html_url stringlengths 48 51 | title stringlengths 1 290 | comments listlengths 0 30 | body stringlengths 0 228k ⌀ | number int64 2 7.08k |
|---|---|---|---|---|
https://github.com/huggingface/datasets/issues/2092 | How to disable making arrow tables in load_dataset ? | [
"Hi ! We plan to add streaming features in the future.\r\n\r\nThis should allow to load a dataset instantaneously without generating the arrow table. The trade-off is that accessing examples from a streaming dataset must be done in an iterative way, and with an additional (but hopefully minor) overhead.\r\nWhat do ... | Is there a way to disable the construction of arrow tables, or to make them on the fly as the dataset is being used ? | 2,092 |
https://github.com/huggingface/datasets/issues/2089 | Add documentaton for dataset README.md files | [
"Hi ! We are using the [datasets-tagging app](https://github.com/huggingface/datasets-tagging) to select the tags to add.\r\n\r\nWe are also adding the full list of tags in #2107 \r\nThis covers multilinguality, language_creators, licenses, size_categories and task_categories.\r\n\r\nIn general if you want to add a... | Hi,
the dataset README files have special headers.
Somehow a documenation of the allowed values and tags is missing.
Could you add that?
Just to give some concrete questions that should be answered imo:
- which values can be passted to multilinguality?
- what should be passed to language_creators?
- which values should licenses have? What do I say when it is a custom license? Should I add a link?
- how should I choose size_categories ? What are valid ranges?
- what are valid task_categories?
Thanks
Philip | 2,089 |
https://github.com/huggingface/datasets/issues/2084 | CUAD - Contract Understanding Atticus Dataset | [
"+1 on this request"
] | ## Adding a Dataset
- **Name:** CUAD - Contract Understanding Atticus Dataset
- **Description:** As one of the only large, specialized NLP benchmarks annotated by experts, CUAD can serve as a challenging research benchmark for the broader NLP community.
- **Paper:** https://arxiv.org/abs/2103.06268
- **Data:** https://github.com/TheAtticusProject/cuad/
- **Motivation:** good domain specific datasets are valuable
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
| 2,084 |
https://github.com/huggingface/datasets/issues/2083 | `concatenate_datasets` throws error when changing the order of datasets to concatenate | [
"Hi,\r\n\r\nthis bug is related to `Dataset.{remove_columns, rename_column, flatten}` not propagating the change to the schema metadata when the info features are updated, so this line is the culprit:\r\n```python\r\ncommon_voice_train = common_voice_train.remove_columns(['client_id', 'up_votes', 'down_votes', 'age... | Hey,
I played around with the `concatenate_datasets(...)` function: https://huggingface.co/docs/datasets/package_reference/main_classes.html?highlight=concatenate_datasets#datasets.concatenate_datasets
and noticed that when the order in which the datasets are concatenated changes an error is thrown where it should not IMO.
Here is a google colab to reproduce the error: https://colab.research.google.com/drive/17VTFU4KQ735-waWZJjeOHS6yDTfV5ekK?usp=sharing | 2,083 |
https://github.com/huggingface/datasets/issues/2080 | Multidimensional arrays in a Dataset | [
"Hi !\r\n\r\nThis is actually supported ! but not yet in `from_pandas`.\r\nYou can use `from_dict` for now instead:\r\n```python\r\nfrom datasets import Dataset, Array2D, Features, Value\r\nimport pandas as pd\r\nimport numpy as np\r\n\r\ndataset = {\r\n 'bbox': [\r\n np.array([[1,2,3,4],[1,2,3,4],[1,2,3,... | Hi,
I'm trying to put together a `datasets.Dataset` to be used with LayoutLM which is available in `transformers`. This model requires as input the bounding boxes of each of the token of a sequence. This is when I realized that `Dataset` does not support multi-dimensional arrays as a value for a column in a row.
The following code results in conversion error in pyarrow (`pyarrow.lib.ArrowInvalid: ('Can only convert 1-dimensional array values', 'Conversion failed for column bbox with type object')`)
```
from datasets import Dataset
import pandas as pd
import numpy as np
dataset = pd.DataFrame({
'bbox': [
np.array([[1,2,3,4],[1,2,3,4],[1,2,3,4]]),
np.array([[1,2,3,4],[1,2,3,4],[1,2,3,4]]),
np.array([[1,2,3,4],[1,2,3,4],[1,2,3,4]]),
np.array([[1,2,3,4],[1,2,3,4],[1,2,3,4]])
],
'input_ids': [1, 2, 3, 4]
})
dataset = Dataset.from_pandas(dataset)
```
Since I wanted to use pytorch for the downstream training task, I also tried a few ways to directly put in a column of 2-D pytorch tensor in a formatted dataset, but I can only have a list of 1-D tensors, or a list of arrays, or a list of lists.
```
import torch
from datasets import Dataset
import pandas as pd
dataset = pd.DataFrame({
'bbox': [
[[1,2,3,4],[1,2,3,4],[1,2,3,4]],
[[1,2,3,4],[1,2,3,4],[1,2,3,4]],
[[1,2,3,4],[1,2,3,4],[1,2,3,4]],
[[1,2,3,4],[1,2,3,4],[1,2,3,4]]
],
'input_ids': [1, 2, 3, 4]
})
dataset = Dataset.from_pandas(dataset)
def test(examples):
return {'bbbox': torch.Tensor(examples['bbox'])}
dataset = dataset.map(test)
print(dataset[0]['bbox'])
print(dataset[0]['bbbox'])
dataset.set_format(type='torch', columns=['input_ids', 'bbox'], output_all_columns=True)
print(dataset[0]['bbox'])
print(dataset[0]['bbbox'])
def test2(examples):
return {'bbbox': torch.stack(examples['bbox'])}
dataset = dataset.map(test2)
print(dataset[0]['bbox'])
print(dataset[0]['bbbox'])
```
Is is possible to support n-D arrays/tensors in datasets?
It seems that it can also be useful for this [feature request](https://github.com/huggingface/datasets/issues/263). | 2,080 |
https://github.com/huggingface/datasets/issues/2078 | MemoryError when computing WER metric | [
"Hi ! Thanks for reporting.\r\nWe're indeed using `jiwer` to compute the WER.\r\n\r\nMaybe instead of calling `jiwer.wer` once for all the preditions/references we can compute the WER iteratively to avoid memory issues ? I'm not too familial with `jiwer` but this must be possible.\r\n\r\nCurrently the code to compu... | Hi, I'm trying to follow the ASR example to try Wav2Vec. This is the code that I use for WER calculation:
```
wer = load_metric("wer")
print(wer.compute(predictions=result["predicted"], references=result["target"]))
```
However, I receive the following exception:
`Traceback (most recent call last):
File "/home/diego/IpGlobal/wav2vec/test_wav2vec.py", line 51, in <module>
print(wer.compute(predictions=result["predicted"], references=result["target"]))
File "/home/diego/miniconda3/envs/wav2vec3.6/lib/python3.6/site-packages/datasets/metric.py", line 403, in compute
output = self._compute(predictions=predictions, references=references, **kwargs)
File "/home/diego/.cache/huggingface/modules/datasets_modules/metrics/wer/73b2d32b723b7fb8f204d785c00980ae4d937f12a65466f8fdf78706e2951281/wer.py", line 94, in _compute
return wer(references, predictions)
File "/home/diego/miniconda3/envs/wav2vec3.6/lib/python3.6/site-packages/jiwer/measures.py", line 81, in wer
truth, hypothesis, truth_transform, hypothesis_transform, **kwargs
File "/home/diego/miniconda3/envs/wav2vec3.6/lib/python3.6/site-packages/jiwer/measures.py", line 192, in compute_measures
H, S, D, I = _get_operation_counts(truth, hypothesis)
File "/home/diego/miniconda3/envs/wav2vec3.6/lib/python3.6/site-packages/jiwer/measures.py", line 273, in _get_operation_counts
editops = Levenshtein.editops(source_string, destination_string)
MemoryError`
My system has more than 10GB of available RAM. Looking at the code, I think that it could be related to the way jiwer does the calculation, as it is pasting all the sentences in a single string before calling Levenshtein editops function.
| 2,078 |
https://github.com/huggingface/datasets/issues/2076 | Issue: Dataset download error | [
"Hi @XuhuiZhou, thanks for reporting this issue. \r\n\r\nIndeed, the old links are no longer valid (404 Not Found error), and the script must be updated with the new links to Google Drive.",
"It would be nice to update the urls indeed !\r\n\r\nTo do this, you just need to replace the urls in `iwslt2017.py` and th... | The download link in `iwslt2017.py` file does not seem to work anymore.
For example, `FileNotFoundError: Couldn't find file at https://wit3.fbk.eu/archive/2017-01-trnted/texts/zh/en/zh-en.tgz`
Would be nice if we could modify it script and use the new downloadable link? | 2,076 |
https://github.com/huggingface/datasets/issues/2075 | ConnectionError: Couldn't reach common_voice.py | [
"Hi @LifaSun, thanks for reporting this issue.\r\n\r\nSometimes, GitHub has some connectivity problems. Could you confirm that the problem persists?",
"@albertvillanova Thanks! It works well now. "
] | When I run:
from datasets import load_dataset, load_metric
common_voice_train = load_dataset("common_voice", "zh-CN", split="train+validation")
common_voice_test = load_dataset("common_voice", "zh-CN", split="test")
Got:
ConnectionError: Couldn't reach https://raw.githubusercontent.com/huggingface/datasets/master/datasets/common_voice/common_voice.py
Version:
1.4.1
Thanks! @lhoestq @LysandreJik @thomwolf | 2,075 |
https://github.com/huggingface/datasets/issues/2071 | Multiprocessing is slower than single process | [
"dupe of #1992"
] | ```python
# benchmark_filter.py
import logging
import sys
import time
from datasets import load_dataset, set_caching_enabled
if __name__ == "__main__":
set_caching_enabled(False)
logging.basicConfig(level=logging.DEBUG)
bc = load_dataset("bookcorpus")
now = time.time()
try:
bc["train"].filter(lambda x: len(x["text"]) < 64, num_proc=int(sys.argv[1]))
except Exception as e:
print(f"cancelled: {e}")
elapsed = time.time() - now
print(elapsed)
```
Running `python benchmark_filter.py 1` (20min+) is faster than `python benchmark_filter.py 2` (2hrs+) | 2,071 |
https://github.com/huggingface/datasets/issues/2070 | ArrowInvalid issue for squad v2 dataset | [
"Hi ! This error happens when you use `map` in batched mode and then your function doesn't return the same number of values per column.\r\n\r\nIndeed since you're using `map` in batched mode, `prepare_validation_features` must take a batch as input (i.e. a dictionary of multiple rows of the dataset), and return a b... | Hello, I am using the huggingface official question answering example notebook (https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/question_answering.ipynb).
In the prepare_validation_features function, I made some modifications to tokenize a new set of quesions with the original contexts and save them in three different list called candidate_input_dis, candidate_attetion_mask and candidate_token_type_ids. When I try to run the next cell for dataset.map, I got the following error:
`ArrowInvalid: Column 1 named candidate_attention_mask expected length 1180 but got length 1178`
My code is as follows:
```
def generate_candidate_questions(examples):
val_questions = examples["question"]
candididate_questions = random.sample(datasets["train"]["question"], len(val_questions))
candididate_questions = [x[:max_length] for x in candididate_questions]
return candididate_questions
def prepare_validation_features(examples, use_mixing=False):
pad_on_right = tokenizer.padding_side == "right"
tokenized_examples = tokenizer(
examples["question" if pad_on_right else "context"],
examples["context" if pad_on_right else "question"],
truncation="only_second" if pad_on_right else "only_first",
max_length=max_length,
stride=doc_stride,
return_overflowing_tokens=True,
return_offsets_mapping=True,
padding="max_length",
)
if use_mixing:
candidate_questions = generate_candidate_questions(examples)
tokenized_candidates = tokenizer(
candidate_questions if pad_on_right else examples["context"],
examples["context"] if pad_on_right else candidate_questions,
truncation="only_second" if pad_on_right else "only_first",
max_length=max_length,
stride=doc_stride,
return_overflowing_tokens=True,
return_offsets_mapping=True,
padding="max_length",
)
sample_mapping = tokenized_examples.pop("overflow_to_sample_mapping")
tokenized_examples["example_id"] = []
if use_mixing:
tokenized_examples["candidate_input_ids"] = tokenized_candidates["input_ids"]
tokenized_examples["candidate_attention_mask"] = tokenized_candidates["attention_mask"]
tokenized_examples["candidate_token_type_ids"] = tokenized_candidates["token_type_ids"]
for i in range(len(tokenized_examples["input_ids"])):
sequence_ids = tokenized_examples.sequence_ids(i)
context_index = 1 if pad_on_right else 0
sample_index = sample_mapping[i]
tokenized_examples["example_id"].append(examples["id"][sample_index])
tokenized_examples["offset_mapping"][i] = [
(o if sequence_ids[k] == context_index else None)
for k, o in enumerate(tokenized_examples["offset_mapping"][i])
]
return tokenized_examples
validation_features = datasets["validation"].map(
lambda xs: prepare_validation_features(xs, True),
batched=True,
remove_columns=datasets["validation"].column_names
)
```
I guess this might happen because of the batched=True. I see similar issues in this repo related to arrow table length mismatch error, but in their cases, the numbers vary a lot. In my case, this error always happens when the expected length and unexpected length are very close. Thanks for the help! | 2,070 |
https://github.com/huggingface/datasets/issues/2068 | PyTorch not available error on SageMaker GPU docker though it is installed | [
"cc @philschmid ",
"Hey @sivakhno,\r\n\r\nhow does your `requirements.txt` look like to install the `datasets` library and which version of it are you running? Can you try to install `datasets>=1.4.0`",
"Hi @philschmid - thanks for suggestion. I am using `datasets==1.4.1`. \r\nI have also tried using `torch=1.6... | I get en error when running data loading using SageMaker SDK
```
File "main.py", line 34, in <module>
run_training()
File "main.py", line 25, in run_training
dm.setup('fit')
File "/opt/conda/lib/python3.6/site-packages/pytorch_lightning/core/datamodule.py", line 92, in wrapped_fn
return fn(*args, **kwargs)
File "/opt/ml/code/data_module.py", line 103, in setup
self.dataset[split].set_format(type="torch", columns=self.columns)
File "/opt/conda/lib/python3.6/site-packages/datasets/fingerprint.py", line 337, in wrapper
out = func(self, *args, **kwargs)
File "/opt/conda/lib/python3.6/site-packages/datasets/arrow_dataset.py", line 995, in set_format
_ = get_formatter(type, **format_kwargs)
File "/opt/conda/lib/python3.6/site-packages/datasets/formatting/__init__.py", line 114, in get_formatter
raise _FORMAT_TYPES_ALIASES_UNAVAILABLE[format_type]
ValueError: PyTorch needs to be installed to be able to return PyTorch tensors.
