_SIZE --task_name=$TASK_NAME --model_name=/gpfswork/rech/toto/transformers_models/$MODEL" done done ```python # Sample code to reproduce the bug dataset_train = load_dataset('imdb', split='train', download_mode="reuse_cache_if_exists") dataset_train = dataset_train.map(lambda e: tokenizer(e['text'], truncation=True, padding='max_length'), batched=True).select(list(range(args.filter))) dataset_val = load_dataset('imdb', split='train', download_mode="reuse_cache_if_exists") dataset_val = dataset_val.map(lambda e: tokenizer(e['text'], truncation=True, padding='max_length'), batched=True).select(list(range(args.filter, args.filter + 5000))) dataset_test = load_dataset('imdb', split='test', download_mode="reuse_cache_if_exists") dataset_test = dataset_test.map(lambda e: tokenizer(e['text'], truncation=True, padding='max_length'), batched=True) ``` ## Expected results I believe I am doing something wrong with the objects. ## Actual results Traceback (most recent call last): File "/gpfslocalsup/pub/anaconda-py3/2020.02/envs/pytorch-gpu-1.7.0/lib/python3.7/site-packages/datasets/builder.py", line 652, in _download_and_prepare self._prepare_split(split_generator, **prepare_split_kwargs) File "/gpfslocalsup/pub/anaconda-py3/2020.02/envs/pytorch-gpu-1.7.0/lib/python3.7/site-packages/datasets/builder.py", line 983, in _prepare_split check_duplicates=True, File "/gpfslocalsup/pub/anaconda-py3/2020.02/envs/pytorch-gpu-1.7.0/lib/python3.7/site-packages/datasets/arrow_writer.py", line 192, in __init__ self.stream = pa.OSFile(self._path, "wb") File "pyarrow/io.pxi", line 829, in pyarrow.lib.OSFile.__cinit__ File "pyarrow/io.pxi", line 844, in pyarrow.lib.OSFile._open_writable File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status File "pyarrow/error.pxi", line 97, in pyarrow.lib.check_status FileNotFoundError: [Errno 2] Failed to open local file '/gpfswork/rech/tts/unm25jp/datasets/paws/labeled_final/1.1.0/09d8fae989bb569009a8f5b879ccf2924d3e5cd55bfe2e89e6dab1c0b50ecd34.incomplete/paws-test.arrow'. Detail: [errno 2] No such file or directory During handling of the above exception, another exception occurred: Traceback (most recent call last): File "compute_measures.py", line 181, in <module> train_loader, val_loader, test_loader = get_dataloader(args) File "/gpfsdswork/projects/rech/toto/intRAOcular/dataset_utils.py", line 69, in get_dataloader dataset_train = load_dataset('paws', "labeled_final", split='train', download_mode="reuse_cache_if_exists") File "/gpfslocalsup/pub/anaconda-py3/2020.02/envs/pytorch-gpu-1.7.0/lib/python3.7/site-packages/datasets/load.py", line 748, in load_dataset use_auth_token=use_auth_token, File "/gpfslocalsup/pub/anaconda-py3/2020.02/envs/pytorch-gpu-1.7.0/lib/python3.7/site-packages/datasets/builder.py", line 575, in download_and_prepare dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs File "/gpfslocalsup/pub/anaconda-py3/2020.02/envs/pytorch-gpu-1.7.0/lib/python3.7/site-packages/datasets/builder.py", line 658, in _download_and_prepare + str(e) OSError: Cannot find data file. Original error: [Errno 2] Failed to open local file '/gpfswork/rech/toto/datasets/paws/labeled_final/1.1.0/09d8fae989bb569009a8f5b879ccf2924d3e5cd55bfe2e89e6dab1c0b50ecd34.incomplete/paws-test.arrow'. Detail: [errno 2] No such file or directory ## Environment info <!-- You can run the command `datasets-cli env` and copy-and-paste its output below. --> - `datasets` version: datasets==1.8.0 - Platform: linux (jeanzay) - Python version: pyarrow==2.0.0 - PyArrow version: 3.7.8

https://api.github.com/repos/huggingface/datasets/issues/2728/timeline

https://api.github.com/repos/huggingface/datasets/issues/2727

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2727/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2727/comments

https://api.github.com/repos/huggingface/datasets/issues/2727/events

https://github.com/huggingface/datasets/issues/2727

MDU6SXNzdWU5NTU4MTIxNDk=

Error in loading the Arabic Billion Words Corpus

closed

CONTRIBUTOR

## Describe the bug I get `IndexError: list index out of range` when trying to load the `Techreen` and `Almustaqbal` configs of the dataset. ## Steps to reproduce the bug ```python load_dataset("arabic_billion_words", "Techreen") load_dataset("arabic_billion_words", "Almustaqbal") ``` ## Expected results The datasets load succefully. ## Actual results ```python _extract_tags(self, sample, tag) 139 if len(out) > 0: 140 break --> 141 return out[0] 142 143 def _clean_text(self, text): IndexError: list index out of range ``` ## Environment info <!-- You can run the command `datasets-cli env` and copy-and-paste its output below. --> - `datasets` version: 1.10.2 - Platform: Ubuntu 18.04.5 LTS - Python version: 3.7.11 - PyArrow version: 3.0.0

https://api.github.com/repos/huggingface/datasets/issues/2727/timeline

https://api.github.com/repos/huggingface/datasets/issues/2726

https://api.github.com/repos/huggingface/datasets

https://api.github.com/repos/huggingface/datasets/issues/2726/labels{/name}

https://api.github.com/repos/huggingface/datasets/issues/2726/comments

https://api.github.com/repos/huggingface/datasets/issues/2726/events

https://github.com/huggingface/datasets/pull/2726

MDExOlB1bGxSZXF1ZXN0Njk5Mzg5MDk1

Typo fix `tokenize_exemple`