id
int64 599M
3.26B
| number
int64 1
7.7k
| title
stringlengths 1
290
| body
stringlengths 0
228k
⌀ | state
stringclasses 2
values | html_url
stringlengths 46
51
| created_at
timestamp[s]date 2020-04-14 10:18:02
2025-07-23 08:04:53
| updated_at
timestamp[s]date 2020-04-27 16:04:17
2025-07-23 18:53:44
| closed_at
timestamp[s]date 2020-04-14 12:01:40
2025-07-23 16:44:42
⌀ | user
dict | labels
listlengths 0
4
| is_pull_request
bool 2
classes | comments
listlengths 0
0
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
1,438,544,617
| 5,211
|
Update Overview.ipynb google colab
|
- removed metrics stuff
- added image example
- added audio example (with ffmpeg instructions)
- updated the "add a new dataset" section
|
closed
|
https://github.com/huggingface/datasets/pull/5211
| 2022-11-07T15:23:52
| 2022-11-29T15:59:48
| 2022-11-29T15:54:17
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
1,438,492,507
| 5,210
|
Tweak readme
|
Tweaked some paragraphs mentioning the modalities we support + added a paragraph on security
|
closed
|
https://github.com/huggingface/datasets/pull/5210
| 2022-11-07T14:51:23
| 2022-11-24T11:35:07
| 2022-11-24T11:26:16
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
1,438,367,678
| 5,209
|
Implement ability to define splits in metadata section of dataset card
|
### Feature request
If you go here: https://huggingface.co/datasets/inria-soda/tabular-benchmark/tree/main you will see bunch of folders that has various CSV files. I’d like dataset viewer to show these files instead of only one dataset like it currently does. (and also people to be able to load them as splits instead of loading through `data_files`)
e.g GLUE has various splits on viewer but it’s too overkill to ask people to implement loading script, so it would be better to let them define these in the README file instead.
Also pinging @polinaeterna @lhoestq @adrinjalali
|
closed
|
https://github.com/huggingface/datasets/issues/5209
| 2022-11-07T13:27:16
| 2023-07-21T14:36:02
| 2023-07-21T14:36:01
|
{
"login": "merveenoyan",
"id": 53175384,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,438,035,707
| 5,208
|
Refactor CI hub fixtures to use monkeypatch instead of patch
|
Minor refactoring of CI to use `pytest` `monkeypatch` instead of `unittest` `patch`.
|
closed
|
https://github.com/huggingface/datasets/pull/5208
| 2022-11-07T09:25:05
| 2022-11-08T06:51:20
| 2022-11-08T06:49:17
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,437,858,506
| 5,207
|
Connection error of the HuggingFace's dataset Hub due to SSLError with proxy
|
### Describe the bug
It's weird. I could not normally connect the dataset Hub of HuggingFace due to a SSLError in my office.
Even when I try to connect using my company's proxy address (e.g., http_proxy and https_proxy),
I'm getting the SSLError issue. What should I do to download the datanet stored in HuggingFace normally?
I welcome any comments. I think those comments will be helpful to me.
* Dataset address - https://huggingface.co/datasets/moyix/debian_csrc/viewer/moyix--debian_csrc
* Log message
```
............ OMISSION ..............
Traceback (most recent call last):
File "/data/home/geunsik-lim/qtlab/./transformers/examples/pytorch/language-modeling/run_clm.py", line 587, in <module>
main()
File "/data/home/geunsik-lim/qtlab/./transformers/examples/pytorch/language-modeling/run_clm.py", line 278, in main
raw_datasets = load_dataset(
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1719, in load_dataset
builder_instance = load_dataset_builder(
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1497, in load_dataset_builder
dataset_module = dataset_module_factory(
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1222, in dataset_module_factory
raise e1 from None
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1179, in dataset_module_factory
raise ConnectionError(f"Couldn't reach '{path}' on the Hub ({type(e).__name__})")
ConnectionError: Couldn't reach 'moyix/debian_csrc' on the Hub (SSLError)
[2022-11-07 15:23:38,476] [INFO] [launch.py:318:sigkill_handler] Killing subprocess 6760
[2022-11-07 15:23:38,476] [ERROR] [launch.py:324:sigkill_handler] ['/home/geunsik-lim/anaconda3/envs/deepspeed/bin/python', '-u', './transformers/examples/pytorch/language-modeling/run_clm.py', '--local_rank=0', '--model_name_or_path=Salesforce/codegen-350M-multi', '--per_device_train_batch_size=1', '--learning_rate', '2e-5', '--num_train_epochs', '1', '--output_dir=./codegen-350M-finetuned', '--overwrite_output_dir', '--dataset_name', 'moyix/debian_csrc', '--cache_dir', '/data/home/geunsik-lim/.cache', '--tokenizer_name', 'Salesforce/codegen-350M-multi', '--block_size', '2048', '--gradient_accumulation_steps', '32', '--do_train', '--fp16', '--deepspeed', 'ds_config_zero2.json'] exits with return code = 1
real 0m7.742s
user 0m4.930s
```
### Steps to reproduce the bug
Steps to reproduce this behavior.
```
(deepspeed) geunsik-lim@ai02:~/qtlab$ ./test_debian_csrc_dataset.py
Traceback (most recent call last):
File "/data/home/geunsik-lim/qtlab/./test_debian_csrc_dataset.py", line 6, in <module>
dataset = load_dataset("moyix/debian_csrc")
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1719, in load_dataset
builder_instance = load_dataset_builder(
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1497, in load_dataset_builder
dataset_module = dataset_module_factory(
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1222, in dataset_module_factory
raise e1 from None
File "/home/geunsik-lim/anaconda3/envs/deepspeed/lib/python3.10/site-packages/datasets/load.py", line 1179, in dataset_module_factory
raise ConnectionError(f"Couldn't reach '{path}' on the Hub ({type(e).__name__})")
ConnectionError: Couldn't reach 'moyix/debian_csrc' on the Hub (SSLError)
(deepspeed) geunsik-lim@ai02:~/qtlab$
(deepspeed) geunsik-lim@ai02:~/qtlab$
(deepspeed) geunsik-lim@ai02:~/qtlab$
(deepspeed) geunsik-lim@ai02:~/qtlab$ cat ./test_debian_csrc_dataset.py
#!/usr/bin/env python
from datasets import load_dataset
dataset = load_dataset("moyix/debian_csrc")
```
1. Adde proxy address of a company in /etc/profile
2. Download dataset with load_dataset() function of datasets package that is provided by HuggingFace.
3. In this case, the address would be "moyix--debian_csrc".
4. I get the "`ConnectionError: Couldn't reach 'moyix/debian_csrc' on the Hub (SSLError`)" error message.
### Expected behavior
* error message:
ConnectionError: Couldn't reach 'moyix/debian_csrc' on the Hub (SSLError)
### Environment info
* software version information:
```
(deepspeed) geunsik-lim@ai02:~$
(deepspeed) geunsik-lim@ai02:~$ conda list -f pytorch
# packages in environment at /home/geunsik-lim/anaconda3/envs/deepspeed:
#
# Name Version Build Channel
pytorch 1.13.0 py3.10_cuda11.7_cudnn8.5.0_0 pytorch
(deepspeed) geunsik-lim@ai02:~$ conda list -f python
# packages in environment at /home/geunsik-lim/anaconda3/envs/deepspeed:
#
# Name Version Build Channel
python 3.10.6 haa1d7c7_1
(deepspeed) geunsik-lim@ai02:~$ conda list -f datasets
# packages in environment at /home/geunsik-lim/anaconda3/envs/deepspeed:
#
# Name Version Build Channel
datasets 2.6.1 py_0 huggingface
(deepspeed) geunsik-lim@ai02:~$ uname -a
Linux ai02 5.4.0-131-generic #147-Ubuntu SMP Fri Oct 14 17:07:22 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux
(deepspeed) geunsik-lim@ai02:~$ cat /etc/lsb-release
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=20.04
DISTRIB_CODENAME=focal
DISTRIB_DESCRIPTION="Ubuntu 20.04.5 LTS"
```
|
open
|
https://github.com/huggingface/datasets/issues/5207
| 2022-11-07T06:56:23
| 2025-03-08T09:04:10
| null |
{
"login": "leemgs",
"id": 82404,
"type": "User"
}
|
[] | false
|
[] |
1,437,223,894
| 5,206
|
Use logging instead of printing to console
|
### Describe the bug
Some logs ([here](https://github.com/huggingface/datasets/blob/4a6e1fe2735505efc7e3a3dbd3e1835da0702575/src/datasets/builder.py#L778), [here](https://github.com/huggingface/datasets/blob/4a6e1fe2735505efc7e3a3dbd3e1835da0702575/src/datasets/builder.py#L786), and [here](https://github.com/huggingface/datasets/blob/4a6e1fe2735505efc7e3a3dbd3e1835da0702575/src/datasets/builder.py#L830)) generated by the `DatasetBuilder` are printed to the console instead of passed to `datasets` logger.
### Steps to reproduce the bug
```python
>> import datasets
>> datasets.load_dataset("some-dataset")
Downloading and preparing dataset csv/data to <path>...
Downloading data files: 100%|██████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 7729.06it/s]
Extracting data files: 100%|████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 527.23it/s]
Dataset csv downloaded and prepared to <path>. Subsequent calls will reuse this data.
```
### Expected behavior
The logs should not be printed to the console directly but passed to the logger so that the user can redirect them wherever he wants.
### Environment info
- `datasets` version: 2.6.1
- Platform: macOS-13.0-x86_64-i386-64bit
- Python version: 3.9.15
- PyArrow version: 10.0.0
- Pandas version: 1.5.1
|
closed
|
https://github.com/huggingface/datasets/issues/5206
| 2022-11-05T23:48:02
| 2022-11-06T00:06:00
| 2022-11-06T00:05:59
|
{
"login": "bilelomrani1",
"id": 16692099,
"type": "User"
}
|
[] | false
|
[] |
1,437,221,987
| 5,205
|
Add missing `DownloadConfig.use_auth_token` value
|
This PR solves https://github.com/huggingface/datasets/issues/5204
Now the `token` is propagated so that `DownloadConfig.use_auth_token` value is set before trying to download private files from existing datasets in the Hub.
|
closed
|
https://github.com/huggingface/datasets/pull/5205
| 2022-11-05T23:36:36
| 2022-11-08T08:13:00
| 2022-11-07T16:20:24
|
{
"login": "alvarobartt",
"id": 36760800,
"type": "User"
}
|
[] | true
|
[] |
1,437,221,259
| 5,204
|
`push_to_hub` not propagating `token` through `DownloadConfig`
|
### Describe the bug
When trying to upload a new 🤗 Dataset to the Hub via Python, and providing the `token` as a parameter to the `Dataset.push_to_hub` function, it just works for the first time, assuming that the dataset didn't exist before.
But when trying to run `Dataset.push_to_hub` again over the same dataset, instead of updating it, it throws a `ConnectionError` when trying to retrieve the `README.md` that may contain some metadata about the dataset, so as to also update it, but since the `token` is not propagated, the `DownloadConfig` provided to the `datasets.utils.file_utils.get_from_cache` function doesn't contain the `use_auth_token` value set to `token`, it's just using the default one which is None/False.
So on, when uploading a dataset via Python with `push_to_hub` with the `token` as a parameter with the HuggingFace API Token as value, it can just be uploaded when the dataset is new, otherwise it fails with to `ConnectionError` due to the `token` not being propagated as `use_auth_token`.
### Steps to reproduce the bug
Let's create a new dataset in our HF account via Python as:
```python
from datasets import Dataset
data = {"a": [1, 2, 3], "b": [4, 5, 6]}
ds = Dataset.from_dict(data)
ds.push_to_hub(repo_id=<HF_USERNAME>/<HF_DATASET>, private=private, token=<HF_TOKEN_HERE>)
```
When we create the `Dataset` for the first time it works and there are no issues, but when trying to actually upload a new version of the same dataset (same name under the same username), we encounter the following issue:
```python
from datasets import Dataset
data = {"a": [1, 2, 3], "b": [4, 5, 6]}
ds = Dataset.from_dict(data)
ds.push_to_hub(repo_id=<HF_USERNAME>/<HF_DATASET>, private=private, token=<HF_TOKEN_HERE>)
>>> ConnectionError: Couldn't reach https://huggingface.co/datasets/alvarobartt/demo/resolve/main/README.md (ConnectionError('Unauthorized for URL https://huggingface.co/datasets/<HF_USERNAME>/<HF_DATASET>/resolve/main/README.md. Please use the parameter `use_auth_token=True` after logging in with `huggingface-cli login`'))
```
### Expected behavior
Ideally, the `token` parameter provided to `push_to_hub` should be propagated and used to download the `README.md` when trying to update a `Dataset`, instead of throwing that exception, so that the authentication can be done directly through code without running `huggingface-cli login`as mentioned at https://huggingface.co/docs/datasets/upload_dataset#upload-with-python.
### Environment info
- `datasets` version: 2.6.1
- Platform: macOS-13.0-arm64-arm-64bit
- Python version: 3.10.8
- PyArrow version: 10.0.0
- Pandas version: 1.5.1
|
closed
|
https://github.com/huggingface/datasets/issues/5204
| 2022-11-05T23:32:20
| 2022-11-08T10:12:09
| 2022-11-08T10:12:08
|
{
"login": "alvarobartt",
"id": 36760800,
"type": "User"
}
|
[] | false
|
[] |
1,436,710,518
| 5,203
|
Update canonical links to Hub links
|
This PR updates some of the canonical dataset links to their corresponding links on the Hub; closes #5200.