```
when trying to execute dataset loading using this notebook https://github.com/PyTorchLightning/pytorch-lightning/blob/master/notebooks/04-transformers-text-classification.ipynb, specifically lines
```
self.columns = [c for c in self.dataset[split].column_names if c in self.loader_columns]
self.dataset[split].set_format(type="torch", columns=self.columns)
```
The SageMaker docker image used is 763104351884.dkr.ecr.eu-central-1.amazonaws.com/pytorch-training:1.4.0-gpu-py3 .
By running container interactively I have checked that torch loading completes successfully by executing `https://github.com/huggingface/datasets/blob/master/src/datasets/config.py#L39`.
Also as a first line in the data loading module I have
```
import os
os.environ["USE_TF"] = "0"
os.environ["USE_TORCH"] = "1"
````
But unfortunately the error stills persists. Any suggestions would be appreciated as I am stack.
Many Thanks!
| 2,068 |
https://github.com/huggingface/datasets/issues/2067 | Multiprocessing windows error | [
"Hi ! Thanks for reporting.\r\nThis looks like a bug, could you try to provide a minimal code example that reproduces the issue ? This would be very helpful !\r\n\r\nOtherwise I can try to run the wav2vec2 code above on my side but probably not this week..",
"```\r\nfrom datasets import load_dataset\r\n\r\ndatase... | As described here https://huggingface.co/blog/fine-tune-xlsr-wav2vec2
When using the num_proc argument on windows the whole Python environment crashes and hanging in loop.
For example at the map_to_array part.
An error occures because the cache file already exists and windows throws and error. After this the log crashes into an loop | 2,067 |
https://github.com/huggingface/datasets/issues/2065 | Only user permission of saved cache files, not group | [
"Hi ! Thanks for reporting.\r\n\r\nCurrently there's no way to specify this.\r\n\r\nWhen loading/processing a dataset, the arrow file is written using a temporary file. Then once writing is finished, it's moved to the cache directory (using `shutil.move` [here](https://github.com/huggingface/datasets/blob/f6b8251eb... | Hello,
It seems when a cached file is saved from calling `dataset.map` for preprocessing, it gets the user permissions and none of the user's group permissions. As we share data files across members of our team, this is causing a bit of an issue as we have to continually reset the permission of the files. Do you know any ways around this or a way to correctly set the permissions? | 2,065 |
https://github.com/huggingface/datasets/issues/2061 | Cannot load udpos subsets from xtreme dataset using load_dataset() | [
"@lhoestq Adding \"_\" to the class labels in the dataset script will fix the issue.\r\n\r\nThe bigger issue IMO is that the data files are in conll format, but the examples are tokens, not sentences.",
"Hi ! Thanks for reporting @adzcodez \r\n\r\n\r\n> @lhoestq Adding \"_\" to the class labels in the dataset scr... | Hello,
I am trying to load the udpos English subset from xtreme dataset, but this faces an error during loading. I am using datasets v1.4.1, pip install. I have tried with other udpos languages which also fail, though loading a different subset altogether (such as XNLI) has no issue. I have also tried on Colab and faced the same error.
Reprex is:
`from datasets import load_dataset `
`dataset = load_dataset('xtreme', 'udpos.English')`
The error is:
`KeyError: '_'`
The full traceback is:
KeyError Traceback (most recent call last)
<ipython-input-5-7181359ea09d> in <module>
1 from datasets import load_dataset
----> 2 dataset = load_dataset('xtreme', 'udpos.English')
~\Anaconda3\envs\mlenv\lib\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, script_version, use_auth_token, **config_kwargs)
738
739 # Download and prepare data
--> 740 builder_instance.download_and_prepare(
741 download_config=download_config,
742 download_mode=download_mode,
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, try_from_hf_gcs, dl_manager, base_path, use_auth_token, **download_and_prepare_kwargs)
576 logger.warning("HF google storage unreachable. Downloading and preparing it from source")
577 if not downloaded_from_gcs:
--> 578 self._download_and_prepare(
579 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
580 )
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
654 try:
655 # Prepare split will record examples associated to the split
--> 656 self._prepare_split(split_generator, **prepare_split_kwargs)
657 except OSError as e:
658 raise OSError(
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\builder.py in _prepare_split(self, split_generator)
977 generator, unit=" examples", total=split_info.num_examples, leave=False, disable=not_verbose
978 ):
--> 979 example = self.info.features.encode_example(record)
980 writer.write(example)
981 finally:
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\features.py in encode_example(self, example)
946 def encode_example(self, example):
947 example = cast_to_python_objects(example)
--> 948 return encode_nested_example(self, example)
949
950 def encode_batch(self, batch):
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\features.py in encode_nested_example(schema, obj)
840 # Nested structures: we allow dict, list/tuples, sequences
841 if isinstance(schema, dict):
--> 842 return {
843 k: encode_nested_example(sub_schema, sub_obj) for k, (sub_schema, sub_obj) in utils.zip_dict(schema, obj)
844 }
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\features.py in <dictcomp>(.0)
841 if isinstance(schema, dict):
842 return {
--> 843 k: encode_nested_example(sub_schema, sub_obj) for k, (sub_schema, sub_obj) in utils.zip_dict(schema, obj)
844 }
845 elif isinstance(schema, (list, tuple)):
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\features.py in encode_nested_example(schema, obj)
868 # ClassLabel will convert from string to int, TranslationVariableLanguages does some checks
869 elif isinstance(schema, (ClassLabel, TranslationVariableLanguages, Value, _ArrayXD)):
--> 870 return schema.encode_example(obj)
871 # Other object should be directly convertible to a native Arrow type (like Translation and Translation)
872 return obj
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\features.py in encode_example(self, example_data)
647 # If a string is given, convert to associated integer
648 if isinstance(example_data, str):
--> 649 example_data = self.str2int(example_data)
650
651 # Allowing -1 to mean no label.
~\Anaconda3\envs\mlenv\lib\site-packages\datasets\features.py in str2int(self, values)
605 if value not in self._str2int:
606 value = value.strip()
--> 607 output.append(self._str2int[str(value)])
608 else:
609 # No names provided, try to integerize
KeyError: '_'
| 2,061 |
https://github.com/huggingface/datasets/issues/2059 | Error while following docs to load the `ted_talks_iwslt` dataset | [
"@skyprince999 as you authored the PR for this dataset, any comments?",
"This has been fixed in #2064 by @mariosasko (thanks again !)\r\n\r\nThe fix is available on the master branch and we'll do a new release very soon :)"
] | I am currently trying to load the `ted_talks_iwslt` dataset into google colab.
The [docs](https://huggingface.co/datasets/ted_talks_iwslt) mention the following way of doing so.
```python
dataset = load_dataset("ted_talks_iwslt", language_pair=("it", "pl"), year="2014")
```
Executing it results in the error attached below.
```
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-6-7dcc67154ef9> in <module>()
----> 1 dataset = load_dataset("ted_talks_iwslt", language_pair=("it", "pl"), year="2014")
4 frames
/usr/local/lib/python3.7/dist-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, script_version, use_auth_token, **config_kwargs)
730 hash=hash,
731 features=features,
--> 732 **config_kwargs,
733 )
734
/usr/local/lib/python3.7/dist-packages/datasets/builder.py in __init__(self, writer_batch_size, *args, **kwargs)
927
928 def __init__(self, *args, writer_batch_size=None, **kwargs):
--> 929 super(GeneratorBasedBuilder, self).__init__(*args, **kwargs)
930 # Batch size used by the ArrowWriter
931 # It defines the number of samples that are kept in memory before writing them
/usr/local/lib/python3.7/dist-packages/datasets/builder.py in __init__(self, cache_dir, name, hash, features, **config_kwargs)
241 name,
242 custom_features=features,
--> 243 **config_kwargs,
244 )
245
/usr/local/lib/python3.7/dist-packages/datasets/builder.py in _create_builder_config(self, name, custom_features, **config_kwargs)
337 if "version" not in config_kwargs and hasattr(self, "VERSION") and self.VERSION:
338 config_kwargs["version"] = self.VERSION
--> 339 builder_config = self.BUILDER_CONFIG_CLASS(**config_kwargs)
340
341 # otherwise use the config_kwargs to overwrite the attributes
/root/.cache/huggingface/modules/datasets_modules/datasets/ted_talks_iwslt/024d06b1376b361e59245c5878ab8acf9a7576d765f2d0077f61751158e60914/ted_talks_iwslt.py in __init__(self, language_pair, year, **kwargs)
219 description=description,
220 version=datasets.Version("1.1.0", ""),
--> 221 **kwargs,
222 )
223
TypeError: __init__() got multiple values for keyword argument 'version'
```
How to resolve this?
PS: Thanks a lot @huggingface team for creating this great library! | 2,059 |
https://github.com/huggingface/datasets/issues/2058 | Is it possible to convert a `tfds` to HuggingFace `dataset`? | [
"Hi! You can either save the TF dataset to one of the formats supported by datasets (`parquet`, `csv`, `json`, ...) or pass a generator function to `Dataset.from_generator` that yields its examples."
] | I was having some weird bugs with `C4`dataset version of HuggingFace, so I decided to try to download `C4`from `tfds`. I would like to know if it is possible to convert a tfds dataset to HuggingFace dataset format :)
I can also open a new issue reporting the bug I'm receiving with `datasets.load_dataset('c4','en')` in the future if you think that it would be useful.
Thanks!
| 2,058 |
https://github.com/huggingface/datasets/issues/2056 | issue with opus100/en-fr dataset | [
"@lhoestq I also deleted the cache and redownload the file and still the same issue, I appreciate any help on this. thanks ",
"Here please find the minimal code to reproduce the issue @lhoestq note this only happens with MT5TokenizerFast\r\n\r\n```\r\nfrom datasets import load_dataset\r\nfrom transformers impor... | Hi
I am running run_mlm.py code of huggingface repo with opus100/fr-en pair, I am getting this error, note that this error occurs for only this pairs and not the other pairs. Any idea why this is occurring? and how I can solve this?
Thanks a lot @lhoestq for your help in advance.
`
thread '<unnamed>' panicked at 'index out of bounds: the len is 617 but the index is 617', /__w/tokenizers/tokenizers/tokenizers/src/tokenizer/normalizer.rs:382:21
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
63%|██████████████████████████████████████████████████████████▊ | 626/1000 [00:27<00:16, 22.69ba/s]
Traceback (most recent call last):
File "run_mlm.py", line 550, in <module>
main()
File "run_mlm.py", line 412, in main
in zip(data_args.dataset_name, data_args.dataset_config_name)]
File "run_mlm.py", line 411, in <listcomp>
logger) for dataset_name, dataset_config_name\
File "/user/dara/dev/codes/seq2seq/data/tokenize_datasets.py", line 96, in get_tokenized_dataset
load_from_cache_file=not data_args.overwrite_cache,
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/dataset_dict.py", line 448, in map
for k, dataset in self.items()
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/dataset_dict.py", line 448, in <dictcomp>
for k, dataset in self.items()
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/arrow_dataset.py", line 1309, in map
update_data=update_data,
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/arrow_dataset.py", line 204, in wrapper
out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/fingerprint.py", line 337, in wrapper
out = func(self, *args, **kwargs)
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/arrow_dataset.py", line 1574, in _map_single
batch, indices, check_same_num_examples=len(self.list_indexes()) > 0, offset=offset
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/arrow_dataset.py", line 1490, in apply_function_on_filtered_inputs
function(*fn_args, effective_indices, **fn_kwargs) if with_indices else function(*fn_args, **fn_kwargs)
File "/user/dara/dev/codes/seq2seq/data/tokenize_datasets.py", line 89, in tokenize_function
return tokenizer(examples[text_column_name], return_special_tokens_mask=True)
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/transformers/tokenization_utils_base.py", line 2347, in __call__
**kwargs,
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/transformers/tokenization_utils_base.py", line 2532, in batch_encode_plus
**kwargs,
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/transformers/tokenization_utils_fast.py", line 384, in _batch_encode_plus
is_pretokenized=is_split_into_words,
pyo3_runtime.PanicException: index out of bounds: the len is 617 but the index is 617
` | 2,056 |
https://github.com/huggingface/datasets/issues/2055 | is there a way to override a dataset object saved with save_to_disk? | [
"Hi\r\nYou can rename the arrow file and update the name in `state.json`",
"I tried this way, but when there is a mapping process to the dataset, it again uses a random cache name. atm, I am trying to use the following method by setting an exact cache file,\r\n\r\n```\r\n dataset_with_embedding =csv_da... | At the moment when I use save_to_disk, it uses the arbitrary name for the arrow file. Is there a way to override such an object? | 2,055 |
https://github.com/huggingface/datasets/issues/2054 | Could not find file for ZEST dataset | [
"The zest dataset url was changed (allenai/zest#3) and #2057 should resolve this.",
"This has been fixed in #2057 by @matt-peters (thanks again !)\r\n\r\nThe fix is available on the master branch and we'll do a new release very soon :)",
"Thanks @lhoestq and @matt-peters ",
"I am closing this issue since its ... | I am trying to use zest dataset from Allen AI using below code in colab,
```
!pip install -q datasets
from datasets import load_dataset
dataset = load_dataset("zest")
```
I am getting the following error,
```
Using custom data configuration default
Downloading and preparing dataset zest/default (download: 5.53 MiB, generated: 19.96 MiB, post-processed: Unknown size, total: 25.48 MiB) to /root/.cache/huggingface/datasets/zest/default/0.0.0/1f7a230fbfc964d979bbca0f0130fbab3259fce547ee758ad8aa4f9c9bec6cca...
---------------------------------------------------------------------------
FileNotFoundError Traceback (most recent call last)
<ipython-input-6-18dbbc1a4b8a> in <module>()
1 from datasets import load_dataset
2
----> 3 dataset = load_dataset("zest")
9 frames
/usr/local/lib/python3.7/dist-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)
612 )
613 elif response is not None and response.status_code == 404:
--> 614 raise FileNotFoundError("Couldn't find file at {}".format(url))
615 _raise_if_offline_mode_is_enabled(f"Tried to reach {url}")
616 raise ConnectionError("Couldn't reach {}".format(url))
FileNotFoundError: Couldn't find file at https://ai2-datasets.s3-us-west-2.amazonaws.com/zest/zest.zip
``` | 2,054 |
https://github.com/huggingface/datasets/issues/2052 | Timit_asr dataset repeats examples | [
"Hi,\r\n\r\nthis was fixed by #1995, so you can wait for the next release or install the package directly from the master branch with the following command: \r\n```bash\r\npip install git+https://github.com/huggingface/datasets\r\n```",
"Ty!"