|
closed
|
https://github.com/huggingface/datasets/pull/5203
| 2022-11-04T22:50:50
| 2022-11-07T18:43:05
| 2022-11-07T18:40:19
|
{
"login": "stevhliu",
"id": 59462357,
"type": "User"
}
|
[] | true
|
[] |
1,435,886,090
| 5,202
|
CI fails after bulk edit of canonical datasets
|
```
______ test_get_dataset_config_info[paws-labeled_final-expected_splits2] _______
[gw0] linux -- Python 3.7.15 /opt/hostedtoolcache/Python/3.7.15/x64/bin/python
path = 'paws', config_name = 'labeled_final'
expected_splits = ['train', 'test', 'validation']
@pytest.mark.parametrize(
"path, config_name, expected_splits",
[
("squad", "plain_text", ["train", "validation"]),
("dalle-mini/wit", "dalle-mini--wit", ["train"]),
("paws", "labeled_final", ["train", "test", "validation"]),
],
)
def test_get_dataset_config_info(path, config_name, expected_splits):
info = get_dataset_config_info(path, config_name=config_name)
assert info.config_name == config_name
> assert list(info.splits.keys()) == expected_splits
E AssertionError: assert ['test', 'tra... 'validation'] == ['train', 'te... 'validation']
E At index 0 diff: 'test' != 'train'
E Full diff:
E - ['train', 'test', 'validation']
E + ['test', 'train', 'validation']
tests/test_inspect.py:45: AssertionError
_ test_get_dataset_info[paws-expected_configs2-expected_splits_in_first_config2] _
[gw0] linux -- Python 3.7.15 /opt/hostedtoolcache/Python/3.7.15/x64/bin/python
path = 'paws'
expected_configs = ['labeled_final', 'labeled_swap', 'unlabeled_final']
expected_splits_in_first_config = ['train', 'test', 'validation']
@pytest.mark.parametrize(
"path, expected_configs, expected_splits_in_first_config",
[
("squad", ["plain_text"], ["train", "validation"]),
("dalle-mini/wit", ["dalle-mini--wit"], ["train"]),
("paws", ["labeled_final", "labeled_swap", "unlabeled_final"], ["train", "test", "validation"]),
],
)
def test_get_dataset_info(path, expected_configs, expected_splits_in_first_config):
infos = get_dataset_infos(path)
assert list(infos.keys()) == expected_configs
expected_config = expected_configs[0]
assert expected_config in infos
info = infos[expected_config]
assert info.config_name == expected_config
> assert list(info.splits.keys()) == expected_splits_in_first_config
E AssertionError: assert ['test', 'tra... 'validation'] == ['train', 'te... 'validation']
E At index 0 diff: 'test' != 'train'
E Full diff:
E - ['train', 'test', 'validation']
E + ['test', 'train', 'validation']
tests/test_inspect.py:90: AssertionError
______ test_get_dataset_split_names[paws-labeled_final-expected_splits2] _______
[gw0] linux -- Python 3.7.15 /opt/hostedtoolcache/Python/3.7.15/x64/bin/python
path = 'paws', expected_config = 'labeled_final'
expected_splits = ['train', 'test', 'validation']
@pytest.mark.parametrize(
"path, expected_config, expected_splits",
[
("squad", "plain_text", ["train", "validation"]),
("dalle-mini/wit", "dalle-mini--wit", ["train"]),
("paws", "labeled_final", ["train", "test", "validation"]),
],
)
def test_get_dataset_split_names(path, expected_config, expected_splits):
infos = get_dataset_infos(path)
assert expected_config in infos
info = infos[expected_config]
assert info.config_name == expected_config
> assert list(info.splits.keys()) == expected_splits
E AssertionError: assert ['test', 'tra... 'validation'] == ['train', 'te... 'validation']
E At index 0 diff: 'test' != 'train'
E Full diff:
E - ['train', 'test', 'validation']
E + ['test', 'train', 'validation']
```
|
closed
|
https://github.com/huggingface/datasets/issues/5202
| 2022-11-04T10:51:20
| 2023-02-16T09:11:10
| 2023-02-16T09:11:10
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,435,881,554
| 5,201
|
Do not sort splits in dataset info
|
I suggest not to sort splits by their names in dataset_info in README so that they are displayed in the order specified in the loading script. Otherwise `test` split is displayed first, see this repo: https://huggingface.co/datasets/paws
What do you think?
But I added sorting in tests to fix CI (for the same dataset).
|
closed
|
https://github.com/huggingface/datasets/pull/5201
| 2022-11-04T10:47:21
| 2022-11-04T14:47:37
| 2022-11-04T14:45:09
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[] | true
|
[] |
1,435,831,559
| 5,200
|
Some links to canonical datasets in the docs are outdated
|
As we don't have canonical datasets in the github repo anymore, some old links to them doesn't work. I don't know how many of them are there, I found link to SuperGlue here: https://huggingface.co/docs/datasets/dataset_script#multiple-configurations, probably there are more of them. These links should be replaced by links to the corresponding datasets on the Hub.
|
closed
|
https://github.com/huggingface/datasets/issues/5200
| 2022-11-04T10:06:21
| 2022-11-07T18:40:20
| 2022-11-07T18:40:20
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[
{
"name": "documentation",
"color": "0075ca"
}
] | false
|
[] |
1,434,818,836
| 5,199
|
Deprecate dummy data generation command
|
Deprecate the `dummy_data` CLI command.
|
closed
|
https://github.com/huggingface/datasets/pull/5199
| 2022-11-03T15:05:54
| 2022-11-04T14:01:50
| 2022-11-04T13:59:47
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
1,434,699,165
| 5,198
|
Add note about the name of a dataset script
|
Add note that a dataset script should has the same name as a repo/dir, a bit related to this issue https://github.com/huggingface/datasets/issues/5193
also fixed two minor issues in audio docs (broken links)
|
closed
|
https://github.com/huggingface/datasets/pull/5198
| 2022-11-03T13:51:32
| 2022-11-04T12:47:59
| 2022-11-04T12:46:01
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[] | true
|
[] |
1,434,676,150
| 5,197
|
[zstd] Use max window log size
|
ZstdDecompressor has a parameter `max_window_size` to limit max memory usage when decompressing zstd files. The default `max_window_size` is not enough when files are compressed by `zstd --ultra` flags.
Change `max_window_size` to the zstd's max window size. NOTE, the `zstd.WINDOWLOG_MAX` is the log_2 value of the max window size.
|
open
|
https://github.com/huggingface/datasets/pull/5197
| 2022-11-03T13:35:58
| 2022-11-03T13:45:19
| null |
{
"login": "reyoung",
"id": 728699,
"type": "User"
}
|
[] | true
|
[] |
1,434,401,646
| 5,196
|
Use hfh hf_hub_url function
|
Small refactoring to use `hf_hub_url` function from `huggingface_hub`.
This PR also creates the `hub` module that will contain all `huggingface_hub` functionalities relevant to `datasets`.
This is a necessary stage before implementing the use of the `hfh` caching system (which uses its `hf_hub_url` under the hood).
EDIT:
~~Finally, we use our `config.HUB_DATASETS_URL` when using `hfh.hf_hub_url`~~
There is a breaking change: the `hfh` `hf_hub_url` function uses
- `hfh` `HUGGINGFACE_CO_URL_TEMPLATE` URL template, different from the `datasets` `config.HUB_DATASETS_URL`
- also, `hfh` `DEFAULT_REVISION`, instead of `datasets` `config.HUB_DEFAULT_VERSION`
|
closed
|
https://github.com/huggingface/datasets/pull/5196
| 2022-11-03T10:08:09
| 2022-12-06T11:38:17
| 2022-11-09T07:15:12
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,434,290,689
| 5,195
|
[wip testing docs]
| null |
closed
|
https://github.com/huggingface/datasets/pull/5195
| 2022-11-03T08:37:34
| 2023-04-04T15:10:37
| 2023-04-04T15:10:33
|
{
"login": "mishig25",
"id": 11827707,
"type": "User"
}
|
[] | true
|
[] |
1,434,206,951
| 5,194
|
Fix docs about dataset_info in YAML
|
This PR fixes some misalignment in the docs after we transferred the dataset_info from `dataset_infos.json` to YAML in the dataset card:
- #4926
Related to:
- #5193
|
closed
|
https://github.com/huggingface/datasets/pull/5194
| 2022-11-03T07:10:23
| 2022-11-03T13:31:27
| 2022-11-03T13:29:21
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,433,883,780
| 5,193
|
"One or several metadata. were found, but not in the same directory or in a parent directory"
|
### Describe the bug
When loading my own dataset, on loading it I get an error.
Here is my dataset link: https://huggingface.co/datasets/corentinm7/MyoQuant-SDH-Data
And the error after loading with:
```python
from datasets import load_dataset
load_dataset("corentinm7/MyoQuant-SDH-Data")
```
```python
Downloading readme: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3.34k/3.34k [00:00<00:00, 4.45MB/s]
Using custom data configuration SDH_16k-53e7301a92ab0025
Downloading and preparing dataset None/SDH_16k to /home/corentin/.cache/huggingface/datasets/corentinm7___imagefolder/SDH_16k-53e7301a92ab0025/0.0.0/37fbb85cc714a338bea574ac6c7d0b5be5aff46c1862c1989b20e0771199e93f...
Downloading data: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3.28M/3.28M [00:00<00:00, 4.31MB/s]
Downloading data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:01<00:00, 1.75s/it]
Downloading data: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.13G/1.13G [00:15<00:00, 74.3MB/s]
Downloading data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:16<00:00, 16.09s/it]
Extracting data files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:13<00:00, 13.16s/it]
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/corentin/code-project/hugging_face_play/.venv/lib/python3.10/site-packages/datasets/load.py", line 1742, in load_dataset
builder_instance.download_and_prepare(
File "/home/corentin/code-project/hugging_face_play/.venv/lib/python3.10/site-packages/datasets/builder.py", line 814, in download_and_prepare
self._download_and_prepare(
File "/home/corentin/code-project/hugging_face_play/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1423, in _download_and_prepare
super()._download_and_prepare(
File "/home/corentin/code-project/hugging_face_play/.venv/lib/python3.10/site-packages/datasets/builder.py", line 905, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/corentin/code-project/hugging_face_play/.venv/lib/python3.10/site-packages/datasets/builder.py", line 1374, in _prepare_split
for key, record in logging.tqdm(
File "/home/corentin/code-project/hugging_face_play/.venv/lib/python3.10/site-packages/tqdm/std.py", line 1195, in __iter__
for obj in iterable:
File "/home/corentin/code-project/hugging_face_play/.venv/lib/python3.10/site-packages/datasets/packaged_modules/folder_based_builder/folder_based_builder.py", line 394, in _generate_examples
raise ValueError(
ValueError: One or several metadata. were found, but not in the same directory or in a parent directory of /home/corentin/.cache/huggingface/datasets/downloads/extracted/60c4aa8d4da3065bb3d310de4373dffd73bd4dc331aedcb4ee867febe4fdb7cd/validation/sick/2_CG_SDH_TAM_Bin1cKO_ko_pla_4_1640.tif.
```
However the test command is working fine. ```datasets-cli test hugging_face_play/ds_test/SDH_16k.py --save_info --all_configs --force_redownload```
```
Using custom data configuration SDH_16k
Testing builder 'SDH_16k' (1/1)
Downloading and preparing dataset sdh_16k/SDH_16k to /home/corentin/.cache/huggingface/datasets/sdh_16k/SDH_16k/1.0.0/21b584239a638aeeda33cba1ac2ca4869d48e4b4f20fb22274d5a5ddc487659d...
Downloading data: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.13G/1.13G [00:14<00:00, 76.5MB/s]
Downloading data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:15<00:00, 15.66s/it]
Downloading data: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3.28M/3.28M [00:02<00:00, 1.44MB/s]
Downloading data files: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:03<00:00, 3.21s/it]
Downloading data files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 11586.48it/s]
Extracting data files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:13<00:00, 13.42s/it]
Dataset sdh_16k downloaded and prepared to /home/corentin/.cache/huggingface/datasets/sdh_16k/SDH_16k/1.0.0/21b584239a638aeeda33cba1ac2ca4869d48e4b4f20fb22274d5a5ddc487659d. Subsequent calls will reuse this data.
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 605.27it/s]
Dataset card saved at hugging_face_play/ds_test/README.md
Test successful.
```
### Steps to reproduce the bug
Simply run on python
```python
from datasets import load_dataset
load_dataset("corentinm7/MyoQuant-SDH-Data")
```
### Expected behavior
As the test command worked, this error should not appear
### Environment info
- `datasets` version: 2.6.1
- Platform: Linux-5.10.16.3-microsoft-standard-WSL2-x86_64-with-glibc2.31
- Python version: 3.10.6
- PyArrow version: 10.0.0
- Pandas version: 1.5.1
|
closed
|
https://github.com/huggingface/datasets/issues/5193
| 2022-11-02T22:46:25
| 2022-11-03T13:39:16
| 2022-11-03T13:35:44
|
{
"login": "lambda-science",
"id": 20109584,
"type": "User"
}
|
[] | false
|
[] |
1,433,199,790
| 5,192
|
Drop labels in Image and Audio folders if files are on different levels in directory or if there is only one label
|
Will close https://github.com/huggingface/datasets/issues/5153
Drop labels by default (`drop_labels=None`) when:
* there are files on different levels of directory hierarchy by checking their path depth
* all files are in the same directory (=only one label was inferred)
First one fixes cases like this:
```
repo
image3.jpg
image4.jpg
data
image1.jpg
image2.jpg
```
Second one fixes cases like this:
```
repo
image1.jpg
image2.jpg
image3.jpg
```
This is mostly to fix the viewer for people who just drop images in the Hub interface into the root dir.
I added tests for both of the cases on local and remote files. **I also changed data files for old test on drop_labels** (`test_generate_examples_drop_labels`). The files I provide to `test_generate_examples_drop_labels` now has "canonical" classification structure (two dirs) in order not to change the logic of the test (=not to check these two cases addressed in the PR).
|
closed
|
https://github.com/huggingface/datasets/pull/5192
| 2022-11-02T14:01:41
| 2022-11-15T16:32:53
| 2022-11-15T16:31:07
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | true
|
[] |
1,433,191,658
| 5,191
|
Make torch.Tensor and spacy models cacheable
|
Override `Pickler.save` to implement deterministic reduction (lazily registered; inspired by https://github.com/uqfoundation/dill/blob/master/dill/_dill.py#L343) functions for `torch.Tensor` and spaCy models.
Fix https://github.com/huggingface/datasets/issues/5170, fix https://github.com/huggingface/datasets/issues/3178
|
closed
|
https://github.com/huggingface/datasets/pull/5191
| 2022-11-02T13:56:18
| 2022-11-02T17:20:48
| 2022-11-02T17:18:42
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
1,433,014,626
| 5,190
|
`path` is `None` when downloading a custom audio dataset from the Hub
|
### Describe the bug
I've created an [audio dataset](https://huggingface.co/datasets/lewtun/audio-test-push) using the `audiofolder` feature desribed in the [docs](https://huggingface.co/docs/datasets/audio_dataset#audiofolder) and then pushed it to the Hub.
Locally, I can see the `audio.path` feature is of the expected form `path/to/data_dir`, but when I download the dataset from the Hub, I see `audio.path` is `None`
Here's an example:
```python
from datasets import load_dataset
ds = load_dataset("lewtun/audio-test-push")
ds["train"][0]
# {
# "audio": {
# "path": None, <-- Is this expected?
# "array": array(
# [
# 3.97140226e-07,
# 7.30310290e-07,
# 7.56406735e-07,
# ...,
# -1.19636677e-01,
# -1.16811886e-01,
# -1.12441722e-01,
# ]
# ),
# "sampling_rate": 44100,
# },
# "song_id": 0,
# "genre_id": 0,
# "genre": "Electronic",
# }
```
Is this expected behaviour? If yes, feel free to close this issue as it's not a true bug then :)
### Steps to reproduce the bug
1. Create an audio dataset with the `audiofolder` feature
2. Push the dataset to the Hub with `push_to_hub()`
3. Download the Hub dataset and inspect the `audio.path` feature
### Expected behavior
`audio.path` points to the file associated with the audio data
### Environment info
- `datasets` version: 2.6.2.dev0
- Platform: macOS-10.16-x86_64-i386-64bit
- Python version: 3.8.13
- PyArrow version: 9.0.0
- Pandas version: 1.5.1
|
closed
|
https://github.com/huggingface/datasets/issues/5190
| 2022-11-02T11:51:25
| 2022-11-02T12:55:02
| 2022-11-02T12:55:02
|
{
"login": "lewtun",
"id": 26859204,
"type": "User"
}
|
[] | false
|
[] |
1,432,769,143
| 5,189
|
Reduce friction in tabular dataset workflow by eliminating having splits when dataset is loaded
|
### Feature request
Sorry for cryptic name but I'd like to explain using code itself. When I want to load a specific dataset from a repository (for instance, this: https://huggingface.co/datasets/inria-soda/tabular-benchmark)
```python
from datasets import load_dataset
dataset = load_dataset("inria-soda/tabular-benchmark", data_files=["reg_cat/house_sales.csv"], streaming=True)
print(next(iter(dataset["train"])))
```
`datasets` library is essentially designed for people who'd like to use benchmark datasets on various modalities to fine-tune their models, and these benchmark datasets usually have pre-defined train and test splits. However, for tabular workflows, having train and test splits usually ends up model overfitting to validation split so usually the users would like to do validation techniques like `StratifiedKFoldCrossValidation` or when they tune for hyperparameters they do `GridSearchCrossValidation` so often the behavior is to create their own splits. Even [in this paper](https://hal.archives-ouvertes.fr/hal-03723551) a benchmark is introduced but the split is done by authors.