] | Summary
When loading timit_asr dataset on datasets 1.4+, every row in the dataset is the same
Steps to reproduce
As an example, on this code there is the text from the training part:
Code snippet:
```
from datasets import load_dataset, load_metric
timit = load_dataset("timit_asr")
timit['train']['text']
#['Would such an act of refusal be useful?',
# 'Would such an act of refusal be useful?',
# 'Would such an act of refusal be useful?',
# 'Would such an act of refusal be useful?',
# 'Would such an act of refusal be useful?',
# 'Would such an act of refusal be useful?',
```
The same behavior happens for other columns
Expected behavior:
Different info on the actual timit_asr dataset
Actual behavior:
When loading timit_asr dataset on datasets 1.4+, every row in the dataset is the same. I've checked datasets 1.3 and the rows are different
Debug info
Streamlit version: (get it with $ streamlit version)
Python version: Python 3.6.12
Using Conda? PipEnv? PyEnv? Pex? Using pip
OS version: Centos-release-7-9.2009.1.el7.centos.x86_64
Additional information
You can check the same behavior on https://huggingface.co/datasets/viewer/?dataset=timit_asr | 2,052 |
https://github.com/huggingface/datasets/issues/2050 | Build custom dataset to fine-tune Wav2Vec2 | [
"@lhoestq - We could simply use the \"general\" json dataset for this no? ",
"Sure you can use the json loader\r\n```python\r\ndata_files = {\"train\": \"path/to/your/train_data.json\", \"test\": \"path/to/your/test_data.json\"}\r\ntrain_dataset = load_dataset(\"json\", data_files=data_files, split=\"train\")\r\n... | Thank you for your recent tutorial on how to finetune Wav2Vec2 on a custom dataset. The example you gave here (https://huggingface.co/blog/fine-tune-xlsr-wav2vec2) was on the CommonVoice dataset. However, what if I want to load my own dataset? I have a manifest (transcript and their audio files) in a JSON file.
| 2,050 |
https://github.com/huggingface/datasets/issues/2048 | github is not always available - probably need a back up | [] | Yesterday morning github wasn't working:
```
:/tmp$ wget https://raw.githubusercontent.com/huggingface/datasets/1.4.1/metrics/sacrebleu/sacrebleu.py--2021-03-12 18:35:59-- https://raw.githubusercontent.com/huggingface/datasets/1.4.1/metrics/sacrebleu/sacrebleu.py
Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.111.133, 185.199.109.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.
HTTP request sent, awaiting response... 500 Internal Server Error
2021-03-12 18:36:11 ERROR 500: Internal Server Error.
```
Suggestion: have a failover system and replicate the data on another system and reach there if gh isn't reachable? perhaps gh can be a master and the replicate a slave - so there is only one true source. | 2,048 |
https://github.com/huggingface/datasets/issues/2046 | add_faisis_index gets very slow when doing it interatively | [
"I think faiss automatically sets the number of threads to use to build the index.\r\nCan you check how many CPU cores are being used when you build the index in `use_own_knowleldge_dataset` as compared to this script ? Are there other programs running (maybe for rank>0) ?",
"Hi,\r\n I am running the add_faiss_in... | As the below code suggests, I want to run add_faisis_index in every nth interaction from the training loop. I have 7.2 million documents. Usually, it takes 2.5 hours (if I run an as a separate process similar to the script given in rag/use_own_knowleldge_dataset.py). Now, this takes usually 5hrs. Is this normal? Any way to make this process faster?
@lhoestq
```
def training_step(self, batch, batch_idx) -> Dict:
if (not batch_idx==0) and (batch_idx%5==0):
print("******************************************************")
ctx_encoder=self.trainer.model.module.module.model.rag.ctx_encoder
model_copy =type(ctx_encoder)(self.config_dpr) # get a new instance #this will be load in the CPU
model_copy.load_state_dict(ctx_encoder.state_dict()) # copy weights and stuff
list_of_gpus = ['cuda:2','cuda:3']
c_dir='/custom/cache/dir'
kb_dataset = load_dataset("csv", data_files=[self.custom_config.csv_path], split="train", delimiter="\t", column_names=["title", "text"],cache_dir=c_dir)
print(kb_dataset)
n=len(list_of_gpus) #nunber of dedicated GPUs
kb_list=[kb_dataset.shard(n, i, contiguous=True) for i in range(n)]
#kb_dataset.save_to_disk('/hpc/gsir059/MY-Test/RAY/transformers/examples/research_projects/rag/haha-dir')
print(self.trainer.global_rank)
dataset_shards = self.re_encode_kb(model_copy.to(device=list_of_gpus[self.trainer.global_rank]),kb_list[self.trainer.global_rank])
output = [None for _ in list_of_gpus]
#self.trainer.accelerator_connector.accelerator.barrier("embedding_process")
dist.all_gather_object(output, dataset_shards)
#This creation and re-initlaization of the new index
if (self.trainer.global_rank==0): #saving will be done in the main process
combined_dataset = concatenate_datasets(output)
passages_path =self.config.passages_path
logger.info("saving the dataset with ")
#combined_dataset.save_to_disk('/hpc/gsir059/MY-Test/RAY/transformers/examples/research_projects/rag/MY-Passage')
combined_dataset.save_to_disk(passages_path)
logger.info("Add faiss index to the dataset that consist of embeddings")
embedding_dataset=combined_dataset
index = faiss.IndexHNSWFlat(768, 128, faiss.METRIC_INNER_PRODUCT)
embedding_dataset.add_faiss_index("embeddings", custom_index=index)
embedding_dataset.get_index("embeddings").save(self.config.index_path)
| 2,046 |
https://github.com/huggingface/datasets/issues/2040 | ValueError: datasets' indices [1] come from memory and datasets' indices [0] come from disk | [
"Hi ! To help me understand the situation, can you print the values of `load_from_disk(PATH_DATA_CLS_A)['train']._indices_data_files` and `load_from_disk(PATH_DATA_CLS_B)['train']._indices_data_files` ?\r\nThey should both have a path to an arrow file\r\n\r\nAlso note that from #2025 concatenating datasets will no... | Hi there,
I am trying to concat two datasets that I've previously saved to disk via `save_to_disk()` like so (note that both are saved as `DataDict`, `PATH_DATA_CLS_*` are `Path`-objects):
```python
concatenate_datasets([load_from_disk(PATH_DATA_CLS_A)['train'], load_from_disk(PATH_DATA_CLS_B)['train']])
```
Yielding the following error:
```python
ValueError: Datasets' indices should ALL come from memory, or should ALL come from disk.
However datasets' indices [1] come from memory and datasets' indices [0] come from disk.
```
Been trying to solve this for quite some time now. Both `DataDict` have been created by reading in a `csv` via `load_dataset` and subsequently processed using the various `datasets` methods (i.e. filter, map, remove col, rename col). Can't figure out tho...
`load_from_disk(PATH_DATA_CLS_A)['train']` yields:
```python
Dataset({
features: ['labels', 'text'],
num_rows: 785
})
```
`load_from_disk(PATH_DATA_CLS_B)['train']` yields:
```python
Dataset({
features: ['labels', 'text'],
num_rows: 3341
})
``` | 2,040 |
https://github.com/huggingface/datasets/issues/2038 | outdated dataset_infos.json might fail verifications | [
"Hi ! Thanks for reporting.\r\n\r\nTo update the dataset_infos.json you can run:\r\n```\r\ndatasets-cli test ./datasets/doc2dial --all_configs --save_infos --ignore_verifications\r\n```",
"Fixed by #2041, thanks again @songfeng !"
] | The [doc2dial/dataset_infos.json](https://github.com/huggingface/datasets/blob/master/datasets/doc2dial/dataset_infos.json) is outdated. It would fail data_loader when verifying download checksum etc..
Could you please update this file or point me how to update this file?
Thank you. | 2,038 |
https://github.com/huggingface/datasets/issues/2036 | Cannot load wikitext | [
"Solved!"
] | when I execute these codes
```
>>> from datasets import load_dataset
>>> test_dataset = load_dataset("wikitext")
```
I got an error,any help?
```
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/xxx/anaconda3/envs/transformer/lib/python3.7/site-packages/datasets/load.py", line 589, in load_dataset
path, script_version=script_version, download_config=download_config, download_mode=download_mode, dataset=True
File "/home/xxx/anaconda3/envs/transformer/lib/python3.7/site-packages/datasets/load.py", line 267, in prepare_module
local_path = cached_path(file_path, download_config=download_config)
File "/home/xxx/anaconda3/envs/transformer/lib/python3.7/site-packages/datasets/utils/file_utils.py", line 308, in cached_path
use_etag=download_config.use_etag,
File "/home/xxx/anaconda3/envs/transformer/lib/python3.7/site-packages/datasets/utils/file_utils.py", line 487, in get_from_cache
raise ConnectionError("Couldn't reach {}".format(url))
ConnectionError: Couldn't reach https://raw.githubusercontent.com/huggingface/datasets/1.1.3/datasets/wikitext/wikitext.py
``` | 2,036 |
https://github.com/huggingface/datasets/issues/2035 | wiki40b/wikipedia for almost all languages cannot be downloaded | [
"Dear @lhoestq for wikipedia dataset I also get the same error, I greatly appreciate if you could have a look into this dataset as well. Below please find the command to reproduce the error:\r\n\r\n```\r\ndataset = load_dataset(\"wikipedia\", \"20200501.bg\")\r\nprint(dataset)\r\n```\r\n\r\nYour library is my only ... | Hi
I am trying to download the data as below:
```
from datasets import load_dataset
dataset = load_dataset("wiki40b", "cs")
print(dataset)
```
I am getting this error. @lhoestq I will be grateful if you could assist me with this error. For almost all languages except english I am getting this error.
I really need majority of languages in this dataset to be able to train my models for a deadline and your great scalable super well-written library is my only hope to train the models at scale while being low on resources.
thank you very much.
```
(fast) dara@vgne046:/user/dara/dev/codes/seq2seq$ python test_data.py
Downloading and preparing dataset wiki40b/cs (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to temp/dara/cache_home_2/datasets/wiki40b/cs/1.1.0/063778187363ffb294896eaa010fc254b42b73e31117c71573a953b0b0bf010f...
Traceback (most recent call last):
File "test_data.py", line 3, in <module>
dataset = load_dataset("wiki40b", "cs")
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/load.py", line 746, in load_dataset
use_auth_token=use_auth_token,
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/builder.py", line 579, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/datasets/builder.py", line 1105, in _download_and_prepare
import apache_beam as beam
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/apache_beam-2.28.0-py3.7-linux-x86_64.egg/apache_beam/__init__.py", line 96, in <module>
from apache_beam import io
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/apache_beam-2.28.0-py3.7-linux-x86_64.egg/apache_beam/io/__init__.py", line 23, in <module>
from apache_beam.io.avroio import *
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/apache_beam-2.28.0-py3.7-linux-x86_64.egg/apache_beam/io/avroio.py", line 55, in <module>
import avro
File "<frozen importlib._bootstrap>", line 983, in _find_and_load
File "<frozen importlib._bootstrap>", line 967, in _find_and_load_unlocked
File "<frozen importlib._bootstrap>", line 668, in _load_unlocked
File "<frozen importlib._bootstrap>", line 638, in _load_backward_compatible
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/avro_python3-1.9.2.1-py3.7.egg/avro/__init__.py", line 34, in <module>
File "/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/avro_python3-1.9.2.1-py3.7.egg/avro/__init__.py", line 30, in LoadResource
NotADirectoryError: [Errno 20] Not a directory: '/user/dara/libs/anaconda3/envs/fast/lib/python3.7/site-packages/avro_python3-1.9.2.1-py3.7.egg/avro/VERSION.txt'
``` | 2,035 |
https://github.com/huggingface/datasets/issues/2032 | Use Arrow filtering instead of writing a new arrow file for Dataset.filter | [
"Actually table.filter returns a new table in memory, which can fill users RAM.\r\n\r\nTherefore it's not a good solution if we want to keep supporting bigger than RAM datastes"
] | Currently the filter method reads the dataset batch by batch to write a new, filtered, arrow file on disk. Therefore all the reading + writing can take some time.
Using a mask directly on the arrow table doesn't do any read or write operation therefore it's significantly quicker.
I think there are two cases:
- if the dataset doesn't have an indices mapping, then one can simply use the arrow filtering on the main arrow table `dataset._data.filter(...)`
- if the dataset an indices mapping, then the mask should be applied on the indices mapping table `dataset._indices.filter(...)`
The indices mapping is used to map between the idx at `dataset[idx]` in `__getitem__` and the idx in the actual arrow table.
The new filter method should therefore be faster, and allow users to pass either a filtering function (that returns a boolean given an example), or directly a mask.
Feel free to discuss this idea in this thread :)
One additional note: the refactor at #2025 would make all the pickle-related stuff work directly with the arrow filtering, so that we only need to change the Dataset.filter method without having to deal with pickle.
cc @theo-m @gchhablani
related issues: #1796 #1949 | 2,032 |
https://github.com/huggingface/datasets/issues/2031 | wikipedia.py generator that extracts XML doesn't release memory | [
"Hi @miyamonz \r\nThanks for investigating this issue, good job !\r\nIt would be awesome to integrate your fix in the library, could you open a pull request ?",
"OK! I'll send it later."
] | I tried downloading Japanese wikipedia, but it always failed because of out of memory maybe.
I found that the generator function that extracts XML data in wikipedia.py doesn't release memory in the loop.
https://github.com/huggingface/datasets/blob/13a5b7db992ad5cf77895e4c0f76595314390418/datasets/wikipedia/wikipedia.py#L464-L502
`root.clear()` intend to clear memory, but it doesn't.
https://github.com/huggingface/datasets/blob/13a5b7db992ad5cf77895e4c0f76595314390418/datasets/wikipedia/wikipedia.py#L490
https://github.com/huggingface/datasets/blob/13a5b7db992ad5cf77895e4c0f76595314390418/datasets/wikipedia/wikipedia.py#L494
I replaced them with `elem.clear()`, then it seems to work correctly.
here is the notebook to reproduce it.
https://gist.github.com/miyamonz/dc06117302b6e85fa51cbf46dde6bb51#file-xtract_content-ipynb | 2,031 |
https://github.com/huggingface/datasets/issues/2029 | Loading a faiss index KeyError | [
"In your code `dataset2` doesn't contain the \"embeddings\" column, since it is created from the pandas DataFrame with columns \"text\" and \"label\".\r\n\r\nTherefore when you call `dataset2[embeddings_name]`, you get a `KeyError`.\r\n\r\nIf you want the \"embeddings\" column back, you can create `dataset2` with\r... | I've recently been testing out RAG and DPR embeddings, and I've run into an issue that is not apparent in the documentation.