It's a bit confusing for average tabular user to try and load a dataset and see `"train"` so it would be nice if we would not load dataset into a split called `train `by default.
```diff
from datasets import load_dataset
dataset = load_dataset("inria-soda/tabular-benchmark", data_files=["reg_cat/house_sales.csv"], streaming=True)
-print(next(iter(dataset["train"])))
+print(next(iter(dataset)))
```
### Motivation
I explained it above 😅
### Your contribution
I think this is quite a big change that seems small (e.g. how to determine datasets that will not be load to train split?), it's best if we discuss first!
|
open
|
https://github.com/huggingface/datasets/issues/5189
| 2022-11-02T09:15:02
| 2022-12-06T12:13:17
| null |
{
"login": "merveenoyan",
"id": 53175384,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,432,477,139
| 5,188
|
add: segmentation guide.
|
Closes #5181
I have opened a PR on Hub (https://huggingface.co/datasets/huggingface/documentation-images/discussions/5) to include the images in our central Hub repository. Once the PR is merged I will edit the image links.
I have also prepared a [Colab Notebook](https://colab.research.google.com/drive/1BMDCfOTBnyshoME5RSxn5iQy-TWeFbOA?usp=sharing) in case anyone wants to play.
- [x] Replace the image links
|
closed
|
https://github.com/huggingface/datasets/pull/5188
| 2022-11-02T04:34:36
| 2022-11-04T18:25:57
| 2022-11-04T18:23:34
|
{
"login": "sayakpaul",
"id": 22957388,
"type": "User"
}
|
[
{
"name": "documentation",
"color": "0075ca"
}
] | true
|
[] |
1,432,375,375
| 5,187
|
chore: add notebook links to img cls and obj det.
|
Closes https://github.com/huggingface/datasets/issues/5182
|
closed
|
https://github.com/huggingface/datasets/pull/5187
| 2022-11-02T02:30:09
| 2022-11-03T01:52:24
| 2022-11-03T01:49:56
|
{
"login": "sayakpaul",
"id": 22957388,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | true
|
[] |
1,432,045,011
| 5,186
|
Incorrect error message when Dataset.from_sql fails and sqlalchemy not installed
|
### Describe the bug
When calling `Dataset.from_sql` (in my case, with sqlite3), it fails with a message ```ValueError: Please pass `features` or at least one example when writing data``` when I don't have `sqlalchemy` installed.
### Steps to reproduce the bug
Make a new sqlite db with `sqlite3` and `pandas` from a remote [URL](https://raw.githubusercontent.com/nytimes/covid-19-data/master/us-states.csv).
```python
import sqlite3
import pandas as pd
from datasets import Dataset
conn = sqlite3.connect('us_covid_data.db')
df = pd.read_csv('https://raw.githubusercontent.com/nytimes/covid-19-data/master/us-states.csv')
df.to_sql('states', conn, if_exists='replace')
```
Then if you try to query this DB like this:
```python
ds = Dataset.from_sql('''SELECT * from states WHERE state=="New York";''', "sqlite:///us_covid_data.db")
```
You run into the error I described above:
```ValueError: Please pass `features` or at least one example when writing data```
However, if you try to pass features, as the error suggests, then you get an error that tells you the underlying problem...
```python
from datasets import Dataset, Features, Value
features = Features({
'date': Value('date32'),
'label': Value('string'),
'fips': Value('int32'),
'cases': Value('int32'),
'deaths': Value('int32')
})
ds = Dataset.from_sql(
'''SELECT * from states WHERE state=="New York";''',
"sqlite:///us_covid_data.db",
features=features
)
```
Which results in the actual underlying error: `ImportError: Using URI string without sqlalchemy installed.`
### Expected behavior
Instead of `ValueError` about needing to pass features, we should provide the actual underlying error about not having SQLAlchemy installed when it isn't found in the environment.
### Environment info
- `datasets` version: 2.6.1
- Platform: macOS-10.16-x86_64-i386-64bit
- Python version: 3.8.10
- PyArrow version: 10.0.0
- Pandas version: 1.2.5
|
closed
|
https://github.com/huggingface/datasets/issues/5186
| 2022-11-01T20:25:51
| 2022-11-15T18:24:39
| 2022-11-15T18:24:39
|
{
"login": "nateraw",
"id": 32437151,
"type": "User"
}
|
[] | false
|
[] |
1,432,021,611
| 5,185
|
Allow passing a subset of output features to Dataset.map
|
### Feature request
Currently, map does one of two things to the features (if I'm not mistaken):
* when you do not pass features, types are assumed to be equal to the input if they can be cast, and inferred otherwise
* when you pass a full specification of features, output features are set to this
However, sometimes you want to just pass some of the output types, particularly when the first of these modes makes an incorrect type. This currently crashes.
### Motivation
To give a little background: this problem appears in converting labels to ids, where the labels happen to be floats rather than strings
Consider the following use of map to convert from float to int
```python
data = Dataset.from_dict({'y':[1.0,2.0,3.0]})
mapped = data.map(lambda r: {'y': int(r['y'])})
mapped['y'] # is floats, not ints
```
The result is a float again, since after the mapping operation it forces the old datatypes back on the data.
Passing `features=Features({"y": Value(dtype="int64")})` to map works in principle, but then extending it a little to e.g.
```python
def format_data(r):
return {**tokenizer(r["text"]), "y": int(r["y"])}
data = Dataset.from_dict({"y": [1.0, 2.0, 3.0], "text": ["one", "two", "three"]})
mapped = data.map(
format_data,
features=Features({'y': Value(dtype="int64")}),
remove_columns=["text"],
)
```
Results in a crash in dataset internals, as it expects either all or no output features to be specified.
Of course one can pass a full feature specification, but this becomes tokenizer specific and very awkward.
### Your contribution
I've looked at `write_batch` and particularly `col_type = features[col] if features else None`, but checking for `col in features` here makes it fail elsewhere, but the structure makes it hard to understand how and why. I do not think I would have the time myself to get to the bottom of this anytime soon.
|
open
|
https://github.com/huggingface/datasets/issues/5185
| 2022-11-01T20:07:20
| 2022-11-01T20:07:34
| null |
{
"login": "sanderland",
"id": 48946947,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,431,418,066
| 5,183
|
Loading an external dataset in a format similar to conll2003
|
I'm trying to load a custom dataset in a Dataset object, it's similar to conll2003 but with 2 columns only (word entity), I used the following script:
features = datasets.Features(
{"tokens": datasets.Sequence(datasets.Value("string")),
"ner_tags": datasets.Sequence(
datasets.features.ClassLabel(
names=["B-PER", .... etc.]))}
)
from datasets import Dataset
INPUT_COLUMNS = "tokens ner_tags".split(" ")
def read_conll(file):
#all_labels = []
example = {col: [] for col in INPUT_COLUMNS}
idx = 0
with open(file) as f:
for line in f:
if line:
if line.startswith("-DOCSTART-") and example["tokens"] != []:
print(idx, example)
yield idx, example
idx += 1
example = {col: [] for col in INPUT_COLUMNS}
elif line == "\n" or (line.startswith("-DOCSTART-") and example["tokens"] == []):
continue
else:
row_cols = line.split(" ")
for i, col in enumerate(example):
example[col] = row_cols[i].rstrip()
dset = Dataset.from_generator(read_conll, gen_kwargs={"file": "/content/new_train.txt"}, features = features)
The following error happened:
[/usr/local/lib/python3.7/dist-packages/datasets/utils/py_utils.py](https://localhost:8080/#) in <genexpr>(.0)
285 for key in unique_values(itertools.chain(*dicts)): # set merge all keys
286 # Will raise KeyError if the dict don't have the same keys
--> 287 yield key, tuple(d[key] for d in dicts)
288
TypeError: tuple indices must be integers or slices, not str
What does this mean and what should I modify?
|
closed
|
https://github.com/huggingface/datasets/issues/5183
| 2022-11-01T13:18:29
| 2022-11-02T11:57:50
| 2022-11-02T11:57:50
|
{
"login": "Taghreed7878",
"id": 112555442,
"type": "User"
}
|
[] | false
|
[] |
1,431,029,547
| 5,182
|
Add notebook / other resource links to the task-specific data loading guides
|
Does it make sense to include links to notebooks / scripts that show how to use a dataset for training / fine-tuning a model?
For example, here in [https://huggingface.co/docs/datasets/image_classification] we could include a mention of https://github.com/huggingface/notebooks/blob/main/examples/image_classification.ipynb.
Applies to https://huggingface.co/docs/datasets/object_detection as well.
Cc: @osanseviero @nateraw
|
closed
|
https://github.com/huggingface/datasets/issues/5182
| 2022-11-01T07:57:26
| 2022-11-03T01:49:57
| 2022-11-03T01:49:57
|
{
"login": "sayakpaul",
"id": 22957388,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,431,027,102
| 5,181
|
Add a guide for semantic segmentation
|
Currently, we have these guides for object detection and image classification:
* https://huggingface.co/docs/datasets/object_detection
* https://huggingface.co/docs/datasets/image_classification
I am proposing adding a similar guide for semantic segmentation.
I am happy to contribute a PR for it.
Cc: @osanseviero @nateraw
|
closed
|
https://github.com/huggingface/datasets/issues/5181
| 2022-11-01T07:54:50
| 2022-11-04T18:23:36
| 2022-11-04T18:23:36
|
{
"login": "sayakpaul",
"id": 22957388,
"type": "User"
}
|
[
{
"name": "documentation",
"color": "0075ca"
}
] | false
|
[] |
1,431,012,438
| 5,180
|
An example or recommendations for creating large image datasets?
|
I know that Apache Beam and `datasets` have [some connector utilities](https://huggingface.co/docs/datasets/beam). But it's a little unclear what we mean by "But if you want to run your own Beam pipeline with Dataflow, here is how:". What does that pipeline do?
As a user, I was wondering if we have this support for creating large image datasets. If so, we should mention that [here](https://huggingface.co/docs/datasets/image_dataset).
Cc @lhoestq
|
open
|
https://github.com/huggingface/datasets/issues/5180
| 2022-11-01T07:38:38
| 2022-11-02T10:17:11
| null |
{
"login": "sayakpaul",
"id": 22957388,
"type": "User"
}
|
[] | false
|
[] |
1,430,826,100
| 5,179
|
`map()` fails midway due to format incompatibility
|
### Describe the bug
I am using the `emotion` dataset from Hub for sequence classification. After training the model, I am using it to generate predictions for all the entries present in the `validation` split of the dataset.
```py
def get_test_accuracy(model):
def fn(batch):
inputs = {k:v.to(device) for k,v in batch.items()
if k in tokenizer.model_input_names}
with torch.no_grad():
output = model(**inputs)
pred_label = torch.argmax(output.logits, axis=-1)
return {"predicted_label": pred_label.cpu().numpy()}
return fn
```
This is how the `get_test_accuracy()` is being used:
```py
emotions = load_dataset("emotion")
def tokenize(batch):
return tokenizer(batch["text"], padding=True, truncation=True)
emotions_encoded = emotions.map(tokenize, batched=True)
emotions_encoded.set_format("torch",
columns=["input_ids", "attention_mask", "label"])
new_dataset = emotions_encoded["validation"].map(
accuracy_fn, batched=True, batch_size=128
)
```
Complete code is available in the Colab Notebook provided below.
The `map()` process fails midway giving:
```shell
AttributeError Traceback (most recent call last)
<ipython-input-8-ad24ac288eb4> in <module>
2
3 new_dataset = emotions_encoded["validation"].map(
----> 4 accuracy_fn, batched=True, batch_size=128
5 )
7 frames
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in map(self, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint, desc)
2588 new_fingerprint=new_fingerprint,
2589 disable_tqdm=disable_tqdm,
-> 2590 desc=desc,
2591 )
2592 else:
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
582 self: "Dataset" = kwargs.pop("self")
583 # apply actual function
--> 584 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
585 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
586 for dataset in datasets:
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
549 }
550 # apply actual function
--> 551 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
552 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
553 # re-apply format to the output
/usr/local/lib/python3.7/dist-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)
478 # Call actual function
479
--> 480 out = func(self, *args, **kwargs)
481
482 # Update fingerprint of in-place transforms + update in-place history of transforms
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in _map_single(self, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, new_fingerprint, rank, offset, disable_tqdm, desc, cache_only)
2970 indices,
2971 check_same_num_examples=len(input_dataset.list_indexes()) > 0,
-> 2972 offset=offset,
2973 )
2974 except NumExamplesMismatchError:
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in apply_function_on_filtered_inputs(inputs, indices, check_same_num_examples, offset)
2850 if with_rank:
2851 additional_args += (rank,)
-> 2852 processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
2853 if update_data is None:
2854 # Check if the function returns updated examples
<ipython-input-6-4e0d280426f6> in fn(batch)
1 def get_test_accuracy(model):
2 def fn(batch):
----> 3 inputs = {k:v.to(device) for k,v in batch.items()
4 if k in tokenizer.model_input_names}
5 with torch.no_grad():
<ipython-input-6-4e0d280426f6> in <dictcomp>(.0)
2 def fn(batch):
3 inputs = {k:v.to(device) for k,v in batch.items()
----> 4 if k in tokenizer.model_input_names}
5 with torch.no_grad():
6 output = model(**inputs)
AttributeError: 'list' object has no attribute 'to'
```
As you'd notice in the notebook, the process fails _midway_ and not at the beginning.
Is this expected?
### Steps to reproduce the bug
Colab Notebook:
https://colab.research.google.com/gist/sayakpaul/d1570d537faf39040d02d77b1ed7de07/scratchpad.ipynb
### Expected behavior
The mapping process should complete as is. If you switch the `split` to `test` it works as expected.
### Environment info
Colab
|
closed
|
https://github.com/huggingface/datasets/issues/5179
| 2022-11-01T03:57:59
| 2022-11-08T11:35:26
| 2022-11-08T11:35:26
|
{
"login": "sayakpaul",
"id": 22957388,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,430,800,810
| 5,178
|
Unable to download the Chinese `wikipedia`, the dumpstatus.json not found!
|
### Describe the bug
I tried:
`data = load_dataset('wikipedia', '20220301.zh', beam_runner='DirectRunner')`
and
`data = load_dataset("wikipedia", language="zh", date="20220301", beam_runner='DirectRunner')`
but both got:
`FileNotFoundError: Couldn't find file at https://dumps.wikimedia.org/zhwiki/20220301/dumpstatus.json`
the full report is:
```
FileNotFoundError Traceback (most recent call last)
<ipython-input-13-d07c5021090c> in <module>
1 from datasets import load_dataset
2
----> 3 data = load_dataset("wikipedia", language="zh", date="20220301", beam_runner='DirectRunner')<?, ?it/s]
/opt/conda/lib/python3.8/site-packages/datasets/load.py in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, ignore_verifications, keep_in_memory, save_infos, revision, use_auth_token, task, streaming, **config_kwargs)
1740
1741 # Download and prepare data
-> 1742 builder_instance.download_and_prepare(
1743 download_config=download_config,
1744 download_mode=download_mode,
/opt/conda/lib/python3.8/site-packages/datasets/builder.py in download_and_prepare(self, output_dir, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, base_path, use_auth_token, file_format, max_shard_size, storage_options, **download_and_prepare_kwargs)
812 **download_and_prepare_kwargs,
813 }
--> 814 self._download_and_prepare(
815 dl_manager=dl_manager,
816 verify_infos=verify_infos,
/opt/conda/lib/python3.8/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_splits_kwargs)
1645 options=beam_options,
1646 )
-> 1647 super()._download_and_prepare(
1648 dl_manager, verify_infos=False, pipeline=pipeline, **prepare_splits_kwargs
1649 ) # TODO handle verify_infos in beam datasets
/opt/conda/lib/python3.8/site-packages/datasets/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
881 split_dict = SplitDict(dataset_name=self.name)
882 split_generators_kwargs = self._make_split_generators_kwargs(prepare_split_kwargs)
--> 883 split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
884
885 # Checksums verification
~/.cache/huggingface/modules/datasets_modules/datasets/wikipedia/aa542ed919df55cc5d3347f42dd4521d05ca68751f50dbc32bae2a7f1e167559/wikipedia.py in _split_generators(self, dl_manager, pipeline)
943 info_url = _base_url(lang) + _INFO_FILE
944 # Use dictionary since testing mock always returns the same result.
--> 945 downloaded_files = dl_manager.download_and_extract({"info": info_url})
946
947 xml_urls = []
/opt/conda/lib/python3.8/site-packages/datasets/download/download_manager.py in download_and_extract(self, url_or_urls)
431 extracted_path(s): `str`, extracted paths of given URL(s).
432 """
--> 433 return self.extract(self.download(url_or_urls))
434
435 def get_recorded_sizes_checksums(self):
/opt/conda/lib/python3.8/site-packages/datasets/download/download_manager.py in download(self, url_or_urls)
308
309 start_time = datetime.now()
--> 310 downloaded_path_or_paths = map_nested(
311 download_func,
312 url_or_urls,
/opt/conda/lib/python3.8/site-packages/datasets/utils/py_utils.py in map_nested(function, data_struct, dict_only, map_list, map_tuple, map_numpy, num_proc, parallel_min_length, types, disable_tqdm, desc)
427 num_proc = 1
428 if num_proc <= 1 or len(iterable) < parallel_min_length:
--> 429 mapped = [
430 _single_map_nested((function, obj, types, None, True, None))
431 for obj in logging.tqdm(iterable, disable=disable_tqdm, desc=desc)
/opt/conda/lib/python3.8/site-packages/datasets/utils/py_utils.py in <listcomp>(.0)
428 if num_proc <= 1 or len(iterable) < parallel_min_length:
429 mapped = [
--> 430 _single_map_nested((function, obj, types, None, True, None))
431 for obj in logging.tqdm(iterable, disable=disable_tqdm, desc=desc)
432 ]
/opt/conda/lib/python3.8/site-packages/datasets/utils/py_utils.py in _single_map_nested(args)
329 # Singleton first to spare some computation
330 if not isinstance(data_struct, dict) and not isinstance(data_struct, types):
--> 331 return function(data_struct)
332
333 # Reduce logging to keep things readable in multiprocessing with tqdm
/opt/conda/lib/python3.8/site-packages/datasets/download/download_manager.py in _download(self, url_or_filename, download_config)
335 # append the relative path to the base_path
336 url_or_filename = url_or_path_join(self._base_path, url_or_filename)
--> 337 return cached_path(url_or_filename, download_config=download_config)
338
339 def iter_archive(self, path_or_buf: Union[str, io.BufferedReader]):
/opt/conda/lib/python3.8/site-packages/datasets/utils/file_utils.py in cached_path(url_or_filename, download_config, **download_kwargs)
186 if is_remote_url(url_or_filename):
187 # URL, so get it from the cache (downloading if necessary)
--> 188 output_path = get_from_cache(
189 url_or_filename,
190 cache_dir=cache_dir,
/opt/conda/lib/python3.8/site-packages/datasets/utils/file_utils.py in get_from_cache(url, cache_dir, force_download, proxies, etag_timeout, resume_download, user_agent, local_files_only, use_etag, max_retries, use_auth_token, ignore_url_params, download_desc)
533 )
534 elif response is not None and response.status_code == 404:
--> 535 raise FileNotFoundError(f"Couldn't find file at {url}")
536 _raise_if_offline_mode_is_enabled(f"Tried to reach {url}")
537 if head_error is not None:
FileNotFoundError: Couldn't find file at https://dumps.wikimedia.org/zhwiki/20220301/dumpstatus.json
```
### Steps to reproduce the bug
`data = load_dataset('wikipedia', '20220301.zh', beam_runner='DirectRunner')`
### Expected behavior
download the data
### Environment info
python3.6
latest datasets/transformers version
|
closed
|
https://github.com/huggingface/datasets/issues/5178
| 2022-11-01T03:17:55
| 2022-11-02T08:27:15
| 2022-11-02T08:24:29
|
{
"login": "beyondguo",
"id": 37113676,
"type": "User"
}
|
[] | false
|
[] |
1,430,238,556
| 5,177
|
Update create image dataset docs
|
Based on @osanseviero and community feedback, it wasn't super clear how to upload a dataset to the Hub after creating something like an image captioning dataset. This PR adds a brief section on how to upload the dataset with `push_to_hub`.
|
closed
|
https://github.com/huggingface/datasets/pull/5177
| 2022-10-31T17:45:56
| 2022-11-02T17:15:22
| 2022-11-02T17:13:02
|
{
"login": "stevhliu",
"id": 59462357,
"type": "User"
}
|
[
{
"name": "documentation",
"color": "0075ca"
}
] | true
|
[] |
1,430,214,539
| 5,176
|
prepare dataset for cloud storage doesn't work
|
### Describe the bug
Following the [documentation](https://huggingface.co/docs/datasets/filesystems#load-and-save-your-datasets-using-your-cloud-storage-filesystem) and [this PR](https://github.com/huggingface/datasets/pull/4724), I was downloading and storing huggingface dataset to cloud storage.
```
from datasets import load_dataset, load_dataset_builder
dataset = load_dataset_builder("wikipedia", "20220301.en", cache_dir='LOCAL_PATH')
dataset.download_and_prepare("gs://Bucket_NAME", file_format="parquet")
```
The above code successfully downloaded dataset, however, it returns error from `download_and_prepare`.
> Traceback (most recent call last):
> File "/shared/zhuiai/research/wiki/wiki/gcsfs.py", line 12, in <module>
> dataset.download_and_prepare("gs://upgen/dataset/wiki", file_format="parquet")
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/datasets/builder.py", line 671, in download_and_prepare
> fs_token_paths = fsspec.get_fs_token_paths(output_dir, storage_options=storage_options)
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/fsspec/core.py", line 635, in get_fs_token_paths
> cls = get_filesystem_class(protocol)
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/fsspec/registry.py", line 234, in get_filesystem_class
> register_implementation(protocol, _import_class(bit["class"]))
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/fsspec/registry.py", line 257, in _import_class
> mod = importlib.import_module(mod)
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/importlib/__init__.py", line 127, in import_module
> return _bootstrap._gcd_import(name[level:], package, level)
> File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
> File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
> File "<frozen importlib._bootstrap>", line 986, in _find_and_load_unlocked
> File "<frozen importlib._bootstrap>", line 680, in _load_unlocked
> File "<frozen importlib._bootstrap_external>", line 850, in exec_module
> File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
> File "/shared/zhuiai/research/wiki/wiki/gcsfs.py", line 12, in <module>
> dataset.download_and_prepare("gs://upgen/dataset/wiki", file_format="parquet")
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/datasets/builder.py", line 671, in download_and_prepare
> fs_token_paths = fsspec.get_fs_token_paths(output_dir, storage_options=storage_options)
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/fsspec/core.py", line 635, in get_fs_token_paths
> cls = get_filesystem_class(protocol)
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/fsspec/registry.py", line 234, in get_filesystem_class
> register_implementation(protocol, _import_class(bit["class"]))
> File "/shared/zhuiai/.conda/envs/wiki/lib/python3.9/site-packages/fsspec/registry.py", line 258, in _import_class
> return getattr(mod, name)
> AttributeError: partially initialized module 'gcsfs' has no attribute 'GCSFileSystem' (most likely due to a circular import)
### Steps to reproduce the bug
1. pip install datasets==2.6.1 gcsfs==2022.8.2
2. Run the following code will reproduce the issue (change `LOCAL_PATH` and `Bucket_NAME` accordingly)
```
from datasets import load_dataset, load_dataset_builder
dataset = load_dataset_builder("wikipedia", "20220301.en", cache_dir='LOCAL_PATH')
dataset.download_and_prepare("gs://Bucket_NAME", file_format="parquet")
```
### Expected behavior
Expecting successful downloading dataset and uploading it to cloud storage.
### Environment info
- `datasets` version: 2.6.1
- Platform: Linux-5.15.0-25-generic-x86_64-with-glibc2.35
- Python version: 3.9.12
- PyArrow version: 7.0.0
- Pandas version: 1.5.1
|
closed
|
https://github.com/huggingface/datasets/issues/5176
| 2022-10-31T17:28:57
| 2023-03-28T09:11:46
| 2023-03-28T09:11:45
|
{
"login": "araonblake",
"id": 27285078,
"type": "User"
}
|
[] | false
|
[] |
1,428,696,231
| 5,175
|
Loading an external NER dataset
|
I need to use huggingface datasets to load a custom dataset similar to conll2003 but with more entities and each the files contain only two columns: word and ner tag.
I tried this code snnipet that I found here as an answer to a similar issue:
from datasets import Dataset
INPUT_COLUMNS = "ID Text NER".split()
def read_conll(file):
example = {col: [] for col in INPUT_COLUMNS}
idx = 0
with open(file) as f:
for line in f:
if line.startswith("-DOCSTART-") or line == "\n" or not line:
if example[next(iter(example))]:
yield idx, example
idx += 1
example = {col: [] for col in INPUT_COLUMNS}
else:
row_cols = line.split()
for i, col in enumerate(example):
example[col] = row_cols[i].rstrip()
train = Dataset.from_generator(read_conll, gen_kwargs={"file": "some_path"})
But the following error happened:
ValueError: Please pass `features` or at least one example when writing data
|
closed
|
https://github.com/huggingface/datasets/issues/5175
| 2022-10-30T09:31:55
| 2022-11-01T13:15:49
| 2022-11-01T13:15:49
|
{
"login": "Taghreed7878",
"id": 112555442,
"type": "User"
}
|
[] | false
|
[] |
1,427,216,416
| 5,174
|
Preserve None in list type cast in PyArrow 10
|
The `ListArray` type in PyArrow 10.0.0 supports the `mask` parameter, which allows us to preserve Nones in nested lists in `cast` instead of replacing them with empty lists.
Fix https://github.com/huggingface/datasets/issues/3676
|
closed
|
https://github.com/huggingface/datasets/pull/5174
| 2022-10-28T12:48:30
| 2022-10-28T13:15:33
| 2022-10-28T13:13:18
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
1,425,880,441
| 5,173
|
Raise ffmpeg warnings only once
|
Our warnings looks nice now.
`librosa` warning that was raised at each decoding:
```
/usr/local/lib/python3.7/dist-packages/librosa/core/audio.py:165: UserWarning: PySoundFile failed. Trying audioread instead.
warnings.warn("PySoundFile failed. Trying audioread instead.")
```
is suppressed with `filterwarnings("ignore")` in a context manager. That means the first warning is also ignored (setting `filterwarnings("once")` didn't work!), so I added info that audioread is used for decoding to our message. Hope it's enough.
Tests failed at first because they used to check if the warning was raised at (each) decoding in `librosa` case but now we throw only one warning (at first decoding). I removed this check for warnings, do you think it's fine?
|
closed
|
https://github.com/huggingface/datasets/pull/5173
| 2022-10-27T15:58:33
| 2022-10-28T16:03:05
| 2022-10-28T16:00:51
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[] | true
|
[] |
1,425,523,114
| 5,172
|
Inconsistency behavior between handling local file protocol and other FS protocols
|
### Describe the bug
These lines us used during load_from_disk:
```
if is_remote_filesystem(fs):
dest_dataset_dict_path = extract_path_from_uri(dataset_dict_path)
else:
fs = fsspec.filesystem("file")
dest_dataset_dict_path = dataset_dict_path
```
If a local FS is given, then it will the URL as the path name. If a remote Fs is given, then it will use the path of the URL. This is an inconsistent behavior when handling a file: when using remote FS, you must write a URL, but for local FS, even if you passed LocalFileSystem as `fs` you still can't use a `file://` URL. It will be recognized as a directory named `file:`.
### Steps to reproduce the bug
```
import fsspec.core
url = "hdfs:///somewhere/MNIST"
# url = "file:///somewhere/MNIST"
fs, path = fsspec.core.url_to_fs(url)
fs.ls(path) # this will always work
load_from_disk(path, fs) # only works for local FS
load_from_disk(url, fs) # only works for remote FS
```
### Expected behavior
one of `url` or `path` should always work
I think we extract path from given URL by using `fsspec.core.url_to_fs` instead of using `is_remote_filesystem` and `extract_path_from_uri` will fix this, since:
```
fsspec.core.url_to_fs("/somewhere/MNIST") -> LocalFs, '/somewhere/MNIST'
fsspec.core.url_to_fs("file:///somewhere/MNIST") -> LocalFs, '/somewhere/MNIST'
fsspec.core.url_to_fs("hdfs:///somewhere/MNIST") -> HDFS, '/somewhere/MNIST'
```
and
```
fsspec.core.url_to_fs("file:///somewhere/MNIST") == fsspec.core.url_to_fs("/somewhere/MNIST")
```
In theory, this wouldn't break anything, since giving local path and remote uri still works. It will only affect local URI (make it works too)
### Environment info
- `datasets` version: 2.5.1
- Platform: Linux-5.4.205.1**HIDDEN**
- Python version: 3.7.10
- PyArrow version: 8.0.0
- Pandas version: 1.2.4
|
open
|
https://github.com/huggingface/datasets/issues/5172
| 2022-10-27T12:03:20
| 2024-05-08T19:31:13
| null |
{
"login": "leoleoasd",
"id": 37735580,
"type": "User"
}
|
[] | false
|
[] |
1,425,355,111
| 5,171
|
Add PB and TB in convert_file_size_to_int
| null |
closed
|
https://github.com/huggingface/datasets/pull/5171
| 2022-10-27T09:50:31
| 2022-10-27T12:14:27
| 2022-10-27T12:12:30
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
1,425,301,835
| 5,170
|
[Caching] Deterministic hashing of torch tensors
|
Currently this fails
```python
import torch
from datasets.fingerprint import Hasher
t = torch.tensor([1.])
def func(x):
return t + x
hash1 = Hasher.hash(func)
t = torch.tensor([1.])
hash2 = Hasher.hash(func)
assert hash1 == hash2
```
Also as noticed in https://discuss.huggingface.co/t/dataset-cant-cache-models-outputs/24945, using a model in a `map` function doesn't work well with caching. Indeed the `bert-base-uncased` model has a different hash every time you reload it. Supporting torch tensors may also help in this case.