The basic steps are:
1. Create a dataset (dataset1)
2. Create an embeddings column using DPR
3. Add a faiss index to the dataset
4. Save faiss index to a file
5. Create a new dataset (dataset2) with the same text and label information as dataset1
6. Try to load the faiss index from file to dataset2
7. Get `KeyError: "Column embeddings not in the dataset"`
I've made a colab notebook that should show exactly what I did. Please switch to GPU runtime; I didn't check on CPU.
https://colab.research.google.com/drive/1X0S9ZuZ8k0ybcoei4w7so6dS_WrABmIx?usp=sharing
Ubuntu Version
VERSION="18.04.5 LTS (Bionic Beaver)"
datasets==1.4.1
faiss==1.5.3
faiss-gpu==1.7.0
torch==1.8.0+cu101
transformers==4.3.3
NVIDIA-SMI 460.56
Driver Version: 460.32.03
CUDA Version: 11.2
Tesla K80
I was basically following the steps here: https://huggingface.co/docs/datasets/faiss_and_ea.html#adding-a-faiss-index
I included the exact code from the documentation at the end of the notebook to show that they don't work either.
| 2,029 |
https://github.com/huggingface/datasets/issues/2026 | KeyError on using map after renaming a column | [
"Hi,\r\n\r\nActually, the error occurs due to these two lines:\r\n```python\r\nraw_dataset.set_format('torch',columns=['img','label'])\r\nraw_dataset = raw_dataset.rename_column('img','image')\r\n```\r\n`Dataset.rename_column` doesn't update the `_format_columns` attribute, previously defined by `Dataset.set_format... | Hi,
I'm trying to use `cifar10` dataset. I want to rename the `img` feature to `image` in order to make it consistent with `mnist`, which I'm also planning to use. By doing this, I was trying to avoid modifying `prepare_train_features` function.
Here is what I try:
```python
transform = Compose([ToPILImage(),ToTensor(),Normalize([0.0,0.0,0.0],[1.0,1.0,1.0])])
def prepare_features(examples):
images = []
labels = []
print(examples)
for example_idx, example in enumerate(examples["image"]):
if transform is not None:
images.append(transform(examples["image"][example_idx].permute(2,0,1)))
else:
images.append(examples["image"][example_idx].permute(2,0,1))
labels.append(examples["label"][example_idx])
output = {"label":labels, "image":images}
return output
raw_dataset = load_dataset('cifar10')
raw_dataset.set_format('torch',columns=['img','label'])
raw_dataset = raw_dataset.rename_column('img','image')
features = datasets.Features({
"image": datasets.Array3D(shape=(3,32,32),dtype="float32"),
"label": datasets.features.ClassLabel(names=[
"airplane",
"automobile",
"bird",
"cat",
"deer",
"dog",
"frog",
"horse",
"ship",
"truck",
]),
})
train_dataset = raw_dataset.map(prepare_features, features = features,batched=True, batch_size=10000)
```
The error:
```python
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
<ipython-input-54-bf29672c53ee> in <module>()
14 ]),
15 })
---> 16 train_dataset = raw_dataset.map(prepare_features, features = features,batched=True, batch_size=10000)
2 frames
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in map(self, function, with_indices, 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)
1287 test_inputs = self[:2] if batched else self[0]
1288 test_indices = [0, 1] if batched else 0
-> 1289 update_data = does_function_return_dict(test_inputs, test_indices)
1290 logger.info("Testing finished, running the mapping function on the dataset")
1291
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in does_function_return_dict(inputs, indices)
1258 fn_args = [inputs] if input_columns is None else [inputs[col] for col in input_columns]
1259 processed_inputs = (
-> 1260 function(*fn_args, indices, **fn_kwargs) if with_indices else function(*fn_args, **fn_kwargs)
1261 )
1262 does_return_dict = isinstance(processed_inputs, Mapping)
<ipython-input-52-b4dccbafb70d> in prepare_features(examples)
3 labels = []
4 print(examples)
----> 5 for example_idx, example in enumerate(examples["image"]):
6 if transform is not None:
7 images.append(transform(examples["image"][example_idx].permute(2,0,1)))
KeyError: 'image'
```
The print statement inside returns this:
```python
{'label': tensor([6, 9])}
```
Apparently, both `img` and `image` do not exist after renaming.
Note that this code works fine with `img` everywhere.
Notebook: https://colab.research.google.com/drive/1SzESAlz3BnVYrgQeJ838vbMp1OsukiA2?usp=sharing
| 2,026 |
https://github.com/huggingface/datasets/issues/2022 | ValueError when rename_column on splitted dataset | [
"Hi,\r\n\r\nThis is a bug so thanks for reporting it. `Dataset.__setstate__` is the problem, which is called when `Dataset.rename_column` tries to copy the dataset with `copy.deepcopy(self)`. This only happens if the `split` arg in `load_dataset` was defined as `ReadInstruction`.\r\n\r\nTo overcome this issue, use... | Hi there,
I am loading `.tsv` file via `load_dataset` and subsequently split the rows into training and test set via the `ReadInstruction` API like so:
```python
split = {
'train': ReadInstruction('train', to=90, unit='%'),
'test': ReadInstruction('train', from_=-10, unit='%')
}
dataset = load_dataset(
path='csv', # use 'text' loading script to load from local txt-files
delimiter='\t', # xxx
data_files=text_files, # list of paths to local text files
split=split, # xxx
)
dataset
```
Part of output:
```python
DatasetDict({
train: Dataset({
features: ['sentence', 'sentiment'],
num_rows: 900
})
test: Dataset({
features: ['sentence', 'sentiment'],
num_rows: 100
})
})
```
Afterwards I'd like to rename the 'sentence' column to 'text' in order to be compatible with my modelin pipeline. If I run the following code I experience a `ValueError` however:
```python
dataset['train'].rename_column('sentence', 'text')
```
```python
/usr/local/lib/python3.7/dist-packages/datasets/splits.py in __init__(self, name)
353 for split_name in split_names_from_instruction:
354 if not re.match(_split_re, split_name):
--> 355 raise ValueError(f"Split name should match '{_split_re}'' but got '{split_name}'.")
356
357 def __str__(self):
ValueError: Split name should match '^\w+(\.\w+)*$'' but got 'ReadInstruction('.
```
In particular, these behavior does not arise if I use the deprecated `rename_column_` method. Any idea what causes the error? Would assume something in the way I defined the split.
Thanks in advance! :) | 2,022 |
https://github.com/huggingface/datasets/issues/2021 | Interactively doing save_to_disk and load_from_disk corrupts the datasets object? | [
"Hi,\r\n\r\nCan you give us a minimal reproducible example? This [part](https://huggingface.co/docs/datasets/master/processing.html#controling-the-cache-behavior) of the docs explains how to control caching."
] | dataset_info.json file saved after using save_to_disk gets corrupted as follows.
Is there a way to disable the cache that will save to /tmp/huggiface/datastes ?
I have a feeling there is a serious issue with cashing. | 2,021 |
https://github.com/huggingface/datasets/issues/2012 | No upstream branch | [
"What's the issue exactly ?\r\n\r\nGiven an `upstream` remote repository with url `https://github.com/huggingface/datasets.git`, you can totally rebase from `upstream/master`.\r\n\r\nIt's mentioned at the beginning how to add the `upstream` remote repository\r\n\r\nhttps://github.com/huggingface/datasets/blob/987df... | Feels like the documentation on adding a new dataset is outdated?
https://github.com/huggingface/datasets/blob/987df6b4e9e20fc0c92bc9df48137d170756fd7b/ADD_NEW_DATASET.md#L49-L54
There is no upstream branch on remote. | 2,012 |
https://github.com/huggingface/datasets/issues/2010 | Local testing fails | [
"I'm not able to reproduce on my side.\r\nCan you provide the full stacktrace please ?\r\nWhat version of `python` and `dill` do you have ? Which OS are you using ?",
"```\r\nco_filename = '<ipython-input-2-e0383a102aae>', returned_obj = [0]\r\n ... | I'm following the CI setup as described in
https://github.com/huggingface/datasets/blob/8eee4fa9e133fe873a7993ba746d32ca2b687551/.circleci/config.yml#L16-L19
in a new conda environment, at commit https://github.com/huggingface/datasets/commit/4de6dbf84e93dad97e1000120d6628c88954e5d4
and getting
```
FAILED tests/test_caching.py::RecurseDumpTest::test_dump_ipython_function - TypeError: an integer is required (got type bytes)
1 failed, 2321 passed, 5109 skipped, 10 warnings in 124.32s (0:02:04)
```
Seems like a discrepancy with CI, perhaps a lib version that's not controlled?
Tried with `pyarrow=={1.0.0,0.17.1,2.0.0}` | 2,010 |
https://github.com/huggingface/datasets/issues/2009 | Ambiguous documentation | [
"Hi @theo-m !\r\n\r\nA few lines above this line, you'll find that the `_split_generators` method returns a list of `SplitGenerator`s objects:\r\n\r\n```python\r\ndatasets.SplitGenerator(\r\n name=datasets.Split.VALIDATION,\r\n # These kwargs will be passed to _generate_examples\r\n gen_kwargs={\r\n ... | https://github.com/huggingface/datasets/blob/2ac9a0d24a091989f869af55f9f6411b37ff5188/templates/new_dataset_script.py#L156-L158
Looking at the template, I find this documentation line to be confusing, the method parameters don't include the `gen_kwargs` so I'm unclear where they're coming from.
Happy to push a PR with a clearer statement when I understand the meaning. | 2,009 |
https://github.com/huggingface/datasets/issues/2007 | How to not load huggingface datasets into memory | [
"So maybe a summary here: \r\nIf I could fit a large model with batch_size = X into memory, is there a way I could train this model for huge datasets with keeping setting the same? thanks ",
"The `datastets` library doesn't load datasets into memory. Therefore you can load a dataset that is terabytes big without ... | Hi
I am running this example from transformers library version 4.3.3:
(Here is the full documentation https://github.com/huggingface/transformers/issues/8771 but the running command should work out of the box)
USE_TF=0 deepspeed run_seq2seq.py --model_name_or_path google/mt5-base --dataset_name wmt16 --dataset_config_name ro-en --source_prefix "translate English to Romanian: " --task translation_en_to_ro --output_dir /test/test_large --do_train --do_eval --predict_with_generate --max_train_samples 500 --max_val_samples 500 --max_source_length 128 --max_target_length 128 --sortish_sampler --per_device_train_batch_size 8 --val_max_target_length 128 --deepspeed ds_config.json --num_train_epochs 1 --eval_steps 25000 --warmup_steps 500 --overwrite_output_dir
(Here please find the script: https://github.com/huggingface/transformers/blob/master/examples/seq2seq/run_seq2seq.py)
If you do not pass max_train_samples in above command to load the full dataset, then I get memory issue on a gpu with 24 GigBytes of memory.
I need to train large-scale mt5 model on large-scale datasets of wikipedia (multiple of them concatenated or other datasets in multiple languages like OPUS), could you help me how I can avoid loading the full data into memory? to make the scripts not related to data size?
In above example, I was hoping the script could work without relying on dataset size, so I can still train the model without subsampling training set.
thank you so much @lhoestq for your great help in advance
| 2,007 |
https://github.com/huggingface/datasets/issues/2005 | Setting to torch format not working with torchvision and MNIST | [
"Adding to the previous information, I think `torch.utils.data.DataLoader` is doing some conversion. \r\nWhat I tried:\r\n```python\r\ntrain_dataset = load_dataset('mnist')\r\n```\r\nI don't use any `map` or `set_format` or any `transform`. I use this directly, and try to load batches using the `DataLoader` with ba... | Hi
I am trying to use `torchvision.transforms` to handle the transformation of the image data in the `mnist` dataset. Assume I have a `transform` variable which contains the `torchvision.transforms` object.
A snippet of what I am trying to do:
```python
def prepare_features(examples):
images = []
labels = []
for example_idx, example in enumerate(examples["image"]):
if transform is not None:
images.append(transform(
np.array(examples["image"][example_idx], dtype=np.uint8)
))
else:
images.append(torch.tensor(np.array(examples["image"][example_idx], dtype=np.uint8)))
labels.append(torch.tensor(examples["label"][example_idx]))
output = {"label":labels, "image":images}
return output
raw_dataset = load_dataset('mnist')
train_dataset = raw_dataset.map(prepare_features, batched=True, batch_size=10000)
train_dataset.set_format("torch",columns=["image","label"])
```
After this, I check the type of the following:
```python
print(type(train_dataset["train"]["label"]))
print(type(train_dataset["train"]["image"][0]))
```
This leads to the following output:
```python
<class 'torch.Tensor'>
<class 'list'>
```
I use `torch.utils.DataLoader` for batches, the type of `batch["train"]["image"]` is also `<class 'list'>`.
I don't understand why only the `label` is converted to a torch tensor, why does the image not get converted? How can I fix this issue?
Thanks,
Gunjan
EDIT:
I just checked the shapes, and the types, `batch[image]` is a actually a list of list of tensors. Shape is (1,28,2,28), where `batch_size` is 2. I don't understand why this is happening. Ideally it should be a tensor of shape (2,1,28,28).
EDIT 2:
Inside `prepare_train_features`, the shape of `images[0]` is `torch.Size([1,28,28])`, the conversion is working. However, the output of the `map` is a list of list of list of list. | 2,005 |
https://github.com/huggingface/datasets/issues/2003 | Messages are being printed to the `stdout` | [
"This is expected to show this message to the user via stdout.\r\nThis way the users see it directly and can cancel the downloading if they want to.\r\nCould you elaborate why it would be better to have it in stderr instead of stdout ?",
"@lhoestq, sorry for the late reply\r\n\r\nI completely understand why you d... | In this code segment, we can see some messages are being printed to the `stdout`.
https://github.com/huggingface/datasets/blob/7e60bb509b595e8edc60a87f32b2bacfc065d607/src/datasets/builder.py#L545-L554
According to the comment, it is done intentionally, but I don't really understand why don't we log it with a higher level or print it directly to the `stderr`.