This can be fixed by registering a custom pickling functions for torch tensors - as we did for other objects such as CodeType, FunctionType and Regex in `py_utils.py`
|
closed
|
https://github.com/huggingface/datasets/issues/5170
| 2022-10-27T09:15:15
| 2022-11-02T17:18:43
| 2022-11-02T17:18:43
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,425,075,254
| 5,169
|
Add "ipykernel" to list of `co_filename`s to remove
|
Should resolve #5157
|
closed
|
https://github.com/huggingface/datasets/pull/5169
| 2022-10-27T05:56:17
| 2022-11-02T15:46:00
| 2022-11-02T15:43:20
|
{
"login": "gpucce",
"id": 32967787,
"type": "User"
}
|
[] | true
|
[] |
1,424,368,572
| 5,168
|
Fix CI require beam
|
This PR:
- Fixes the CI `require_beam`: before it was requiring PyTorch instead
```python
def require_beam(test_case):
if not config.TORCH_AVAILABLE:
test_case = unittest.skip("test requires PyTorch")(test_case)
return test_case
```
- Fixes a missing `require_beam` in `test_beam_based_builder_download_and_prepare_as_parquet`
- Refactors `require_beam` to use `pytest` (`skipif`) instead
|
closed
|
https://github.com/huggingface/datasets/pull/5168
| 2022-10-26T16:49:33
| 2022-10-27T09:25:19
| 2022-10-27T09:23:26
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,424,124,477
| 5,167
|
Add ffmpeg4 installation instructions in warnings
|
Adds instructions on how to install `ffmpeg=4` on Linux (relevant for Colab users).
Looks pretty ugly because I didn't find a way to check `ffmpeg` version from python (without `subprocess.call()`; `ctypes.util.find_library` doesn't work`), so the warning is raised on each decoding. Any suggestions on how to make it look nice are welcome!
This is how it looks on Colab:
|
closed
|
https://github.com/huggingface/datasets/pull/5167
| 2022-10-26T14:21:14
| 2022-10-27T09:01:12
| 2022-10-27T08:58:58
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[] | true
|
[] |
1,423,629,582
| 5,166
|
Support dill 0.3.6
|
This PR:
- ~~Unpins dill to allow installing dill>=0.3.6~~
- ~~Removes the fix on dill for >=0.3.6 because they implemented a deterministic mode (to be confirmed by @anivegesana)~~
- Pins dill<0.3.7 to allow latest dill 0.3.6
- Implements a fix for dill `save_function` for dill 0.3.6
- Additionally had to implement a fix for dill `save_code` and `_save_regex` for dill 0.3.6
- Fixes the CI so that the latest dill version is tested (besides the minimum 0.3.1.1 required by apache-beam 2.42.0)
Fix #5162.
|
closed
|
https://github.com/huggingface/datasets/pull/5166
| 2022-10-26T08:24:59
| 2022-10-28T05:41:05
| 2022-10-28T05:38:14
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,423,616,677
| 5,165
|
Memory explosion when trying to access 4d tensors in datasets cast to torch or np
|
### Describe the bug
When trying to access an item by index, in a datasets.Dataset cast to torch/np using `set_format` or `with_format`, we get a memory explosion if the item contains 4d (or above) tensors.
### Steps to reproduce the bug
MWE:
```python
from datasets import load_dataset
import numpy as np
def create_4d_tensor(item):
i = item["num_nodes"]
item["x_big"] = np.random.rand(i, 2*i, int(i/2), 1) + 1 # we create a big 4d tensor
return item
if __name__ == "__main__":
dataset = load_dataset(path=f"graphs-datasets/PROTEINS")
# This works
print(dataset["train"].format)
print(dataset["train"][0].keys())
dataset = dataset.map(
create_4d_tensor,
batched=False,
writer_batch_size=100,
)
# This works
print(dataset["train"].format)
print(dataset["train"][0].keys())
dataset.set_format("torch")
print(dataset["train"].format)
# This gets killed :(
print(dataset["train"][0].keys())
```
The problem likely comes from `format_table` [here](https://cs.github.com/huggingface/datasets/blob/f09f781be3278156ce3aa6ec90c1926b1846a78f/src/datasets/arrow_dataset.py#L2328)
### Expected behavior
No memory explosion when trying to access dataset items after cast.
### Environment info
- `datasets` version: 2.3.2
- Platform: Linux-5.14.0-1054-oem-x86_64-with-glibc2.29
- Python version: 3.8.10
- PyArrow version: 8.0.0
- Pandas version: 1.4.3
|
open
|
https://github.com/huggingface/datasets/issues/5165
| 2022-10-26T08:14:47
| 2022-10-26T08:14:47
| null |
{
"login": "clefourrier",
"id": 22726840,
"type": "User"
}
|
[] | false
|
[] |
1,422,813,247
| 5,164
|
WIP: drop labels in Image and Audio folders by default
|
will fix https://github.com/huggingface/datasets/issues/5153 and redundant labels displaying for most of the images datasets on the Hub (which are used just to store files)
TODO: discuss adding `drop_labels` (and `drop_metadata`) params to yaml
|
closed
|
https://github.com/huggingface/datasets/pull/5164
| 2022-10-25T17:21:49
| 2022-11-16T14:21:16
| 2022-11-02T14:03:02
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[] | true
|
[] |
1,422,540,337
| 5,163
|
Reduce default max `writer_batch_size`
|
Reduce the default writer_batch_size from 10k to 1k examples. Additionally, align the default values of `batch_size` and `writer_batch_size` in `Dataset.cast` with the values from the corresponding docstring.
|
closed
|
https://github.com/huggingface/datasets/pull/5163
| 2022-10-25T14:14:52
| 2022-10-27T12:19:27
| 2022-10-27T12:16:47
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
1,422,461,112
| 5,162
|
Pip-compile: Could not find a version that matches dill<0.3.6,>=0.3.6
|
### Describe the bug
When using `pip-compile` (part of `pip-tools`) to generate a pinned requirements file that includes `datasets`, a version conflict of `dill` appears.
It is caused by a transitive dependency conflict between `datasets` and `multiprocess`.
### Steps to reproduce the bug
```bash
$ echo "datasets" > requirements.in
$ pip install pip-tools
$ pip-compile requirements.in
Could not find a version that matches dill<0.3.6,>=0.3.6 (from datasets==2.6.1->-r requirements.in (line 1))
Tried: 0.2, 0.2, 0.2.1, 0.2.1, 0.2.2, 0.2.2, 0.2.3, 0.2.3, 0.2.4, 0.2.4, 0.2.5, 0.2.5, 0.2.6, 0.2.7, 0.2.7.1, 0.2.8, 0.2.8.1, 0.2.8.2, 0.2.9, 0.3.0, 0.3.1, 0.3.1.1, 0.3.2, 0.3.3, 0.3.3, 0.3.4, 0.3.4, 0.3.5, 0.3.5, 0.3.5.1, 0.3.5.1, 0.3.6, 0.3.6
Skipped pre-versions: 0.1a1, 0.2a1, 0.2a1, 0.2b1, 0.2b1
There are incompatible versions in the resolved dependencies:
dill<0.3.6 (from datasets==2.6.1->-r requirements.in (line 1))
dill>=0.3.6 (from multiprocess==0.70.14->datasets==2.6.1->-r requirements.in (line 1))
```
### Expected behavior
A correctly generated file `requirements.txt` with pinned dependencies
### Environment info
Tested with versions `2.6.1, 2.6.0, 2.5.2` on Python 3.8 and 3.10 on Ubuntu 20.04LTS and Python 3.10 on MacOS 12.6 (M1).
|
closed
|
https://github.com/huggingface/datasets/issues/5162
| 2022-10-25T13:23:50
| 2022-11-14T08:25:37
| 2022-10-28T05:38:15
|
{
"login": "Rijgersberg",
"id": 8604946,
"type": "User"
}
|
[] | false
|
[] |
1,422,371,748
| 5,161
|
Dataset can’t cache model’s outputs
|
### Describe the bug
Hi,
I try to cache some outputs of teacher model( Knowledge Distillation ) by using map function of Dataset library, while every time I run my code, I still recompute all the sequences. I tested Bert Model like this, I got different hash every single run, so any idea to deal with this?
### Steps to reproduce the bug
1. run below code
2. get different hash
```
from transformers import BertModel
from transformers import AutoTokenizer
import torch
token = ['hello']
model = BertModel.from_pretrained("bert-base-uncased").eval()
tok = AutoTokenizer.from_pretrained("bert-base-uncased")
def abcd():
with torch.no_grad():
out = model(**tok(token,return_tensors='pt'))[0]
# out = tok(token)
return out
from datasets.fingerprint import Hasher
my_func = abcd
print(Hasher.hash(my_func))
print(abcd())
```
### Expected behavior
I wanna cache all the model output
### Environment info
datasets:2.5.0
|
closed
|
https://github.com/huggingface/datasets/issues/5161
| 2022-10-25T12:19:00
| 2022-11-03T16:12:52
| 2022-11-03T16:12:51
|
{
"login": "jongjyh",
"id": 37979232,
"type": "User"
}
|
[] | false
|
[] |
1,422,193,938
| 5,160
|
Automatically add filename for image/audio folder
|
### Feature request
When creating a custom audio of image dataset, it would be great to automatically have access to the filename. It should be both:
a) Automatically displayed in the viewer
b) Automatically added as a column to the dataset when doing `load_dataset`
In `diffusers` our test rely quite heavily on images and audio files now and it's a bit tedious at the moment to download specific images from a datasets repo.
E.g. we have a dataset of images for tests in `diffusers`: https://huggingface.co/datasets/hf-internal-testing/diffusers-images
where it would be extremely nice to have direct access to the filename both visually on the datasets page (@severo ) as well as via the `load_datasets` function. We currently have some akward functionality to download images by path name: https://github.com/huggingface/diffusers/blob/2fb8fafa4b761f6fc144cf75a6f6f0ea6af3a1c1/src/diffusers/utils/testing_utils.py#L131
It would be much nicer to just go over `load_dataset(...)`
### Motivation
Intuitively the filename is something people understand directly. E.g if you upload a folder of images online, it's nice if you recognize the image as well as the filename next to it directly and that you're able to use it right away.
The label on the other hand is less intuitive to understand as you haven't added it yourself.
### Your contribution
Not sure if I have the time to add it myself anytime soon, but it would help us a lot for `diffusers`.
|
open
|
https://github.com/huggingface/datasets/issues/5160
| 2022-10-25T09:56:49
| 2022-10-26T16:51:46
| null |
{
"login": "patrickvonplaten",
"id": 23423619,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,422,172,080
| 5,159
|
fsspec lock reset in multiprocessing
|
`fsspec` added a clean way of resetting its lock - instead of doing it manually
|
closed
|
https://github.com/huggingface/datasets/pull/5159
| 2022-10-25T09:41:59
| 2022-11-03T20:51:15
| 2022-11-03T20:48:53
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
1,422,059,287
| 5,158
|
Fix language and license tag names in all Hub datasets
|
While working on this:
- #5137
we realized there are still many datasets with deprecated "languages" and "licenses" tag names (instead of "language" and "license").
This is a blocking issue: no subsequent PR can be opened to modify their metadata: a ValueError will be thrown.
We should fix the "language" and "license" tag names in all Hub datasets.
TODO:
- [x] Fix language and license tag names in 402 Hub datasets
CC: @julien-c
|
closed
|
https://github.com/huggingface/datasets/issues/5158
| 2022-10-25T08:19:29
| 2022-10-25T11:27:26
| 2022-10-25T10:42:19
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "dataset contribution",
"color": "0e8a16"
}
] | false
|
[] |
1,421,703,577
| 5,157
|
Consistent caching between python and jupyter
|
### Feature request
I hope this is not my mistake, currently if I use `load_dataset` from a python session on a custom dataset to do the preprocessing, it will be saved in the cache and in other python sessions it will be loaded from the cache, however calling the same from a jupyter notebook does not work, meaning the preprocessing starts from scratch.
If adjusting the hashes is impossible, is there a way to manually set dataset fingerprint to "force" this behaviour?
### Motivation
If this is not already the case and I am doing something wrong, it would be useful to have the two fingerprints consistent so one can create the dataset once and then try small things on jupyter without preprocessing everything again.
### Your contribution
I am happy to try a PR if you give me some pointers where the changes should happen
|
closed
|
https://github.com/huggingface/datasets/issues/5157
| 2022-10-25T01:34:33
| 2022-11-02T15:43:22
| 2022-11-02T15:43:22
|
{
"login": "gpucce",
"id": 32967787,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,421,667,125
| 5,156
|
Unable to download dataset using Azure Data Lake Gen 2
|
### Describe the bug
When using the DatasetBuilder method with the credentials for the cloud storage Azure Data Lake (adl) Gen2, the following error is showed:
```
Traceback (most recent call last):
File "download_hf_dataset.py", line 143, in <module>
main()
File "download_hf_dataset.py", line 102, in main
builder.download_and_prepare(save_dir, storage_options=storage_options, max_shard_size="250MB", file_format="parquet")
File "/home/clarisses/miniconda3/envs/hf_datasets_env/lib/python3.8/site-packages/datasets/builder.py", line 671, in download_and_prepare
fs_token_paths = fsspec.get_fs_token_paths(output_dir, storage_options=storage_options)
File "/home/clarisses/miniconda3/envs/hf_datasets_env/lib/python3.8/site-packages/fsspec/core.py", line 639, in get_fs_token_paths
fs = cls(**options)
File "/home/clarisses/miniconda3/envs/hf_datasets_env/lib/python3.8/site-packages/fsspec/spec.py", line 76, in __call__
obj = super().__call__(*args, **kwargs)
TypeError: __init__() got an unexpected keyword argument 'account_name'
```
If I don't pass the storage_options argument (leave it as None), it requires the credentials used in ADL Gen 1:
`TypeError: __init__() missing 3 required positional arguments: 'tenant_id', 'client_id', and 'client_secret'`
Thus, it is not possible to download a dataset from the cloud using Azure Data Lake (adl) Gen2.
### Steps to reproduce the bug
Assuming that you have an account on Azure and at Storage Account that can be used for reproduce:
1. Create a dict with the format to connect to Azure Data Lake Gen 2
```
storage_options = {"account_name": ACCOUNT_NAME, "account_key": ACCOUNT_KEY) # gen 2 filesystem
```
2. Create a dataset builder for any HF hosted dataset
```
builder = load_dataset_builder(dataset_name)
```
3. Try to download the dataset passing the storage_options as an argument
```
save_dir = 'adl://my_save_dir'
builder.download_and_prepare(save_dir, storage_options=storage_options, max_shard_size="250MB", file_format="parquet")
```
### Expected behavior
Not seeing the error mentioned above and being able to download the dataset to the provided path on ADL
### Environment info
- `datasets` version: 2.6.1
- Platform: Linux-5.15.0-46-generic-x86_64-with-glibc2.17
- Python version: 3.8.13
- PyArrow version: 9.0.0
- Pandas version: 1.5.1
|
closed
|
https://github.com/huggingface/datasets/issues/5156
| 2022-10-25T00:43:18
| 2024-02-15T09:48:36
| 2022-11-17T23:37:08
|
{
"login": "clarissesimoes",
"id": 87379512,
"type": "User"
}
|
[] | false
|
[] |
1,421,278,748
| 5,155
|
TextConfig: added "errors"
|
This patch adds the ability to set the `errors` option of `open` for loading text datasets. I needed it because some data I had scraped had bad bytes in it, so I needed `errors='ignore'`.
|
closed
|
https://github.com/huggingface/datasets/pull/5155
| 2022-10-24T18:56:52
| 2022-11-03T13:38:13
| 2022-11-03T13:35:35
|
{
"login": "NightMachinery",
"id": 36224762,
"type": "User"
}
|
[] | true
|
[] |
1,421,161,992
| 5,154
|
Test latest fsspec in CI
|
Following the discussion in https://discuss.huggingface.co/t/attributeerror-module-fsspec-has-no-attribute-asyn/19255 I think we need to test the latest fsspec in the CI
|
closed
|
https://github.com/huggingface/datasets/pull/5154
| 2022-10-24T17:18:13
| 2023-09-24T10:06:06
| 2022-10-25T09:30:45
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
1,420,833,457
| 5,153
|
default Image/AudioFolder infers labels when there is no metadata files even if there is only one dir
|
### Describe the bug
By default FolderBasedBuilder infers labels if there is not metadata files, even if it's meaningless (for example, they are in a single directory or in the root folder, see this repo as an example: https://huggingface.co/datasets/patrickvonplaten/audios
As this is a corner case for quick exploration of images or audios on the Hub.
### Steps to reproduce the bug
If you have directory like this:
```
repo
image1.jpg
image2.jpg
image3.jpg
```
or
```
repo
data
image1.jpg
image2.jpg
image3.jpg
```
doing `ds = load_dataset(repo)` would create `label` feature:
```python
print(ds["train"][0])
>> {'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=500x375 at 0x7FB5326468E0>, 'label': 0}
```
Also, if you have the following structure:
```
repo
data
image1.jpg
image2.jpg
image3.jpg
image4.jpg
image5.jpg
image6.jpg
```
it will infer two labels:
```python
print(ds["train"][0])
print(ds["train"][-1])
>> {'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=500x375 at 0x7FB5326468E0>, 'label': 1}
>> {'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=500x415 at 0x7FB5326555B0>, 'label': 0}
```
### Expected behavior
We should have only one base feature (Image/Audio) in such cases.
### Environment info
all versions of `datasets`
|
closed
|
https://github.com/huggingface/datasets/issues/5153
| 2022-10-24T13:28:18
| 2022-11-15T16:31:10
| 2022-11-15T16:31:09
|
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,420,808,919
| 5,152
|
refactor FolderBasedBuilder and Image/AudioFolder tests
|
Tests for FolderBasedBuilder, ImageFolder and AudioFolder are mostly duplicating each other. They need to be refactored and Audio/ImageFolder should have only tests specific to the loader.
|
open
|
https://github.com/huggingface/datasets/issues/5152
| 2022-10-24T13:11:52
| 2022-10-24T13:11:52
| null |
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[
{
"name": "refactoring",
"color": "B67A40"
}
] | false
|
[] |
1,420,791,163
| 5,151
|
Add support to create different configs with `push_to_hub` (+ inferring configs from directories with package managers?)
|
Now one can push only different splits within one default config of a dataset.
Would be nice to allow something like:
```
ds.push_to_hub(repo_name, config=config_name)
```
I'm not sure, but this will probably require changes in `data_files.py` patterns. If so, it would also allow to create different configs for packaged modules datasets.
|
open
|
https://github.com/huggingface/datasets/issues/5151
| 2022-10-24T12:59:18
| 2022-11-04T14:55:20
| null |
{
"login": "polinaeterna",
"id": 16348744,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,420,684,999
| 5,150
|
Problems after upgrading to 2.6.1
|
### Describe the bug
Loading a dataset_dict from disk with `load_from_disk` is now creating a `KeyError "length"` that was not occurring in v2.5.2.
Context:
- Each individual dataset in the dict is created with `Dataset.from_pandas`
- The dataset_dict is create from a dict of `Dataset`s, e.g., `DatasetDict({"train": train_ds, "validation": val_ds})
- The pandas dataframe, besides text columns, has a column with a dictionary inside and potentially different keys in each row. Correctly the `Dataset.from_pandas` function adds `key: None` to all dictionaries in each row so that the schema can be correctly inferred.
### Steps to reproduce the bug
Steps to reproduce:
- Upgrade to datasets==2.6.1
- Create a dataset from pandas dataframe with `Dataset.from_pandas`
- Create a dataset_dict from a dict of `Dataset`s, e.g., `DatasetDict({"train": train_ds, "validation": val_ds})
- Save to disk with the `save` function
### Expected behavior
Same as in v2.5.2, that is load from disk without errors
### Environment info
- `datasets` version: 2.6.1
- Platform: Linux-5.4.209-129.367.amzn2int.x86_64-x86_64-with-glibc2.26
- Python version: 3.9.13
- PyArrow version: 9.0.0
- Pandas version: 1.5.1
|
open
|
https://github.com/huggingface/datasets/issues/5150
| 2022-10-24T11:32:36
| 2024-05-12T07:40:03
| null |
{
"login": "pietrolesci",
"id": 61748653,
"type": "User"
}
|
[] | false
|
[] |
1,420,415,639
| 5,149
|
Make iter_files deterministic
|
Fix #5145.
|
closed
|
https://github.com/huggingface/datasets/pull/5149
| 2022-10-24T08:16:27
| 2022-10-27T09:53:23
| 2022-10-27T09:51:09
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,420,219,222
| 5,148
|
Cannot find the rvl_cdip dataset
|
Hi,
I am trying to use load_dataset to load the official "rvl_cdip" dataset but getting an error.
dataset = load_dataset("rvl_cdip")
Couldn't find 'rvl_cdip' on the Hugging Face Hub either: FileNotFoundError: Couldn't find the file at https://raw.githubusercontent.com/huggingface/datasets/master/datasets/rvl_cdip/rvl_cdip.py
Regards,
|
closed
|
https://github.com/huggingface/datasets/issues/5148
| 2022-10-24T04:57:42
| 2022-10-24T12:23:47
| 2022-10-24T06:25:28
|
{
"login": "santule",
"id": 20509836,
"type": "User"
}
|
[] | false
|
[] |
1,419,522,275
| 5,147
|
Allow ignoring kwargs inside fn_kwargs during dataset.map's fingerprinting
|
### Feature request
`dataset.map` accepts a `fn_kwargs` that is passed to `fn`. Currently, the whole `fn_kwargs` is used by `fingerprint_transform` to calculate the new fingerprint.
I'd like to be able to inform `fingerprint_transform` which `fn_kwargs` shoud/shouldn't be taken into account during hashing.
Of course, users should be aware to properly use this new feature, just like the internal usages of `fingerprint_transform` [does](https://github.com/huggingface/datasets/blob/2699593b33ee63d17aad2a2bfddedd38a8df57b8/src/datasets/arrow_dataset.py#L2700).
### Motivation
This is originally motivated by https://github.com/huggingface/transformers/pull/18351#issuecomment-1263588680.
Nonetheless, consider a more general processing function that accepts a kwarg that does not influence it's output:
```python
def fn(example, verbose=False):
...
```
Then `dataset.map(fn, verbose=True)` would not benefit from dataset caching.
I'm not sure if other methods in the `Dataset` API could benefit from this feature.
### Your contribution
Based on `fingerprint_transform `'s `wrapper` function [here](https://github.com/huggingface/datasets/blob/c59cc34fcd2a369d27b77cc678017f5976a926a9/src/datasets/fingerprint.py#L443), it seems to me that it should be possible to make `.map`/`._map_single` accept something like `fn_use_fingerprint_kwargs`/`fn_ignore_fingerprint_kwargs` (probably another arg name). This would then be used by `fingerprint_transform.wrapper` to better/more flexibly hash the transformation.
I could contribute with a PR if this feature and approach look good to you.
|
open
|
https://github.com/huggingface/datasets/issues/5147
| 2022-10-22T21:46:38
| 2022-11-01T22:19:07
| null |
{
"login": "falcaopetri",
"id": 8387736,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,418,331,282
| 5,146
|
Delete duplicate issue template file
|
A conflict between two PRs:
- #5116
- #5136
was not properly resolved, resulting in a duplicate issue template.
This PR removes the duplicate template.
|
closed
|
https://github.com/huggingface/datasets/pull/5146
| 2022-10-21T13:18:46
| 2022-10-21T13:52:30
| 2022-10-21T13:50:04
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,418,005,452
| 5,145
|
Dataset order is not deterministic with ZIP archives and `iter_files`
|
### Describe the bug
For the `beans` dataset (did not try on other), the order of samples is not the same on different machines. Tested on my local laptop, github actions machine, and ec2 instance. The three yield a different order.
### Steps to reproduce the bug
In a clean docker container or conda environment with datasets==2.6.1, run
```python
from datasets import load_dataset
from pprint import pprint
data = load_dataset("beans", split="validation")
pprint(data["image_file_path"])
```
### Expected behavior
The order of the images is the same on all machines.
### Environment info
On the EC2 instance:
```
- `datasets` version: 2.6.1
- Platform: Linux-4.14.291-218.527.amzn2.x86_64-x86_64-with-glibc2.2.5
- Python version: 3.7.10
- PyArrow version: 9.0.0
- Pandas version: 1.3.5
- Numpy version: not checked
```
On my local laptop:
```
- `datasets` version: 2.6.1
- Platform: Linux-5.15.0-50-generic-x86_64-with-glibc2.35
- Python version: 3.9.12
- PyArrow version: 7.0.0
- Pandas version: 1.3.5
- Numpy version: 1.23.1
```
On github actions:
```
- `datasets` version: 2.6.1
- Platform: Linux-5.15.0-1022-azure-x86_64-with-glibc2.2.5
- Python version: 3.8.14
- PyArrow version: 9.0.0
- Pandas version: 1.5.1
- Numpy version: 1.23.4
```
|
closed
|
https://github.com/huggingface/datasets/issues/5145
| 2022-10-21T09:00:03
| 2022-10-27T09:51:49
| 2022-10-27T09:51:10
|
{
"login": "fxmarty",
"id": 9808326,
"type": "User"
}
|
[] | false
|
[] |
1,417,974,731
| 5,144
|
Inconsistent documentation on map remove_columns
|
### Describe the bug
The page [process](https://huggingface.co/docs/datasets/process) says this about the parameter `remove_columns` of the function `map`:
When you remove a column, it is only removed after the example has been provided to the mapped function.
So it seems that the `remove_columns` parameter removes after the mapped functions.
However, another page, [the documentation of the function map](https://huggingface.co/docs/datasets/v2.6.1/en/package_reference/main_classes#datasets.Dataset.map.remove_columns) says:
Columns will be removed before updating the examples with the output of `function`, i.e. if `function` is adding columns with names in remove_columns, these columns will be kept.
So one page says "after the mapped function" and another says "before the mapped function."
Is there something wrong?
### Steps to reproduce the bug
Not about code.
### Expected behavior
consistent about the descriptions of the behavior of the parameter `remove_columns` in the function `map`.
### Environment info
datasets V2.6.0
|
closed
|
https://github.com/huggingface/datasets/issues/5144
| 2022-10-21T08:37:53
| 2022-11-15T14:15:10
| 2022-11-15T14:15:10
|
{
"login": "zhaowei-wang-nlp",
"id": 22047467,
"type": "User"
}
|
[
{
"name": "documentation",
"color": "0075ca"
},
{
"name": "duplicate",
"color": "cfd3d7"
},
{
"name": "good first issue",
"color": "7057ff"
},
{
"name": "hacktoberfest",
"color": "DF8D62"
}
] | false
|
[] |
1,416,837,186
| 5,143
|
DownloadManager Git LFS support
|
### Feature request
Maybe I'm mistaken but the `DownloadManager` does not support extracting git lfs files out of the box right?
Using `dl_manager.download()` or `dl_manager.download_and_extract()` still returns lfs files afaict.
Is there a good way to write a dataset loading script for a repo with lfs files?
### Motivation
/
### Your contribution
/
|
closed
|
https://github.com/huggingface/datasets/issues/5143
| 2022-10-20T15:29:29
| 2022-10-20T17:17:10
| 2022-10-20T17:17:10
|
{
"login": "Muennighoff",
"id": 62820084,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false
|
[] |
1,416,317,678
| 5,142
|
Deprecate num_proc parameter in DownloadManager.extract
|
fixes #5132 : Deprecated the `num_proc` parameter in `DownloadManager.extract` by passing `num_proc` parameter to `map_nested` .
|
closed
|
https://github.com/huggingface/datasets/pull/5142
| 2022-10-20T09:52:52
| 2022-10-25T18:06:56
| 2022-10-25T15:56:45
|
{
"login": "ayushthe1",
"id": 114604338,
"type": "User"
}
|
[] | true
|
[] |
1,415,479,438
| 5,141
|
Raise ImportError instead of OSError
|
fixes #5134 : Replaced OSError with ImportError if required extraction library is not installed.
|
closed
|
https://github.com/huggingface/datasets/pull/5141
| 2022-10-19T19:30:05
| 2022-10-25T15:59:25
| 2022-10-25T15:56:58
|
{
"login": "ayushthe1",
"id": 114604338,
"type": "User"
}
|
[] | true
|
[] |
1,415,075,530
| 5,140
|
Make the KeyHasher FIPS compliant
|
MD5 is not FIPS compliant thus I am proposing this minimal change to make datasets package FIPS compliant
|
closed
|
https://github.com/huggingface/datasets/pull/5140
| 2022-10-19T14:25:52
| 2022-11-07T16:20:43
| 2022-11-07T16:20:43
|
{
"login": "vvalouch",
"id": 22592860,
"type": "User"
}
|
[] | true
|
[] |
1,414,642,723
| 5,137
|
Align task tags in dataset metadata
|
## Describe
Once we have agreed on a common naming for task tags for all open source projects, we should align on them.
## Steps
- [x] Align task tags in canonical datasets
- [x] task_categories: 4 datasets
- [x] task_ids (by @lhoestq)
- [x] Open PRs in community datasets
- [x] task_categories: 451 datasets
- [x] task_ids: 556 datasets
|
closed
|
https://github.com/huggingface/datasets/issues/5137
| 2022-10-19T09:41:42
| 2022-11-10T05:25:58
| 2022-10-25T06:17:00
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "dataset contribution",
"color": "0e8a16"
}
] | false
|
[] |
1,414,492,139
| 5,136
|
Update docs once dataset scripts transferred to the Hub
|
Todo:
- [x] Update docs:
- [x] Datasets on GitHub (legacy)
- [x] Load: offline
- [x] About dataset load:
- [x] Maintaining integrity
- [x] Security
- [x] Update docstrings:
- [x] Inspect:
- [x] get_dataset_config_info
- [x] get_dataset_split_names
- [x] Load:
- [x] dataset_module_factory
- [x] load_dataset_builder
- [x] load_dataset
- [x] Remove `ADD_NEW_DATASET.md`
- [x] Update `.github/ISSUE_TEMPLATE/config.yml`
Fix #5135.
|
closed
|
https://github.com/huggingface/datasets/pull/5136
| 2022-10-19T07:58:27
| 2022-10-20T08:12:21
| 2022-10-20T08:10:00
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,414,413,519
| 5,135
|
Update docs once dataset scripts transferred to the Hub
|
## Describe the bug
As discussed in:
- https://github.com/huggingface/hub-docs/pull/423#pullrequestreview-1146083701
we should update our docs once dataset scripts have been transferred to the Hub (and removed from GitHub):
- #4974
Concretely:
- [x] Datasets on GitHub (legacy): https://huggingface.co/docs/datasets/main/en/share#datasets-on-github-legacy
- [x] ADD_NEW_DATASET: https://github.com/huggingface/datasets/blob/main/ADD_NEW_DATASET.md
- ...