In my opinion, this kind of messages should never printed to the stdout. At least some configuration/flag should make it possible to provide in order to explicitly prevent the package to contaminate the stdout.
| 2,003 |
https://github.com/huggingface/datasets/issues/2001 | Empty evidence document ("provenance") in KILT ELI5 dataset | [
"Why did you close this issue? How did you end up finding the evidence documents? I'm running into a similar issue with other KILT tasks."
] | In the original KILT benchmark(https://github.com/facebookresearch/KILT),
all samples has its evidence document (i.e. wikipedia page id) for prediction.
For example, a sample in ELI5 dataset has the format including provenance (=evidence document) like this
`{"id": "1kiwfx", "input": "In Trading Places (1983, Akroyd/Murphy) how does the scheme at the end of the movie work? Why would buying a lot of OJ at a high price ruin the Duke Brothers?", "output": [{"answer": "I feel so old. People have been askinbg what happened at the end of this movie for what must be the last 15 years of my life. It never stops. Every year/month/fortnight, I see someone asking what happened, and someone explaining. Andf it will keep on happening, until I am 90yrs old, in a home, with nothing but the Internet and my bladder to keep me going. And there it will be: \"what happens at the end of Trading Places?\""}, {"provenance": [{"wikipedia_id": "242855", "title": "Futures contract", "section": "Section::::Abstract.", "start_paragraph_id": 1, "start_character": 14, "end_paragraph_id": 1, "end_character": 612, "bleu_score": 0.9232808519770748}]}], "meta": {"partial_evidence": [{"wikipedia_id": "520990", "title": "Trading Places", "section": "Section::::Plot.\n", "start_paragraph_id": 7, "end_paragraph_id": 7, "meta": {"evidence_span": ["On television, they learn that Clarence Beeks is transporting a secret USDA report on orange crop forecasts.", "On television, they learn that Clarence Beeks is transporting a secret USDA report on orange crop forecasts. Winthorpe and Valentine recall large payments made to Beeks by the Dukes and realize that the Dukes plan to obtain the report to corner the market on frozen orange juice.", "Winthorpe and Valentine recall large payments made to Beeks by the Dukes and realize that the Dukes plan to obtain the report to corner the market on frozen orange juice."]}}]}}`
However, KILT ELI5 dataset from huggingface datasets library only contain empty list of provenance.
`{'id': '1oy5tc', 'input': 'in football whats the point of wasting the first two plays with a rush - up the middle - not regular rush plays i get those', 'meta': {'left_context': '', 'mention': '', 'obj_surface': [], 'partial_evidence': [], 'right_context': '', 'sub_surface': [], 'subj_aliases': [], 'template_questions': []}, 'output': [{'answer': 'In most cases the O-Line is supposed to make a hole for the running back to go through. If you run too many plays to the outside/throws the defense will catch on.\n\nAlso, 2 5 yard plays gets you a new set of downs.', 'meta': {'score': 2}, 'provenance': []}, {'answer': "I you don't like those type of plays, watch CFL. We only get 3 downs so you can't afford to waste one. Lots more passing.", 'meta': {'score': 2}, 'provenance': []}]}
`
should i perform other procedure to obtain evidence documents? | 2,001 |
https://github.com/huggingface/datasets/issues/2000 | Windows Permission Error (most recent version of datasets) | [
"Hi @itsLuisa !\r\n\r\nCould you give us more information about the error you're getting, please?\r\nA copy-paste of the Traceback would be nice to get a better understanding of what is wrong :) ",
"Hello @SBrandeis , this is it:\r\n```\r\nTraceback (most recent call last):\r\n File \"C:\\Users\\Luisa\\AppData\\... | Hi everyone,
Can anyone help me with why the dataset loading script below raises a Windows Permission Error? I stuck quite closely to https://github.com/huggingface/datasets/blob/master/datasets/conll2003/conll2003.py , only I want to load the data from three local three-column tsv-files (id\ttokens\tpos_tags\n). I am using the most recent version of datasets. Thank you in advance!
Luisa
My script:
```
import datasets
import csv
logger = datasets.logging.get_logger(__name__)
class SampleConfig(datasets.BuilderConfig):
def __init__(self, **kwargs):
super(SampleConfig, self).__init__(**kwargs)
class Sample(datasets.GeneratorBasedBuilder):
BUILDER_CONFIGS = [
SampleConfig(name="conll2003", version=datasets.Version("1.0.0"), description="Conll2003 dataset"),
]
def _info(self):
return datasets.DatasetInfo(
description="Dataset with words and their POS-Tags",
features=datasets.Features(
{
"id": datasets.Value("string"),
"tokens": datasets.Sequence(datasets.Value("string")),
"pos_tags": datasets.Sequence(
datasets.features.ClassLabel(
names=[
"''",
",",
"-LRB-",
"-RRB-",
".",
":",
"CC",
"CD",
"DT",
"EX",
"FW",
"HYPH",
"IN",
"JJ",
"JJR",
"JJS",
"MD",
"NN",
"NNP",
"NNPS",
"NNS",
"PDT",
"POS",
"PRP",
"PRP$",
"RB",
"RBR",
"RBS",
"RP",
"TO",
"UH",
"VB",
"VBD",
"VBG",
"VBN",
"VBP",
"VBZ",
"WDT",
"WP",
"WRB",
"``"
]
)
),
}
),
supervised_keys=None,
homepage="https://catalog.ldc.upenn.edu/LDC2011T03",
citation="Weischedel, Ralph, et al. OntoNotes Release 4.0 LDC2011T03. Web Download. Philadelphia: Linguistic Data Consortium, 2011.",
)
def _split_generators(self, dl_manager):
loaded_files = dl_manager.download_and_extract(self.config.data_files)
return [
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": loaded_files["train"]}),
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": loaded_files["test"]}),
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": loaded_files["val"]})
]
def _generate_examples(self, filepath):
logger.info("generating examples from = %s", filepath)
with open(filepath, encoding="cp1252") as f:
data = csv.reader(f, delimiter="\t")
ids = list()
tokens = list()
pos_tags = list()
for id_, line in enumerate(data):
#print(line)
if len(line) == 1:
if tokens:
yield id_, {"id": ids, "tokens": tokens, "pos_tags": pos_tags}
ids = list()
tokens = list()
pos_tags = list()
else:
ids.append(line[0])
tokens.append(line[1])
pos_tags.append(line[2])
# last example
yield id_, {"id": ids, "tokens": tokens, "pos_tags": pos_tags}
def main():
dataset = datasets.load_dataset(
"data_loading.py", data_files={
"train": "train.tsv",
"test": "test.tsv",
"val": "val.tsv"
}
)
#print(dataset)
if __name__=="__main__":
main()
```
| 2,000 |
https://github.com/huggingface/datasets/issues/1997 | from datasets import MoleculeDataset, GEOMDataset | [] | I met the ImportError: cannot import name 'MoleculeDataset' from 'datasets'. Have anyone met the similar issues? Thanks! | 1,997 |
https://github.com/huggingface/datasets/issues/1996 | Error when exploring `arabic_speech_corpus` | [
"Thanks for reporting! We'll fix that as soon as possible",
"Actually soundfile is not a dependency of this dataset.\r\nThe error comes from a bug that was fixed in this commit: https://github.com/huggingface/datasets/pull/1767/commits/c304e63629f4453367de2fd42883a78768055532\r\nBasically the library used to cons... | Navigate to https://huggingface.co/datasets/viewer/?dataset=arabic_speech_corpus
Error:
```
ImportError: To be able to use this dataset, you need to install the following dependencies['soundfile'] using 'pip install soundfile' for instance'
Traceback:
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/streamlit/script_runner.py", line 332, in _run_script
exec(code, module.__dict__)
File "/home/sasha/nlp-viewer/run.py", line 233, in <module>
configs = get_confs(option)
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/streamlit/caching.py", line 604, in wrapped_func
return get_or_create_cached_value()
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/streamlit/caching.py", line 588, in get_or_create_cached_value
return_value = func(*args, **kwargs)
File "/home/sasha/nlp-viewer/run.py", line 145, in get_confs
module_path = nlp.load.prepare_module(path, dataset=True
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/datasets/load.py", line 342, in prepare_module
f"To be able to use this {module_type}, you need to install the following dependencies"
``` | 1,996 |
https://github.com/huggingface/datasets/issues/1994 | not being able to get wikipedia es language | [
"@lhoestq I really appreciate if you could help me providiing processed datasets, I do not really have access to enough resources to run the apache-beam and need to run the codes on these datasets. Only en/de/fr currently works, but I need all the languages more or less. thanks ",
"Hi @dorost1234, I think I can ... | Hi
I am trying to run a code with wikipedia of config 20200501.es, getting:
Traceback (most recent call last):
File "run_mlm_t5.py", line 608, in <module>
main()
File "run_mlm_t5.py", line 359, in main
datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
File "/dara/libs/anaconda3/envs/success432/lib/python3.7/site-packages/datasets-1.2.1-py3.7.egg/datasets/load.py", line 612, in load_dataset
ignore_verifications=ignore_verifications,
File "/dara/libs/anaconda3/envs/success432/lib/python3.7/site-packages/datasets-1.2.1-py3.7.egg/datasets/builder.py", line 527, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/dara/libs/anaconda3/envs/success432/lib/python3.7/site-packages/datasets-1.2.1-py3.7.egg/datasets/builder.py", line 1050, in _download_and_prepare
"\n\t`{}`".format(usage_example)
datasets.builder.MissingBeamOptions: Trying to generate a dataset using Apache Beam, yet no Beam Runner or PipelineOptions() has been provided in `load_dataset` or in the builder arguments. For big datasets it has to run on large-scale data processing tools like Dataflow, Spark, etc. More information about Apache Beam runners at https://beam.apache.org/documentation/runners/capability-matrix/
If you really want to run it locally because you feel like the Dataset is small enough, you can use the local beam runner called `DirectRunner` (you may run out of memory).
Example of usage:
`load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
thanks @lhoestq for any suggestion/help | 1,994 |
https://github.com/huggingface/datasets/issues/1993 | How to load a dataset with load_from disk and save it again after doing transformations without changing the original? | [
"Hi ! That looks like a bug, can you provide some code so that we can reproduce ?\r\nIt's not supposed to update the original dataset",
"Hi, I experimented with RAG. \r\n\r\nActually, you can run the [use_own_knowldge_dataset.py](https://github.com/shamanez/transformers/blob/rag-end-to-end-retrieval/examples/rese... | I am using the latest datasets library. In my work, I first use **load_from_disk** to load a data set that contains 3.8Gb information. Then during my training process, I update that dataset object and add new elements and save it in a different place.
When I save the dataset with **save_to_disk**, the original dataset which is already in the disk also gets updated. I do not want to update it. How to prevent from this?
| 1,993 |
https://github.com/huggingface/datasets/issues/1992 | `datasets.map` multi processing much slower than single processing | [
"Hi @hwijeen, you might want to look at issues #1796 and #1949. I think it could be something related to the I/O operations being performed.",
"I see that many people are experiencing the same issue. Is this problem considered an \"official\" bug that is worth a closer look? @lhoestq",
"Yes this looks like a bu... | Hi, thank you for the great library.
I've been using datasets to pretrain language models, and it often involves datasets as large as ~70G.
My data preparation step is roughly two steps: `load_dataset` which splits corpora into a table of sentences, and `map` converts a sentence into a list of integers, using a tokenizer.
I noticed that `map` function with `num_proc=mp.cpu_count() //2` takes more than 20 hours to finish the job where as `num_proc=1` gets the job done in about 5 hours. The machine I used has 40 cores, with 126G of RAM. There were no other jobs when `map` function was running.
What could be the reason? I would be happy to provide information necessary to spot the reason.
p.s. I was experiencing the imbalance issue mentioned in [here](https://github.com/huggingface/datasets/issues/610#issuecomment-705177036) when I was using multi processing.
p.s.2 When I run `map` with `num_proc=1`, I see one tqdm bar but all the cores are working. When `num_proc=20`, only 20 cores work.
| 1,992 |
https://github.com/huggingface/datasets/issues/1990 | OSError: Memory mapping file failed: Cannot allocate memory | [
"Do you think this is trying to bring the dataset into memory and if I can avoid it to save on memory so it only brings a batch into memory? @lhoestq thank you",
"It's not trying to bring the dataset into memory.\r\n\r\nActually, it's trying to memory map the dataset file, which is different. It allows to load l... | Hi,
I am trying to run a code with a wikipedia dataset, here is the command to reproduce the error. You can find the codes for run_mlm.py in huggingface repo here: https://github.com/huggingface/transformers/blob/v4.3.2/examples/language-modeling/run_mlm.py
```
python run_mlm.py --model_name_or_path bert-base-multilingual-cased --dataset_name wikipedia --dataset_config_name 20200501.en --do_train --do_eval --output_dir /dara/test --max_seq_length 128
```
I am using transformer version: 4.3.2
But I got memory erorr using this dataset, is there a way I could save on memory with dataset library with wikipedia dataset?
Specially I need to train a model with multiple of wikipedia datasets concatenated. thank you very much @lhoestq for your help and suggestions:
```
File "run_mlm.py", line 441, in <module>
main()
File "run_mlm.py", line 233, in main
split=f"train[{data_args.validation_split_percentage}%:]",
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/load.py", line 750, in load_dataset
ds = builder_instance.as_dataset(split=split, ignore_verifications=ignore_verifications, in_memory=keep_in_memory)
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/builder.py", line 740, in as_dataset
map_tuple=True,
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/utils/py_utils.py", line 225, in map_nested
return function(data_struct)
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/builder.py", line 757, in _build_single_dataset
in_memory=in_memory,
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/builder.py", line 829, in _as_dataset
in_memory=in_memory,
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/arrow_reader.py", line 215, in read
return self.read_files(files=files, original_instructions=instructions, in_memory=in_memory)
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/arrow_reader.py", line 236, in read_files
pa_table = self._read_files(files, in_memory=in_memory)
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/arrow_reader.py", line 171, in _read_files
pa_table: pa.Table = self._get_dataset_from_filename(f_dict, in_memory=in_memory)
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/arrow_reader.py", line 302, in _get_dataset_from_filename
pa_table = ArrowReader.read_table(filename, in_memory=in_memory)
File "/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/arrow_reader.py", line 322, in read_table
stream = stream_from(filename)
File "pyarrow/io.pxi", line 782, in pyarrow.lib.memory_map
File "pyarrow/io.pxi", line 743, in pyarrow.lib.MemoryMappedFile._open
File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 99, in pyarrow.lib.check_status
OSError: Memory mapping file failed: Cannot allocate memory
```
| 1,990 |
https://github.com/huggingface/datasets/issues/1989 | Question/problem with dataset labels | [
"It seems that I get parsing errors for various fields in my data. For example now I get this:\r\n```\r\n File \"../../../models/tr-4.3.2/run_puppets.py\", line 523, in <module>\r\n main()\r\n File \"../../../models/tr-4.3.2/run_puppets.py\", line 249, in main\r\n datasets = load_dataset(\"csv\", data_files... | Hi, I'm using a dataset with two labels "nurse" and "not nurse". For whatever reason (that I don't understand), I get an error that I think comes from the datasets package (using csv). Everything works fine if the labels are "nurse" and "surgeon".