This PR complements the work of:
- #5067
This PR is a follow-up of PRs:
- #3777
CC: @julien-c
|
closed
|
https://github.com/huggingface/datasets/issues/5135
| 2022-10-19T06:58:19
| 2022-10-20T08:10:01
| 2022-10-20T08:10:01
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "documentation",
"color": "0075ca"
}
] | false
|
[] |
1,413,623,687
| 5,134
|
Raise ImportError instead of OSError if required extraction library is not installed
|
According to the official Python docs, `OSError` should be thrown in the following situations:
> This exception is raised when a system function returns a system-related error, including I/O failures such as “file not found” or “disk full” (not for illegal argument types or other incidental errors).
Hence, it makes more sense to raise `ImportError` instead of `OSError` when the required extraction/decompression library is not installed.
|
closed
|
https://github.com/huggingface/datasets/issues/5134
| 2022-10-18T17:53:46
| 2022-10-25T15:56:59
| 2022-10-25T15:56:59
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
},
{
"name": "good first issue",
"color": "7057ff"
},
{
"name": "hacktoberfest",
"color": "DF8D62"
}
] | false
|
[] |
1,413,623,462
| 5,133
|
Tensor operation not functioning in dataset mapping
|
## Describe the bug
I'm doing a torch.mean() operation in data preprocessing, and it's not working.
## Steps to reproduce the bug
```
from transformers import pipeline
import torch
import numpy as np
from datasets import load_dataset
device = 'cuda:0'
raw_dataset = load_dataset("glue", "sst2")
feature_extraction = pipeline('feature-extraction', 'bert-base-uncased', device=device)
def extracted_data(examples):
# feature = torch.tensor(feature_extraction(examples['sentence'], batch_size=16), device=device)
# feature = torch.mean(feature, dim=1)
feature = np.asarray(feature_extraction(examples['sentence'], batch_size=16)).squeeze().mean(1)
print(feature.shape)
return {'feature': feature}
extracted_dataset = raw_dataset.map(extracted_data, batched=True, batch_size=16)
```
## Results
When running with torch.mean(), the shape printed out is [16, seq_len, 768], which is exactly the same before the operation. While numpy works just fine, which gives [16, 768].
## Environment info
- `datasets` version: 2.6.1
- Platform: Linux-4.4.0-142-generic-x86_64-with-glibc2.31
- Python version: 3.10.6
- PyArrow version: 9.0.0
- Pandas version: 1.5.0
|
closed
|
https://github.com/huggingface/datasets/issues/5133
| 2022-10-18T17:53:35
| 2022-10-19T04:15:45
| 2022-10-19T04:15:44
|
{
"login": "xinghaow99",
"id": 50691954,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,413,607,306
| 5,132
|
Depracate `num_proc` parameter in `DownloadManager.extract`
|
The `num_proc` parameter is only present in `DownloadManager.extract` but not in `StreamingDownloadManager.extract`, making it impossible to support streaming in the dataset scripts that use it (`openwebtext` and `the_pile_stack_exchange`). We can avoid this situation by deprecating this parameter and passing `DownloadConfig`'s `num_proc` to `map_nested` instead, as it's done in `DownloadManager.download`.
|
closed
|
https://github.com/huggingface/datasets/issues/5132
| 2022-10-18T17:41:05
| 2022-10-25T15:56:46
| 2022-10-25T15:56:46
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
},
{
"name": "good first issue",
"color": "7057ff"
},
{
"name": "hacktoberfest",
"color": "DF8D62"
}
] | false
|
[] |
1,413,534,863
| 5,131
|
WikiText 103 tokenizer hangs
|
See issue here: https://github.com/huggingface/transformers/issues/19702
|
closed
|
https://github.com/huggingface/datasets/issues/5131
| 2022-10-18T16:44:00
| 2023-08-08T08:42:40
| 2023-07-21T14:41:51
|
{
"login": "TrentBrick",
"id": 12433427,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,413,435,000
| 5,130
|
Avoid extra cast in `class_encode_column`
|
Pass the updated features to `map` to avoid the `cast` in `class_encode_column`.
|
closed
|
https://github.com/huggingface/datasets/pull/5130
| 2022-10-18T15:31:24
| 2022-10-19T11:53:02
| 2022-10-19T11:50:46
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
1,413,031,664
| 5,129
|
unexpected `cast` or `class_encode_column` result after `rename_column`
|
## Describe the bug
When invoke `cast` or `class_encode_column` to a colunm renamed by `rename_column` , it will convert all the variables in this column into one variable. I also run this script in version 2.5.2, this bug does not appear. So I switched to the older version.
## Steps to reproduce the bug
```python
from datasets import load_dataset
dataset = load_dataset("amazon_reviews_multi", "en")
data = dataset['train']
data = data.remove_columns(
[
"review_id",
"product_id",
"reviewer_id",
"review_title",
"language",
"product_category",
]
)
data = data.rename_column("review_body", "text")
data1 = data.class_encode_column("stars")
print(set(data1.data.columns[0]))
# output: {<pyarrow.Int64Scalar: 4>, <pyarrow.Int64Scalar: 2>, <pyarrow.Int64Scalar: 3>, <pyarrow.Int64Scalar: 0>, <pyarrow.Int64Scalar: 1>}
data = data.rename_column("stars", "label")
print(set(data.data.columns[0]))
# output: {<pyarrow.Int32Scalar: 5>, <pyarrow.Int32Scalar: 4>, <pyarrow.Int32Scalar: 1>, <pyarrow.Int32Scalar: 3>, <pyarrow.Int32Scalar: 2>}
data2 = data.class_encode_column("label")
print(set(data2.data.columns[0]))
# output: {<pyarrow.Int64Scalar: 0>}
```
## Expected results
the last print should be:
{<pyarrow.Int64Scalar: 4>, <pyarrow.Int64Scalar: 2>, <pyarrow.Int64Scalar: 3>, <pyarrow.Int64Scalar: 0>, <pyarrow.Int64Scalar: 1>}
## Actual results
but it output:
{<pyarrow.Int64Scalar: 0>}
## Environment info
- `datasets` version: 2.6.1
- Platform: macOS-12.5.1-arm64-arm-64bit
- Python version: 3.10.6
- PyArrow version: 9.0.0
- Pandas version: 1.5.0
|
closed
|
https://github.com/huggingface/datasets/issues/5129
| 2022-10-18T11:15:24
| 2022-10-19T03:02:26
| 2022-10-19T03:02:26
|
{
"login": "quaeast",
"id": 35144675,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,412,783,855
| 5,128
|
Make filename matching more robust
|
Fix #5046
|
closed
|
https://github.com/huggingface/datasets/pull/5128
| 2022-10-18T08:22:48
| 2022-10-28T13:07:38
| 2022-10-28T13:05:06
|
{
"login": "riccardobucco",
"id": 9295277,
"type": "User"
}
|
[] | true
|
[] |
1,411,897,544
| 5,127
|
[WIP] WebDataset export
|
I added a first draft of the `IterableDataset.to_wds` method.
You can use it to savea dataset loaded in streamign mode as a webdataset locally.
The API can be further improved to allow to export to a cloud storage like the HF Hub.
I also included sharding with a default max shard size of 500MB (uncompressed), and it is single-processed fo rnow.
Choosing the number of shards is not implemented yet - though if we know the size of the `IterableDataset` this is probably doable`.
For example
```python
>>> from datasets import load_dataset
>>> ds = load_dataset("rotten_tomatoes", split="train", streaming=True)
>>> ds.to_wds("output_dir", compress=True)
>>> import webdataset as wds
>>> ds = wds.WebDataset("output_dir/rotten_tomatoes-train-000000.tar.gz").decode()
>>> next(iter(ds))
{'__key__': '0',
'__url__': 'output_dir/rotten_tomatoes-train-000000.tar.gz',
'label.cls': 1,
'text.txt': 'the rock is destined to be the 21st century\'s new ..., jean-claud van damme or steven segal .'}
```
### Implementation details
The WebDataset format is made of TAR archives containing a series of files per example. For example one pair of `image.jpg` and `label.cls` for image classification.
WebDataset automatically decodes serialized data based on the extension of the files, and output a dictionary. For example `{"image.png": np.array(...), "label.cls": 0}` if you choose the numpy decoding.
To use the automatic decoding, I store each field of each example as a file with its corresponding extension (jpg, json, cls, etc.)
While this is useful to end up with a dictionary with one key per column and appropriate decoding, it can create huge TAR archives if the dataset is made of small samples of text - probably because of useless TAR metadata for each file. This also makes loading super slow: iterating on SQuAD takes 50sec vs 7sec using `datasets` in streaming mode.
I haven't taken a look at alternatives for text datasets made out of small samples, but for image datasets this can already be used to run some benchmarks.
|
closed
|
https://github.com/huggingface/datasets/pull/5127
| 2022-10-17T16:50:22
| 2024-01-11T06:27:04
| 2024-01-08T14:25:43
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
1,411,757,124
| 5,126
|
Fix class name of symbolic link
|
Fix #5098
|
closed
|
https://github.com/huggingface/datasets/pull/5126
| 2022-10-17T15:11:02
| 2022-11-14T14:40:18
| 2022-11-14T14:40:18
|
{
"login": "riccardobucco",
"id": 9295277,
"type": "User"
}
|
[] | true
|
[] |
1,411,602,813
| 5,125
|
Add `pyproject.toml` for `black`
|
Add `pyproject.toml` as a config file for the `black` tool to support VS Code's auto-formatting on save (and to be more consistent with the other HF projects).
|
closed
|
https://github.com/huggingface/datasets/pull/5125
| 2022-10-17T13:38:47
| 2024-11-20T13:36:11
| 2022-10-17T14:21:09
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
1,411,159,725
| 5,124
|
Install tensorflow-macos dependency conditionally
|
Fix #5118.
|
closed
|
https://github.com/huggingface/datasets/pull/5124
| 2022-10-17T08:45:08
| 2022-10-19T09:12:17
| 2022-10-19T09:10:06
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,410,828,756
| 5,123
|
datasets freezes with streaming mode in multiple-gpu
|
## Describe the bug
Hi. I am using this dataloader, which is for processing large datasets in streaming mode mentioned in one of examples of huggingface. I am using it to read c4: https://github.com/huggingface/transformers/blob/b48ac1a094e572d6076b46a9e4ed3e0ebe978afc/examples/research_projects/codeparrot/scripts/codeparrot_training.py#L22
During using multi-gpu in accelerator in one node, the code freezes, but works for 1 GPU:
```
10/16/2022 14:18:46 - INFO - datasets.info - Loading Dataset Infos from /home/jack/.cache/huggingface/modules/datasets_modules/datasets/c4/df532b158939272d032cc63ef19cd5b83e9b4d00c922b833e4cb18b2e9869b01
Steps: 0%| | 0/400000 [00:00<?, ?it/s]10/16/2022 14:18:47 - INFO - torch.utils.data.dataloader - Shared seed (135290893754684706) sent to store on rank 0
```
# Code to reproduce
please run this code with `accelerate launch code.py`
```
from accelerate import Accelerator
from accelerate.logging import get_logger
from datasets import load_dataset
from torch.utils.data.dataloader import DataLoader
import torch
from datasets import load_dataset
from transformers import AutoTokenizer
import torch
from accelerate.logging import get_logger
from torch.utils.data import IterableDataset
from torch.utils.data.datapipes.iter.combinatorics import ShufflerIterDataPipe
logger = get_logger(__name__)
class ConstantLengthDataset(IterableDataset):
"""
Iterable dataset that returns constant length chunks of tokens from stream of text files.
Args:
tokenizer (Tokenizer): The processor used for proccessing the data.
dataset (dataset.Dataset): Dataset with text files.
infinite (bool): If True the iterator is reset after dataset reaches end else stops.
max_seq_length (int): Length of token sequences to return.
num_of_sequences (int): Number of token sequences to keep in buffer.
chars_per_token (int): Number of characters per token used to estimate number of tokens in text buffer.
"""
def __init__(
self,
tokenizer,
dataset,
infinite=False,
max_seq_length=1024,
num_of_sequences=1024,
chars_per_token=3.6,
):
self.tokenizer = tokenizer
# self.concat_token_id = tokenizer.bos_token_id
self.dataset = dataset
self.max_seq_length = max_seq_length
self.epoch = 0
self.infinite = infinite
self.current_size = 0
self.max_buffer_size = max_seq_length * chars_per_token * num_of_sequences
self.content_field = "text"
def __iter__(self):
iterator = iter(self.dataset)
more_examples = True
while more_examples:
buffer, buffer_len = [], 0
while True:
if buffer_len >= self.max_buffer_size:
break
try:
buffer.append(next(iterator)[self.content_field])
buffer_len += len(buffer[-1])
except StopIteration:
if self.infinite:
iterator = iter(self.dataset)
self.epoch += 1
logger.info(f"Dataset epoch: {self.epoch}")
else:
more_examples = False
break
tokenized_inputs = self.tokenizer(buffer, truncation=False)["input_ids"]
all_token_ids = []
for tokenized_input in tokenized_inputs:
all_token_ids.extend(tokenized_input)
for i in range(0, len(all_token_ids), self.max_seq_length):
input_ids = all_token_ids[i : i + self.max_seq_length]
if len(input_ids) == self.max_seq_length:
self.current_size += 1
yield torch.tensor(input_ids)
def shuffle(self, buffer_size=1000):
return ShufflerIterDataPipe(self, buffer_size=buffer_size)
def create_dataloaders(tokenizer, accelerator):
ds_kwargs = {"streaming": True}
# In distributed training, the load_dataset function gaurantees that only one process
# can concurrently download the dataset.
datasets = load_dataset(
"c4",
"en",
cache_dir="cache_dir",
**ds_kwargs,
)
train_data, valid_data = datasets["train"], datasets["validation"]
with accelerator.main_process_first():
train_data = train_data.shuffle(buffer_size=10000, seed=None)
train_dataset = ConstantLengthDataset(
tokenizer,
train_data,
infinite=True,
max_seq_length=256,
)
valid_dataset = ConstantLengthDataset(
tokenizer,
valid_data,
infinite=False,
max_seq_length=256,
)
train_dataset = train_dataset.shuffle(buffer_size=10000)
train_dataloader = DataLoader(train_dataset, batch_size=160, shuffle=True)
eval_dataloader = DataLoader(valid_dataset, batch_size=160)
return train_dataloader, eval_dataloader
def main():
# Accelerator.
logging_dir = "data_save_dir/log"
accelerator = Accelerator(
gradient_accumulation_steps=1,
mixed_precision="bf16",
log_with="tensorboard",
logging_dir=logging_dir,
)
# We need to initialize the trackers we use, and also store our configuration.
# The trackers initializes automatically on the main process.
if accelerator.is_main_process:
accelerator.init_trackers("test")
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
# Load datasets and create dataloaders.
train_dataloader, _ = create_dataloaders(tokenizer, accelerator)
train_dataloader = accelerator.prepare(train_dataloader)
for step, batch in enumerate(train_dataloader, start=1):
print(step)
accelerator.end_training()
if __name__ == "__main__":
main()
```
## Results expected
Being able to run the code for streamining datasets with multi-gpu
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 2.5.2
- Platform: linux
- Python version: 3.9.12
- PyArrow version: 9.0.0
@lhoestq I do not have any idea why this freezing happens, and I removed the streaming mode and this was working fine, so I know this is caused by streaming mode of the dataloader part not working well with multi-gpu setting. Since datasets are large, I hope to keep the streamining mode. I very much appreciate your help.