This is the trace I get:
```
File "../../../models/tr-4.3.2/run_puppets.py", line 523, in <module>
main()
File "../../../models/tr-4.3.2/run_puppets.py", line 249, in main
datasets = load_dataset("csv", data_files=data_files)
File "/dccstor/redrug_ier/envs/last-tr/lib/python3.8/site-packages/datasets/load.py", line 740, in load_dataset
builder_instance.download_and_prepare(
File "/dccstor/redrug_ier/envs/last-tr/lib/python3.8/site-packages/datasets/builder.py", line 572, in download_and_prepare
self._download_and_prepare(
File "/dccstor/redrug_ier/envs/last-tr/lib/python3.8/site-packages/datasets/builder.py", line 650, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/dccstor/redrug_ier/envs/last-tr/lib/python3.8/site-packages/datasets/builder.py", line 1028, in _prepare_split
writer.write_table(table)
File "/dccstor/redrug_ier/envs/last-tr/lib/python3.8/site-packages/datasets/arrow_writer.py", line 292, in write_table
pa_table = pa_table.cast(self._schema)
File "pyarrow/table.pxi", line 1311, in pyarrow.lib.Table.cast
File "pyarrow/table.pxi", line 265, in pyarrow.lib.ChunkedArray.cast
File "/dccstor/redrug_ier/envs/last-tr/lib/python3.8/site-packages/pyarrow/compute.py", line 87, in cast
return call_function("cast", [arr], options)
File "pyarrow/_compute.pyx", line 298, in pyarrow._compute.call_function
File "pyarrow/_compute.pyx", line 192, in pyarrow._compute.Function.call
File "pyarrow/error.pxi", line 122, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 84, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: Failed to parse string: not nurse
```
Any ideas how to fix this? For now, I'll probably make them numeric. | 1,989 |
https://github.com/huggingface/datasets/issues/1988 | Readme.md is misleading about kinds of datasets? | [
"Hi ! Yes it's possible to use image data. There are already a few of them available (MNIST, CIFAR..)"
] | Hi!
At the README.MD, you say: "efficient data pre-processing: simple, fast and reproducible data pre-processing for the above public datasets as well as your own local datasets in CSV/JSON/text. "
But here:
https://github.com/huggingface/datasets/blob/master/templates/new_dataset_script.py#L82-L117
You mention other kinds of datasets, with images and so on. I'm confused.
Is it possible to use it to store, say, imagenet locally? | 1,988 |
https://github.com/huggingface/datasets/issues/1987 | wmt15 is broken | [
"It's reachable for the viewer and me, so I suppose it was down at that moment?"
] | While testing the hotfix, I tried a random other wmt release and found wmt15 to be broken:
```
python -c 'from datasets import load_dataset; load_dataset("wmt15", "de-en")'
Downloading: 2.91kB [00:00, 818kB/s]
Downloading: 3.02kB [00:00, 897kB/s]
Downloading: 41.1kB [00:00, 19.1MB/s]
Downloading and preparing dataset wmt15/de-en (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /home/stas/.cache/huggingface/datasets/wmt15/de-en/1.0.0/39ad5f9262a0910a8ad7028ad432731ad23fdf91f2cebbbf2ba4776b9859e87f...
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/load.py", line 740, in load_dataset
builder_instance.download_and_prepare(
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/builder.py", line 578, in download_and_prepare
self._download_and_prepare(
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/builder.py", line 634, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/home/stas/.cache/huggingface/modules/datasets_modules/datasets/wmt15/39ad5f9262a0910a8ad7028ad432731ad23fdf91f2cebbbf2ba4776b9859e87f/wmt_utils.py", line 757, in _split_generators
downloaded_files = dl_manager.download_and_extract(urls_to_download)
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/download_manager.py", line 283, in download_and_extract
return self.extract(self.download(url_or_urls))
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/download_manager.py", line 191, in download
downloaded_path_or_paths = map_nested(
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/py_utils.py", line 203, in map_nested
mapped = [
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/py_utils.py", line 204, in <listcomp>
_single_map_nested((function, obj, types, None, True)) for obj in tqdm(iterable, disable=disable_tqdm)
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/py_utils.py", line 160, in _single_map_nested
mapped = [_single_map_nested((function, v, types, None, True)) for v in pbar]
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/py_utils.py", line 160, in <listcomp>
mapped = [_single_map_nested((function, v, types, None, True)) for v in pbar]
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/py_utils.py", line 142, in _single_map_nested
return function(data_struct)
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/download_manager.py", line 214, in _download
return cached_path(url_or_filename, download_config=download_config)
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/file_utils.py", line 274, in cached_path
output_path = get_from_cache(
File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/datasets/utils/file_utils.py", line 614, in get_from_cache
raise FileNotFoundError("Couldn't find file at {}".format(url))
FileNotFoundError: Couldn't find file at https://huggingface.co/datasets/wmt/wmt15/resolve/main/training-parallel-nc-v10.tgz
``` | 1,987 |
https://github.com/huggingface/datasets/issues/1986 | wmt datasets fail to load | [
"caching issue, seems to work again.."
] | ~\.cache\huggingface\modules\datasets_modules\datasets\wmt14\43e717d978d2261502b0194999583acb874ba73b0f4aed0ada2889d1bb00f36e\wmt_utils.py in _split_generators(self, dl_manager)
758 # Extract manually downloaded files.
759 manual_files = dl_manager.extract(manual_paths_dict)
--> 760 extraction_map = dict(downloaded_files, **manual_files)
761
762 for language in self.config.language_pair:
TypeError: type object argument after ** must be a mapping, not list | 1,986 |
https://github.com/huggingface/datasets/issues/1984 | Add tests for WMT datasets | [
"Dummy data generation is deprecated now. Closing."
] | As requested in #1981, we need tests for WMT datasets, using dummy data. | 1,984 |
https://github.com/huggingface/datasets/issues/1983 | The size of CoNLL-2003 is not consistant with the official release. | [
"Hi,\r\n\r\nif you inspect the raw data, you can find there are 946 occurrences of `-DOCSTART- -X- -X- O` in the train split and `14041 + 946 = 14987`, which is exactly the number of sentences the authors report. `-DOCSTART-` is a special line that acts as a boundary between two different documents and is filtered ... | Thanks for the dataset sharing! But when I use conll-2003, I meet some questions.
The statistics of conll-2003 in this repo is :
\#train 14041 \#dev 3250 \#test 3453
While the official statistics is:
\#train 14987 \#dev 3466 \#test 3684
Wish for your reply~ | 1,983 |
https://github.com/huggingface/datasets/issues/1981 | wmt datasets fail to load | [
"@stas00 Mea culpa... May I fix this tomorrow morning?",
"yes, of course, I reverted to the version before that and it works ;)\r\n\r\nbut since a new release was just made you will probably need to make a hotfix.\r\n\r\nand add the wmt to the tests?",
"Sure, I will implement a regression test!",
"@stas00 it ... | on master:
```
python -c 'from datasets import load_dataset; load_dataset("wmt14", "de-en")'
Downloading and preparing dataset wmt14/de-en (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /home/stas/.cache/huggingface/datasets/wmt14/de-en/1.0.0/43e717d978d2261502b0194999583acb874ba73b0f4aed0ada2889d1bb00f36e...
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/load.py", line 740, in load_dataset
builder_instance.download_and_prepare(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/builder.py", line 578, in download_and_prepare
self._download_and_prepare(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/builder.py", line 634, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/home/stas/.cache/huggingface/modules/datasets_modules/datasets/wmt14/43e717d978d2261502b0194999583acb874ba73b0f4aed0ada2889d1bb00f36e/wmt_utils.py", line 760, in _split_generators
extraction_map = dict(downloaded_files, **manual_files)
```
it worked fine recently. same problem if I try wmt16.
git bisect points to this commit from Feb 25 as the culprit https://github.com/huggingface/datasets/commit/792f1d9bb1c5361908f73e2ef7f0181b2be409fa
@albertvillanova | 1,981 |
https://github.com/huggingface/datasets/issues/1977 | ModuleNotFoundError: No module named 'apache_beam' for wikipedia datasets | [
"I sometimes also get this error with other languages of the same dataset:\r\n\r\n File \"/dara/libs/anaconda3/envs/code/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/arrow_reader.py\", line 322, in read_table\r\n stream = stream_from(filename)\r\n File \"pyarrow/io.pxi\", line 782, in pyarrow.... | Hi
I am trying to run run_mlm.py code [1] of huggingface with following "wikipedia"/ "20200501.aa" dataset:
`python run_mlm.py --model_name_or_path bert-base-multilingual-cased --dataset_name wikipedia --dataset_config_name 20200501.aa --do_train --do_eval --output_dir /tmp/test-mlm --max_seq_length 256
`
I am getting this error, but as per documentation, huggingface dataset provide processed version of this dataset and users can load it without requiring setup extra settings for apache-beam. could you help me please to load this dataset?
Do you think I can run run_ml.py with this dataset? or anyway I could subsample and train the model? I greatly appreciate providing the processed version of all languages for this dataset, which allow the user to use them without setting up apache-beam,. thanks
I really appreciate your help.
@lhoestq
thanks.
[1] https://github.com/huggingface/transformers/blob/master/examples/language-modeling/run_mlm.py
error I get:
```
>>> import datasets
>>> datasets.load_dataset("wikipedia", "20200501.aa")
Downloading and preparing dataset wikipedia/20200501.aa (download: Unknown size, generated: Unknown size, post-processed: Unknown size, total: Unknown size) to /dara/temp/cache_home_2/datasets/wikipedia/20200501.aa/1.0.0/4021357e28509391eab2f8300d9b689e7e8f3a877ebb3d354b01577d497ebc63...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/dara/temp/libs/anaconda3/envs/codes/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/load.py", line 746, in load_dataset
use_auth_token=use_auth_token,
File "/dara/temp/libs/anaconda3/envs/codes/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/builder.py", line 573, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/dara/temp/libs/anaconda3/envs/codes/lib/python3.7/site-packages/datasets-1.3.0-py3.7.egg/datasets/builder.py", line 1099, in _download_and_prepare
import apache_beam as beam
ModuleNotFoundError: No module named 'apache_beam'
``` | 1,977 |
https://github.com/huggingface/datasets/issues/1973 | Question: what gets stored in the datasets cache and why is it so huge? | [
"Echo'ing this observation: I have a few datasets in the neighborhood of 2GB CSVs uncompressed, and when I use something like `Dataset.save_to_disk()` it's ~18GB on disk.\r\n\r\nIf this is unexpected behavior, would be happy to help run debugging as needed.",
"Thanks @ioana-blue for pointing out this problem (and... | I'm running several training jobs (around 10) with a relatively large dataset (3M samples). The datasets cache reached 178G and it seems really large. What is it stored in there and why is it so large? I don't think I noticed this problem before and seems to be related to the new version of the datasets library. Any insight? Thank you! | 1,973 |
https://github.com/huggingface/datasets/issues/1972 | 'Dataset' object has no attribute 'rename_column' | [
"Hi ! `rename_column` has been added recently and will be available in the next release"
] | 'Dataset' object has no attribute 'rename_column' | 1,972 |
https://github.com/huggingface/datasets/issues/1965 | Can we parallelized the add_faiss_index process over dataset shards ? | [
"Hi !\r\nAs far as I know not all faiss indexes can be computed in parallel and then merged. \r\nFor example [here](https://github.com/facebookresearch/faiss/wiki/Special-operations-on-indexes#splitting-and-merging-indexes) is is mentioned that only IndexIVF indexes can be merged.\r\nMoreover faiss already works us... | I am thinking of making the **add_faiss_index** process faster. What if we run the add_faiss_index process on separate dataset shards and then combine them before (dataset.concatenate) saving the faiss.index file ?
I feel theoretically this will reduce the accuracy of retrieval since it affects the indexing process.
@lhoestq
| 1,965 |
https://github.com/huggingface/datasets/issues/1964 | Datasets.py function load_dataset does not match squad dataset | [
"Hi !\r\n\r\nTo fix 1, an you try to run this code ?\r\n```python\r\nfrom datasets import load_dataset\r\n\r\nload_dataset(\"squad\", download_mode=\"force_redownload\")\r\n```\r\nMaybe the file your downloaded was corrupted, in this case redownloading this way should fix your issue 1.\r\n\r\nRegarding your 2nd poi... | ### 1 When I try to train lxmert,and follow the code in README that --dataset name:
```shell
python examples/question-answering/run_qa.py --model_name_or_path unc-nlp/lxmert-base-uncased --dataset_name squad --do_train --do_eval --per_device_train_batch_size 12 --learning_rate 3e-5 --num_train_epochs 2 --max_seq_length 384 --doc_stride 128 --output_dir /home2/zhenggo1/checkpoint/lxmert_squad
```
the bug is that:
```
Downloading and preparing dataset squad/plain_text (download: 33.51 MiB, generated: 85.75 MiB, post-processed: Unknown size, total: 119.27 MiB) to /home2/zhenggo1/.cache/huggingface/datasets/squad/plain_text/1.0.0/4c81550d83a2ac7c7ce23783bd8ff36642800e6633c1f18417fb58c3ff50cdd7...
Traceback (most recent call last):
File "examples/question-answering/run_qa.py", line 501, in <module>
main()
File "examples/question-answering/run_qa.py", line 217, in main
datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
File "/home2/zhenggo1/anaconda3/envs/lpot/lib/python3.7/site-packages/datasets/load.py", line 746, in load_dataset
use_auth_token=use_auth_token,
File "/home2/zhenggo1/anaconda3/envs/lpot/lib/python3.7/site-packages/datasets/builder.py", line 573, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home2/zhenggo1/anaconda3/envs/lpot/lib/python3.7/site-packages/datasets/builder.py", line 633, in _download_and_prepare
self.info.download_checksums, dl_manager.get_recorded_sizes_checksums(), "dataset source files"
File "/home2/zhenggo1/anaconda3/envs/lpot/lib/python3.7/site-packages/datasets/utils/info_utils.py", line 39, in verify_checksums
raise NonMatchingChecksumError(error_msg + str(bad_urls))
datasets.utils.info_utils.NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v1.1.json']
```
And I try to find the [checksum link](https://github.com/huggingface/datasets/blob/master/datasets/squad/dataset_infos.json)
,is the problem plain_text do not have a checksum?
### 2 When I try to train lxmert,and use local dataset:
```
python examples/question-answering/run_qa.py --model_name_or_path unc-nlp/lxmert-base-uncased --train_file $SQUAD_DIR/train-v1.1.json --validation_file $SQUAD_DIR/dev-v1.1.json --do_train --do_eval --per_device_train_batch_size 12 --learning_rate 3e-5 --num_train_epochs 2 --max_seq_length 384 --doc_stride 128 --output_dir /home2/zhenggo1/checkpoint/lxmert_squad
```
The bug is that
```
['title', 'paragraphs']
Traceback (most recent call last):
File "examples/question-answering/run_qa.py", line 501, in <module>
main()
File "examples/question-answering/run_qa.py", line 273, in main
answer_column_name = "answers" if "answers" in column_names else column_names[2]
IndexError: list index out of range
```
I print the answer_column_name and find that local squad dataset need the package datasets to preprocessing so that the code below can work:
```
if training_args.do_train:
column_names = datasets["train"].column_names
else:
column_names = datasets["validation"].column_names
print(datasets["train"].column_names)
question_column_name = "question" if "question" in column_names else column_names[0]
context_column_name = "context" if "context" in column_names else column_names[1]
answer_column_name = "answers" if "answers" in column_names else column_names[2]
```
## Please tell me how to fix the bug,thks a lot! | 1,964 |
https://github.com/huggingface/datasets/issues/1963 | bug in SNLI dataset | [
"Hi ! The labels -1 correspond to the examples without gold labels in the original snli dataset.\r\nFeel free to remove these examples if you don't need them by using\r\n```python\r\ndata = data.filter(lambda x: x[\"label\"] != -1)\r\n```"
] | Hi
There is label of -1 in train set of SNLI dataset, please find the code below:
```
import numpy as np
import datasets
data = datasets.load_dataset("snli")["train"]
labels = []
for d in data:
labels.append(d["label"])
print(np.unique(labels))
```
and results:
`[-1 0 1 2]`
version of datasets used:
`datasets 1.2.1 <pip>
`
thanks for your help. @lhoestq | 1,963 |
https://github.com/huggingface/datasets/issues/1959 | Bug in skip_rows argument of load_dataset function ? | [
"Hi,\r\n\r\ntry `skiprows` instead. This part is not properly documented in the docs it seems.\r\n\r\n@lhoestq I'll fix this as part of a bigger PR that fixes typos in the docs."
] | Hello everyone,
I'm quite new to Git so sorry in advance if I'm breaking some ground rules of issues posting... :/
I tried to use the load_dataset function, from Huggingface datasets library, on a csv file using the skip_rows argument described on Huggingface page to skip the first row containing column names
`test_dataset = load_dataset('csv', data_files=['test_wLabel.tsv'], delimiter='\t', column_names=["id", "sentence", "label"], skip_rows=1)`
But I got the following error message
`__init__() got an unexpected keyword argument 'skip_rows'`
Have I used the wrong argument ? Am I missing something or is this a bug ?
Thank you very much for your time,
Best regards,
Arthur | 1,959 |
https://github.com/huggingface/datasets/issues/1958 | XSum dataset download link broken | [
"Never mind, I ran it again and it worked this time. Strange."
] | I did
```
from datasets import load_dataset
dataset = load_dataset("xsum")
```
This returns
`ConnectionError: Couldn't reach http://bollin.inf.ed.ac.uk/public/direct/XSUM-EMNLP18-Summary-Data-Original.tar.gz` | 1,958 |
https://github.com/huggingface/datasets/issues/1956 | [distributed env] potentially unsafe parallel execution | [
"You can pass the same `experiment_id` for all the metrics of the same group, and use another `experiment_id` for the other groups.\r\nMaybe we can add an environment variable that sets the default value for `experiment_id` ? What do you think ?",
"Ah, you're absolutely correct, @lhoestq - it's exactly the equiva... | ```
metric = load_metric('glue', 'mrpc', num_process=num_process, process_id=rank)
```
presumes that there is only one set of parallel processes running - and will intermittently fail if you have multiple sets running as they will surely overwrite each other. Similar to https://github.com/huggingface/datasets/issues/1942 (but for a different reason).
That's why dist environments use some unique to a group identifier so that each group is dealt with separately.
e.g. the env-way of pytorch dist syncing is done with a unique per set `MASTER_ADDRESS+MASTER_PORT`
So ideally this interface should ask for a shared secret to do the right thing.
I'm not reporting an immediate need, but am only flagging that this will hit someone down the road.
This problem can be remedied by adding a new optional `shared_secret` option, which can then be used to differentiate different groups of processes. and this secret should be part of the file lock name and the experiment.
Thank you | 1,956 |
https://github.com/huggingface/datasets/issues/1954 | add a new column | [
"Hi\r\nnot sure how change the lable after creation, but this is an issue not dataset request. thanks ",
"Hi ! Currently you have to use `map` . You can see an example of how to do it in this comment: https://github.com/huggingface/datasets/issues/853#issuecomment-727872188\r\n\r\nIn the future we'll add support ... | Hi
I'd need to add a new column to the dataset, I was wondering how this can be done? thanks
@lhoestq | 1,954 |
https://github.com/huggingface/datasets/issues/1949 | Enable Fast Filtering using Arrow Dataset | [
"Hi @gchhablani :)\r\nThanks for proposing your help !\r\n\r\nI'll be doing a refactor of some parts related to filtering in the scope of https://github.com/huggingface/datasets/issues/1877\r\nSo I would first wait for this refactor to be done before working on the filtering. In particular because I plan to make th... | Hi @lhoestq,
As mentioned in Issue #1796, I would love to work on enabling fast filtering/mapping. Can you please share the expectations? It would be great if you could point me to the relevant methods/files involved. Or the docs or maybe an overview of `arrow_dataset.py`. I only ask this because I am having trouble getting started ;-;
Any help would be appreciated.
Thanks,
Gunjan | 1,949 |
https://github.com/huggingface/datasets/issues/1948 | dataset loading logger level | [
"These warnings are showed when there's a call to `.map` to say to the user that a dataset is reloaded from the cache instead of being recomputed.\r\nThey are warnings since we want to make sure the users know that it's not recomputed.",
"Thank you for explaining the intention, @lhoestq \r\n\r\n1. Could it be the... | on master I get this with `--dataset_name wmt16 --dataset_config ro-en`:
```
WARNING:datasets.arrow_dataset:Loading cached processed dataset at /home/stas/.cache/huggingface/datasets/wmt16/ro-en/1.0.0/9dc00622c30446e99c4c63d12a484ea4fb653f2f37c867d6edcec839d7eae50f/cache-2e01bead8cf42e26.arrow
WARNING:datasets.arrow_dataset:Loading cached processed dataset at /home/stas/.cache/huggingface/datasets/wmt16/ro-en/1.0.0/9dc00622c30446e99c4c63d12a484ea4fb653f2f37c867d6edcec839d7eae50f/cache-ac3bebaf4f91f776.arrow
WARNING:datasets.arrow_dataset:Loading cached processed dataset at /home/stas/.cache/huggingface/datasets/wmt16/ro-en/1.0.0/9dc00622c30446e99c4c63d12a484ea4fb653f2f37c867d6edcec839d7eae50f/cache-810c3e61259d73a9.arrow
```
why are those WARNINGs? Should be INFO, no?
warnings should only be used when a user needs to pay attention to something, this is just informative - I'd even say it should be DEBUG, but definitely not WARNING.
Thank you.
| 1,948 |
https://github.com/huggingface/datasets/issues/1945 | AttributeError: 'DatasetDict' object has no attribute 'concatenate_datasets' | [
"sorry my mistake, datasets were overwritten closing now, thanks a lot"
] | Hi
I am trying to concatenate a list of huggingface datastes as:
` train_dataset = datasets.concatenate_datasets(train_datasets)
`
Here is the `train_datasets` when I print:
```
[Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 120361
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 2670
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 6944
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 38140
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 173711
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 1655
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 4274
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 2019
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 2109
}), Dataset({
features: ['attention_mask', 'idx', 'input_ids', 'label', 'question1', 'question2', 'token_type_ids'],
num_rows: 11963
})]
```
I am getting the following error:
`AttributeError: 'DatasetDict' object has no attribute 'concatenate_datasets'
`
I was wondering if you could help me with this issue, thanks a lot | 1,945 |
https://github.com/huggingface/datasets/issues/1942 | [experiment] missing default_experiment-1-0.arrow | [
"Hi !\r\n\r\nThe cache at `~/.cache/huggingface/metrics` stores the users data for metrics computations (hence the arrow files).\r\n\r\nHowever python modules (i.e. dataset scripts, metric scripts) are stored in `~/.cache/huggingface/modules/datasets_modules`.\r\n\r\nIn particular the metrics are cached in `~/.cach... | the original report was pretty bad and incomplete - my apologies!
Please see the complete version here: https://github.com/huggingface/datasets/issues/1942#issuecomment-786336481
------------
As mentioned here https://github.com/huggingface/datasets/issues/1939 metrics don't get cached, looking at my local `~/.cache/huggingface/metrics` - there are many `*.arrow.lock` files but zero metrics files.
w/o the network I get:
```
FileNotFoundError: [Errno 2] No such file or directory: '~/.cache/huggingface/metrics/sacrebleu/default/default_experiment-1-0.arrow
```
there is just `~/.cache/huggingface/metrics/sacrebleu/default/default_experiment-1-0.arrow.lock`
I did run the same `run_seq2seq.py` script on the instance with network and it worked just fine, but only the lock file was left behind.
this is with master.
Thank you. | 1,942 |
https://github.com/huggingface/datasets/issues/1941 | Loading of FAISS index fails for index_name = 'exact' | [
"Thanks for reporting ! I'm taking a look",
"Index training was missing, I fixed it here: https://github.com/huggingface/datasets/commit/f5986c46323583989f6ed1dabaf267854424a521\r\n\r\nCan you try again please ?",
"Works great 👍 I just put a minor comment on the commit, I think you meant to pass the `train_siz... | Hi,
It looks like loading of FAISS index now fails when using index_name = 'exact'.
For example, from the RAG [model card](https://huggingface.co/facebook/rag-token-nq?fbclid=IwAR3bTfhls5U_t9DqsX2Vzb7NhtRHxJxfQ-uwFT7VuCPMZUM2AdAlKF_qkI8#usage).
Running `transformers==4.3.2` and datasets installed from source on latest `master` branch.
```bash
(venv) sergey_mkrtchyan datasets (master) $ python
Python 3.8.6 (v3.8.6:db455296be, Sep 23 2020, 13:31:39)
[Clang 6.0 (clang-600.0.57)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from transformers import RagTokenizer, RagRetriever, RagTokenForGeneration
>>> tokenizer = RagTokenizer.from_pretrained("facebook/rag-token-nq")
>>> retriever = RagRetriever.from_pretrained("facebook/rag-token-nq", index_name="exact", use_dummy_dataset=True)
Using custom data configuration dummy.psgs_w100.nq.no_index-dummy=True,with_index=False
Reusing dataset wiki_dpr (/Users/sergey_mkrtchyan/.cache/huggingface/datasets/wiki_dpr/dummy.psgs_w100.nq.no_index-dummy=True,with_index=False/0.0.0/8a97e0f4fa5bc46e179474db6a61b09d5d2419d2911835bd3f91d110c936d8bb)
Using custom data configuration dummy.psgs_w100.nq.exact-50b6cda57ff32ab4
Reusing dataset wiki_dpr (/Users/sergey_mkrtchyan/.cache/huggingface/datasets/wiki_dpr/dummy.psgs_w100.nq.exact-50b6cda57ff32ab4/0.0.0/8a97e0f4fa5bc46e179474db6a61b09d5d2419d2911835bd3f91d110c936d8bb)
0%| | 0/10 [00:00<?, ?it/s]
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/Users/sergey_mkrtchyan/workspace/cformers/venv/lib/python3.8/site-packages/transformers/models/rag/retrieval_rag.py", line 425, in from_pretrained
return cls(
File "/Users/sergey_mkrtchyan/workspace/cformers/venv/lib/python3.8/site-packages/transformers/models/rag/retrieval_rag.py", line 387, in __init__
self.init_retrieval()
File "/Users/sergey_mkrtchyan/workspace/cformers/venv/lib/python3.8/site-packages/transformers/models/rag/retrieval_rag.py", line 458, in init_retrieval
self.index.init_index()
File "/Users/sergey_mkrtchyan/workspace/cformers/venv/lib/python3.8/site-packages/transformers/models/rag/retrieval_rag.py", line 284, in init_index
self.dataset = load_dataset(
File "/Users/sergey_mkrtchyan/workspace/huggingface/datasets/src/datasets/load.py", line 750, in load_dataset
ds = builder_instance.as_dataset(split=split, ignore_verifications=ignore_verifications, in_memory=keep_in_memory)
File "/Users/sergey_mkrtchyan/workspace/huggingface/datasets/src/datasets/builder.py", line 734, in as_dataset
datasets = utils.map_nested(
File "/Users/sergey_mkrtchyan/workspace/huggingface/datasets/src/datasets/utils/py_utils.py", line 195, in map_nested
return function(data_struct)
File "/Users/sergey_mkrtchyan/workspace/huggingface/datasets/src/datasets/builder.py", line 769, in _build_single_dataset
post_processed = self._post_process(ds, resources_paths)
File "/Users/sergey_mkrtchyan/.cache/huggingface/modules/datasets_modules/datasets/wiki_dpr/8a97e0f4fa5bc46e179474db6a61b09d5d2419d2911835bd3f91d110c936d8bb/wiki_dpr.py", line 205, in _post_process
dataset.add_faiss_index("embeddings", custom_index=index)
File "/Users/sergey_mkrtchyan/workspace/huggingface/datasets/src/datasets/arrow_dataset.py", line 2516, in add_faiss_index
super().add_faiss_index(
File "/Users/sergey_mkrtchyan/workspace/huggingface/datasets/src/datasets/search.py", line 416, in add_faiss_index
faiss_index.add_vectors(self, column=column, train_size=train_size, faiss_verbose=faiss_verbose)
File "/Users/sergey_mkrtchyan/workspace/huggingface/datasets/src/datasets/search.py", line 281, in add_vectors
self.faiss_index.add(vecs)
File "/Users/sergey_mkrtchyan/workspace/cformers/venv/lib/python3.8/site-packages/faiss/__init__.py", line 104, in replacement_add
self.add_c(n, swig_ptr(x))
File "/Users/sergey_mkrtchyan/workspace/cformers/venv/lib/python3.8/site-packages/faiss/swigfaiss.py", line 3263, in add
return _swigfaiss.IndexHNSW_add(self, n, x)
RuntimeError: Error in virtual void faiss::IndexHNSW::add(faiss::Index::idx_t, const float *) at /Users/runner/work/faiss-wheels/faiss-wheels/faiss/faiss/IndexHNSW.cpp:356: Error: 'is_trained' failed
>>>
```
The issue seems to be related to the scalar quantization in faiss added in this commit: 8c5220307c33f00e01c3bf7b8. Reverting it fixes the issue.
| 1,941 |
https://github.com/huggingface/datasets/issues/1940 | Side effect when filtering data due to `does_function_return_dict` call in `Dataset.map()` | [
"Thanks for the report !\r\n\r\nCurrently we don't have a way to let the user easily disable this behavior.\r\nHowever I agree that we should support stateful processing functions, ideally by removing `does_function_return_dict`.\r\n\r\nWe needed this function in order to know whether the `map` functions needs to w... | Hi there!