|
open
|
https://github.com/huggingface/datasets/issues/5123
| 2022-10-17T03:28:16
| 2023-05-14T06:55:20
| null |
{
"login": "jackfeinmann5",
"id": 59409879,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,410,732,403
| 5,122
|
Add warning
|
Fixes: #5105
I think removing the directory with warning is a better solution for this issue. Because if we decide to keep existing files in directory, then we should deal with the case providing same directory for several datasets! Which we know is not possible since `dataset_info.json` exists in that directory.
|
closed
|
https://github.com/huggingface/datasets/pull/5122
| 2022-10-17T01:30:37
| 2022-11-05T12:23:53
| 2022-11-05T12:23:53
|
{
"login": "Salehbigdeli",
"id": 34204311,
"type": "User"
}
|
[] | true
|
[] |
1,410,681,067
| 5,121
|
Bugfix ignore function when creating new_fingerprint for caching
|
maybe fixes: #5109
|
closed
|
https://github.com/huggingface/datasets/pull/5121
| 2022-10-17T00:03:43
| 2022-10-17T12:39:36
| 2022-10-17T12:39:36
|
{
"login": "Salehbigdeli",
"id": 34204311,
"type": "User"
}
|
[] | true
|
[] |
1,410,641,221
| 5,120
|
Fix `tqdm` zip bug
|
This PR solves #5117, by wrapping the entire `zip` clause in tqdm.
For more information, please checkout this Stack Overflow thread:
https://stackoverflow.com/questions/41171191/tqdm-progressbar-and-zip-built-in-do-not-work-together
|
closed
|
https://github.com/huggingface/datasets/pull/5120
| 2022-10-16T22:19:18
| 2022-10-23T10:27:53
| 2022-10-19T08:53:17
|
{
"login": "david1542",
"id": 9879252,
"type": "User"
}
|
[] | true
|
[] |
1,410,561,363
| 5,119
|
[TYPO] Update new_dataset_script.py
| null |
closed
|
https://github.com/huggingface/datasets/pull/5119
| 2022-10-16T17:36:49
| 2022-10-19T09:48:19
| 2022-10-19T09:45:59
|
{
"login": "cakiki",
"id": 3664563,
"type": "User"
}
|
[] | true
|
[] |
1,410,547,373
| 5,118
|
Installing `datasets` on M1 computers
|
## Describe the bug
I wanted to install `datasets` dependencies on my M1 (in order to start contributing to the project). However, I got an error regarding `tensorflow`.
On M1, `tensorflow-macos` needs to be installed instead. Can we add a conditional requirement, so that `tensorflow-macos` would be installed on M1?
## Steps to reproduce the bug
Fresh clone this project (on m1), create a virtualenv and run this:
```python
pip install -e ".[dev]"
```
## Expected results
Installation should be smooth, and all the dependencies should be installed on M1.
## Actual results
You should receive an error, saying pip couldn't find a version that matches this pattern:
```
tensorflow>=2.3,!=2.6.0,!=2.6.1
```
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 2.6.2.dev0
- Platform: macOS-12.6-arm64-arm-64bit
- Python version: 3.9.6
- PyArrow version: 7.0.0
- Pandas version: 1.5.0
|
closed
|
https://github.com/huggingface/datasets/issues/5118
| 2022-10-16T16:50:08
| 2022-10-19T09:10:08
| 2022-10-19T09:10:08
|
{
"login": "david1542",
"id": 9879252,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,409,571,346
| 5,117
|
Progress bars have color red and never completed to 100%
|
## Describe the bug
Progress bars after transformative operations turn in red and never be completed to 100%
## Steps to reproduce the bug
```python
from datasets import load_dataset
load_dataset('rotten_tomatoes', split='test').filter(lambda o: True)
```
## Expected results
Progress bar should be 100% and green
## Actual results
Progress bar turn in red and never completed to 100%
## Environment info
- `datasets` version: 2.6.1
- Platform: Linux-5.10.133+-x86_64-with-Ubuntu-18.04-bionic
- Python version: 3.7.14
- PyArrow version: 6.0.1
- Pandas version: 1.3.5
|
closed
|
https://github.com/huggingface/datasets/issues/5117
| 2022-10-14T16:12:30
| 2024-06-19T19:03:42
| 2022-10-23T12:58:41
|
{
"login": "echatzikyriakidis",
"id": 63857529,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,409,549,471
| 5,116
|
Use yaml for issue templates + revamp
|
Use YAML instead of markdown (more expressive) for the issue templates. In addition, update their structure/fields to be more aligned with Transformers.
PS: also removes the "add_dataset" PR template, as we no longer accept such PRs.
|
closed
|
https://github.com/huggingface/datasets/pull/5116
| 2022-10-14T15:53:13
| 2022-10-19T13:05:49
| 2022-10-19T13:03:22
|
{
"login": "mariosasko",
"id": 47462742,
"type": "User"
}
|
[] | true
|
[] |
1,409,250,020
| 5,115
|
Fix iter_batches
|
The `pa.Table.to_reader()` method available in `pyarrow>=8.0.0` may return chunks of size < `max_chunksize`, therefore `iter_batches` can return batches smaller than the `batch_size` specified by the user
Therefore batched `map` couldn't always use batches of the right size, e.g. this fails because it runs only on one batch of one element:
```python
from datasets import Dataset, concatenate_datasets
ds = concatenate_datasets([Dataset.from_dict({"a": [i]}) for i in range(10)])
ds2 = ds.map(lambda _: {}, batched=True)
assert list(ds2) == list(ds)
```
This was introduced in https://github.com/huggingface/datasets/pull/5030
Close https://github.com/huggingface/datasets/issues/5111
This will require a patch release along with https://github.com/huggingface/datasets/pull/5113
TODO:
- [x] fix tests
- [x] add more tests
|
closed
|
https://github.com/huggingface/datasets/pull/5115
| 2022-10-14T12:06:14
| 2022-10-14T15:02:15
| 2022-10-14T14:59:58
|
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true
|
[] |
1,409,236,738
| 5,114
|
load_from_disk with remote filesystem fails due to a wrong temporary local folder path
|
## Describe the bug
The function load_from_disk fails when using a remote filesystem because of a wrong temporary path generation in the load_from_disk method of arrow_dataset.py:
```python
if is_remote_filesystem(fs):
src_dataset_path = extract_path_from_uri(dataset_path)
dataset_path = Dataset._build_local_temp_path(src_dataset_path)
fs.download(src_dataset_path, dataset_path.as_posix(), recursive=True)
```
If _dataset_path_ is `gs://speech/mydataset/train`, then _src_dataset_path_ will be `speech/mydataset/train` and _dataset_path_ will be something like `/var/folders/9s/gf0b/T/tmp6t/speech/mydataset/train`
Then, after downloading the **folder** _src_dataset_path_, you will get a path like `/var/folders/9s/gf0b/T/tmp6t/speech/mydataset/train/train/state.json` (notice we have train twice)
Instead of downloading the remote folder we should be downloading all the files in the folder for the path to be right:
```python
fs.download(os.path.join(src_dataset_path,*), dataset_path.as_posix(), recursive=True)
```
## Steps to reproduce the bug
```python
fs = gcsfs.GCSFileSystem(**storage_options)
dataset = load_from_disk("common_voice_processed") # loading local dataset previously saved locally, works fine
dataset.save_to_disk(output_dir, fs=fs) #works fine
dataset = load_from_disk(output_dir, fs=fs) # crashes
```
## Expected results
The dataset is loaded
## Actual results
FileNotFoundError: [Errno 2] No such file or directory: '/var/folders/9s/gf0b9jz15d517yrf7m3nvlxr0000gn/T/tmp6t5e221_/speech/datasets/tests/common_voice_processed/train/state.json'
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: datasets-2.6.1.dev0
- Platform: mac os monterey 12.5.1
- Python version: 3.8.13
- PyArrow version:pyarrow==9.0.0
|
open
|
https://github.com/huggingface/datasets/issues/5114
| 2022-10-14T11:54:53
| 2022-11-19T07:13:10
| null |
{
"login": "bruno-hays",
"id": 48770768,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,409,207,607
| 5,113
|
Fix filter indices when batched
|
This PR fixes a bug introduced by:
- #5030
Fix #5112.
|
closed
|
https://github.com/huggingface/datasets/pull/5113
| 2022-10-14T11:30:03
| 2022-10-24T06:21:09
| 2022-10-14T12:11:44
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true
|
[] |
1,409,143,409
| 5,112
|
Bug with filtered indices
|
## Describe the bug
As reported by @PartiallyTyped (and by @Muennighoff):
- https://github.com/huggingface/datasets/issues/5111#issuecomment-1278652524
There is an issue with the indices of a filtered dataset.
## Steps to reproduce the bug
```python
ds = Dataset.from_dict({"num": [0, 1, 2, 3]})
ds = ds.filter(lambda num: num % 2 == 0, input_columns="num", batch_size=2)
assert all(item["num"] % 2 == 0 for item in ds)
```
## Expected results
The indices of the filtered dataset should correspond to the examples with "language" equals to "english".
## Actual results
Indices to items with other languages are included in the filtered dataset indices
## Preliminar investigation
It seems a bug introduced by:
- #5030
|
closed
|
https://github.com/huggingface/datasets/issues/5112
| 2022-10-14T10:35:47
| 2022-10-14T13:55:03
| 2022-10-14T12:11:45
|
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,408,143,170
| 5,111
|
map and filter not working properly in multiprocessing with the new release 2.6.0
|
## Describe the bug
When mapping is used on a dataset with more than one process, there is a weird behavior when trying to use `filter` , it's like only the samples from one worker are retrieved, one needs to specify the same `num_proc` in filter for it to work properly. This doesn't happen with `datasets` version 2.5.2
In the code below the data is filtered differently when we increase `num_proc` used in `map` although the datsets before and after mapping have identical elements.
## Steps to reproduce the bug
```python
import datasets
from datasets import load_dataset
def preprocess(example):
return example
ds = load_dataset("codeparrot/codeparrot-clean-valid", split="train").select([i for i in range(10)])
ds1 = ds.map(preprocess, num_proc=2)
ds2 = ds.map(preprocess)
# the datasets elements are the same
for i in range(len(ds1)):
assert ds1[i]==ds2[i]
print(f'Target column before filtering {ds1["autogenerated"]}')
print(f'Target column before filtering {ds2["autogenerated"]}')
print(f"datasets version {datasets.__version__}")
ds_filtered_1 = ds1.filter(lambda x: not x["autogenerated"])
ds_filtered_2 = ds2.filter(lambda x: not x["autogenerated"])
# all elements in Target column are false so they should all be kept, but for ds2 only the first 5=num_samples/num_proc are kept
print(ds_filtered_1)
print(ds_filtered_2)
```
```
Target column before filtering [False, False, False, False, False, False, False, False, False, False]
Target column before filtering [False, False, False, False, False, False, False, False, False, False]
Dataset({
features: ['repo_name', 'path', 'copies', 'size', 'content', 'license', 'hash', 'line_mean', 'line_max', 'alpha_frac', 'autogenerated'],
num_rows: 5
})
Dataset({
features: ['repo_name', 'path', 'copies', 'size', 'content', 'license', 'hash', 'line_mean', 'line_max', 'alpha_frac', 'autogenerated'],
num_rows: 10
})
```
## Expected results
Increasing `num_proc` in mapping shouldn't alter filtering. With the previous version 2.5.2 this doesn't happen
## Actual results
Filtering doesn't work properly when we increase `num_proc` in mapping but not when calling `filter`
## Environment info
<!-- You can run the command `datasets-cli env` and copy-and-paste its output below. -->
- `datasets` version: 2.6.0
- Platform: Linux-4.19.0-22-cloud-amd64-x86_64-with-glibc2.28
- Python version: 3.9.13
- PyArrow version: 8.0.0
- Pandas version: 1.4.2
|
closed
|
https://github.com/huggingface/datasets/issues/5111
| 2022-10-13T17:00:55
| 2022-10-17T08:26:59
| 2022-10-14T14:59:59
|
{
"login": "loubnabnl",
"id": 44069155,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,407,434,706
| 5,109
|
Map caching not working for some class methods
|
## Describe the bug
The cache loading is not working as expected for some class methods with a model stored in an attribute.
The new fingerprint for `_map_single` is not the same at each run. The hasher generate a different hash for the class method.
This comes from `dumps` function in `datasets.utils.py_utils` which generates a different dump at each run.
## Steps to reproduce the bug
```python
from datasets import load_dataset
from transformers import AutoConfig, AutoModel, AutoTokenizer
dataset = load_dataset("ethos", "binary")
BASE_MODELNAME = "sentence-transformers/all-MiniLM-L6-v2"
class Object:
def __init__(self):
config = AutoConfig.from_pretrained(BASE_MODELNAME)
self.bert = AutoModel.from_config(config=config, add_pooling_layer=False)
self.tok = AutoTokenizer.from_pretrained(BASE_MODELNAME)
def tokenize(self, examples):
tokenized_texts = self.tok(
examples["text"],
padding="max_length",
truncation=True,
max_length=256,
)
return tokenized_texts
instance = Object()
result = dict()
for phase in ["train"]:
result[phase] = dataset[phase].map(instance.tokenize, batched=True, load_from_cache_file=True, num_proc=2)
```
## Expected results
Load cache instead of recompute result.
## Actual results
Result recomputed from scratch at each run.
The cache works fine when deleting `bert` attribute.
## Environment info
- `datasets` version: 2.5.3.dev0
- Platform: macOS-10.16-x86_64-i386-64bit
- Python version: 3.9.13
- PyArrow version: 7.0.0
- Pandas version: 1.5.0
|
closed
|
https://github.com/huggingface/datasets/issues/5109
| 2022-10-13T09:12:58
| 2022-10-17T10:38:45
| 2022-10-17T10:38:45
|
{
"login": "Mouhanedg56",
"id": 23029765,
"type": "User"
}
|
[
{
"name": "bug",
"color": "d73a4a"
}
] | false
|
[] |
1,407,044,107
| 5,108
|
Fix a typo in arrow_dataset.py
| null |
closed
|
https://github.com/huggingface/datasets/pull/5108
| 2022-10-13T02:33:55
| 2022-10-14T09:47:28
| 2022-10-14T09:47:27
|
{
"login": "yangky11",
"id": 5431913,
"type": "User"
}
|
[] | true
|
[] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.