In my codebase I have a function to filter rows in a dataset, selecting only a certain number of examples per class. The function passes a extra argument to maintain a counter of the number of dataset rows/examples already selected per each class, which are the ones I want to keep in the end:
```python
def fill_train_examples_per_class(example, per_class_limit: int, counter: collections.Counter):
label = int(example['label'])
current_counter = counter.get(label, 0)
if current_counter < per_class_limit:
counter[label] = current_counter + 1
return True
return False
```
At some point I invoke it through the `Dataset.filter()` method in the `arrow_dataset.py` module like this:
```python
...
kwargs = {"per_class_limit": train_examples_per_class_limit, "counter": Counter()}
datasets['train'] = datasets['train'].filter(fill_train_examples_per_class, num_proc=1, fn_kwargs=kwargs)
...
```
The problem is that, passing a stateful container (the counter,) provokes a side effect in the new filtered dataset obtained. This is due to the fact that at some point in `filter()`, the `map()`'s function `does_function_return_dict` is invoked in line [1290](https://github.com/huggingface/datasets/blob/96578adface7e4bc1f3e8bafbac920d72ca1ca60/src/datasets/arrow_dataset.py#L1290).
When this occurs, the state of the counter is initially modified by the effects of the function call on the 1 or 2 rows selected in lines 1288 and 1289 of the same file (which are marked as `test_inputs` & `test_indices` respectively in lines 1288 and 1289. This happens out of the control of the user (which for example can't reset the state of the counter before continuing the execution,) provoking in the end an undesired side effect in the results obtained.
In my case, the resulting dataset -despite of the counter results are ok- lacks an instance of the classes 0 and 1 (which happen to be the classes of the first two examples of my dataset.) The rest of the classes I have in my dataset, contain the right number of examples as they were not affected by the effects of `does_function_return_dict` call.
I've debugged my code extensively and made a workaround myself hardcoding the necessary stuff (basically putting `update_data=True` in line 1290,) and then I obtain the results I expected without the side effect.
Is there a way to avoid that call to `does_function_return_dict` in map()'s line 1290 ? (e.g. extracting the required information that `does_function_return_dict` returns without making the testing calls to the user function on dataset rows 0 & 1)
Thanks in advance,
Francisco Perez-Sorrosal
| 1,940 |
https://github.com/huggingface/datasets/issues/1939 | [firewalled env] OFFLINE mode | [
"Thanks for reporting and for all the details and suggestions.\r\n\r\nI'm totally in favor of having a HF_DATASETS_OFFLINE env variable to disable manually all the connection checks, remove retries etc.\r\n\r\nMoreover you may know that the use case that you are mentioning is already supported from `datasets` 1.3.0... | This issue comes from a need to be able to run `datasets` in a firewalled env, which currently makes the software hang until it times out, as it's unable to complete the network calls.
I propose the following approach to solving this problem, using the example of `run_seq2seq.py` as a sample program. There are 2 possible ways to going about it.
## 1. Manual
manually prepare data and metrics files, that is transfer to the firewalled instance the dataset and the metrics and run:
```
DATASETS_OFFLINE=1 run_seq2seq.py --train_file xyz.csv --validation_file xyz.csv ...
```
`datasets` must not make any network calls and if there is a logic to do that and something is missing it should assert that this or that action requires network and therefore it can't proceed.
## 2. Automatic
In some clouds one can prepare a datastorage ahead of time with a normal networked environment but which doesn't have gpus and then one switches to the gpu instance which is firewalled, but it can access all the cached data. This is the ideal situation, since in this scenario we don't have to do anything manually, but simply run the same application twice:
1. on the non-firewalled instance:
```
run_seq2seq.py --dataset_name wmt16 --dataset_config ro-en ...
```
which should download and cached everything.
2. and then immediately after on the firewalled instance, which shares the same filesystem
```
DATASETS_OFFLINE=1 run_seq2seq.py --dataset_name wmt16 --dataset_config ro-en ...
```
and the metrics and datasets should be cached by the invocation number 1 and any network calls be skipped and if the logic is missing data it should assert and not try to fetch any data from online.
## Common Issues
1. for example currently `datasets` tries to look up online datasets if the files contain json or csv, despite the paths already provided
```
if dataset and path in _PACKAGED_DATASETS_MODULES:
```
2. it has an issue with metrics. e.g. I had to manually copy `rouge/rouge.py` from the `datasets` repo to the current dir - or it was hanging.
I had to comment out `head_hf_s3(...)` calls to make things work. So all those `try: head_hf_s3(...)` shouldn't be tried with `DATASETS_OFFLINE=1`
Here is the corresponding issue for `transformers`: https://github.com/huggingface/transformers/issues/10379
Thanks. | 1,939 |
https://github.com/huggingface/datasets/issues/1937 | CommonGen dataset page shows an error OSError: [Errno 28] No space left on device | [
"Facing the same issue for [Squad](https://huggingface.co/datasets/viewer/?dataset=squad) and [TriviaQA](https://huggingface.co/datasets/viewer/?dataset=trivia_qa) datasets as well.",
"We just fixed the issue, thanks for reporting !"
] | The page of the CommonGen data https://huggingface.co/datasets/viewer/?dataset=common_gen shows
| 1,937 |
https://github.com/huggingface/datasets/issues/1934 | Add Stanford Sentiment Treebank (SST) | [
"Dataset added in release [1.5.0](https://github.com/huggingface/datasets/releases/tag/1.5.0), I think I can close this."
] | I am going to add SST:
- **Name:** The Stanford Sentiment Treebank
- **Description:** The first corpus with fully labeled parse trees that allows for a complete analysis of the compositional effects of sentiment in language
- **Paper:** [Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank](https://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf)
- **Data:** https://nlp.stanford.edu/sentiment/index.html
- **Motivation:** Already requested in #353, SST is a popular dataset for Sentiment Classification
What's the difference with the [_SST-2_](https://huggingface.co/datasets/viewer/?dataset=glue&config=sst2) dataset included in GLUE? Essentially, SST-2 is a version of SST where:
- the labels were mapped from real numbers in [0.0, 1.0] to a binary label: {0, 1}
- the labels of the *sub-sentences* were included only in the training set
- the labels in the test set are obfuscated
So there is a lot more information in the original SST. The tricky bit is, the data is scattered into many text files and, for one in particular, I couldn't find the original encoding ([*but I'm not the only one*](https://groups.google.com/g/word2vec-toolkit/c/QIUjLw6RqFk/m/_iEeyt428wkJ) 🎵). The only solution I found was to manually replace all the è, ë, ç and so on into an `utf-8` copy of the text file. I uploaded the result in my Dropbox and I am using that as the main repo for the dataset.
Also, the _sub-sentences_ are built at run-time from the information encoded in several text files, so generating the examples is a bit more cumbersome than usual. Luckily, the dataset is not enormous.
I plan to divide the dataset in 2 configs: one with just whole sentences with their labels, the other with sentences _and their sub-sentences_ with their labels. Each config will be split in train, validation and test. Hopefully this makes sense, we may discuss it in the PR I'm going to submit.
| 1,934 |
https://github.com/huggingface/datasets/issues/1924 | Anonymous Dataset Addition (i.e Anonymous PR?) | [
"Hi !\r\nI guess you can add a dataset without the fields that must be kept anonymous, and then update those when the anonymity period is over.\r\nYou can also make the PR from an anonymous org.\r\nPinging @yjernite just to make sure it's ok",
"Hello,\r\nI would prefer to do the reverse: adding a link to an anony... | Hello,
Thanks a lot for your librairy.
We plan to submit a paper on OpenReview using the Anonymous setting. Is it possible to add a new dataset without breaking the anonimity, with a link to the paper ?
Cheers
@eusip | 1,924 |
https://github.com/huggingface/datasets/issues/1922 | How to update the "wino_bias" dataset | [
"Hi @JieyuZhao !\r\n\r\nYou can edit the dataset card of wino_bias to update the URL via a Pull Request. This would be really appreciated :)\r\n\r\nThe dataset card is the README.md file you can find at https://github.com/huggingface/datasets/tree/master/datasets/wino_bias\r\nAlso the homepage url is also mentioned... | Hi all,
Thanks for the efforts to collect all the datasets! But I think there is a problem with the wino_bias dataset. The current link is not correct. How can I update that?
Thanks! | 1,922 |
https://github.com/huggingface/datasets/issues/1919 | Failure to save with save_to_disk | [
"Hi thanks for reporting and for proposing a fix :)\r\n\r\nI just merged a fix, feel free to try it from the master branch !",
"Closing since this has been fixed by #1923"
] | When I try to save a dataset locally using the `save_to_disk` method I get the error:
```bash
FileNotFoundError: [Errno 2] No such file or directory: '/content/squad/train/squad-train.arrow'
```
To replicate:
1. Install `datasets` from master
2. Run this code:
```python
from datasets import load_dataset
squad = load_dataset("squad") # or any other dataset
squad.save_to_disk("squad") # error here
```
The problem is that the method is not creating a directory with the name `dataset_path` for saving the dataset in (i.e. it's not creating the *train* and *validation* directories in this case). After creating the directory the problem resolves.
I'll open a PR soon doing that and linking this issue.
| 1,919 |
https://github.com/huggingface/datasets/issues/1917 | UnicodeDecodeError: windows 10 machine | [
"upgraded to php 3.9.2 and it works!"
] | Windows 10
Php 3.6.8
when running
```
import datasets
oscar_am = datasets.load_dataset("oscar", "unshuffled_deduplicated_am")
print(oscar_am["train"][0])
```
I get the following error
```
file "C:\PYTHON\3.6.8\lib\encodings\cp1252.py", line 23, in decode
return codecs.charmap_decode(input,self.errors,decoding_table)[0]
UnicodeDecodeError: 'charmap' codec can't decode byte 0x9d in position 58: character maps to <undefined>
``` | 1,917 |
https://github.com/huggingface/datasets/issues/1915 | Unable to download `wiki_dpr` | [
"Thanks for reporting ! This is a bug. For now feel free to set `ignore_verifications=False` in `load_dataset`.\r\nI'm working on a fix",
"I just merged a fix :)\r\n\r\nWe'll do a patch release soon. In the meantime feel free to try it from the master branch\r\nThanks again for reporting !",
"Closing since this... | I am trying to download the `wiki_dpr` dataset. Specifically, I want to download `psgs_w100.multiset.no_index` with no embeddings/no index. In order to do so, I ran:
`curr_dataset = load_dataset("wiki_dpr", embeddings_name="multiset", index_name="no_index")`
However, I got the following error:
`datasets.utils.info_utils.UnexpectedDownloadedFile: {'embeddings_index'}`
I tried adding in flags `with_embeddings=False` and `with_index=False`:
`curr_dataset = load_dataset("wiki_dpr", with_embeddings=False, with_index=False, embeddings_name="multiset", index_name="no_index")`
But I got the following error:
`raise ExpectedMoreDownloadedFiles(str(set(expected_checksums) - set(recorded_checksums)))
datasets.utils.info_utils.ExpectedMoreDownloadedFiles: {‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_5’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_15’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_30’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_36’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_18’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_41’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_13’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_48’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_10’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_23’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_14’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_34’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_43’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_40’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_47’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_3’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_24’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_7’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_33’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_46’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_42’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_27’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_29’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_26’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_22’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_4’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_20’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_39’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_6’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_16’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_8’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_35’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_49’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_17’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_25’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_0’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_38’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_12’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_44’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_1’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_32’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_19’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_31’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_37’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_9’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_11’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_21’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_28’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_45’, ‘https://dl.fbaipublicfiles.com/rag/rag_multiset_embeddings/wiki_passages_2’}`
Is there anything else I need to set to download the dataset?
**UPDATE**: just running `curr_dataset = load_dataset("wiki_dpr", with_embeddings=False, with_index=False)` gives me the same error.
| 1,915 |
https://github.com/huggingface/datasets/issues/1911 | Saving processed dataset running infinitely | [
"@thomwolf @lhoestq can you guys please take a look and recommend some solution.",
"am suspicious of this thing? what's the purpose of this? pickling and unplickling\r\n`self = pickle.loads(pickle.dumps(self))`\r\n\r\n```\r\n def save_to_disk(self, dataset_path: str, fs=None):\r\n \"\"\"\r\n Save... | I have a text dataset of size 220M.
For pre-processing, I need to tokenize this and filter rows with the large sequence.
My tokenization took roughly 3hrs. I used map() with batch size 1024 and multi-process with 96 processes.
filter() function was way to slow, so I used a hack to use pyarrow filter table function, which is damm fast. Mentioned [here](https://github.com/huggingface/datasets/issues/1796)
```dataset._data = dataset._data.filter(...)```
It took 1 hr for the filter.
Then i use `save_to_disk()` on processed dataset and it is running forever.
I have been waiting since 8 hrs, it has not written a single byte.
Infact it has actually read from disk more than 100GB, screenshot below shows the stats using `iotop`.
Second process is the one.
<img width="1672" alt="Screenshot 2021-02-19 at 6 36 53 PM" src="https://user-images.githubusercontent.com/20911334/108508197-7325d780-72e1-11eb-8369-7c057d137d81.png">
I am not able to figure out, whether this is some issue with dataset library or that it is due to my hack for filter() function. | 1,911 |
https://github.com/huggingface/datasets/issues/1907 | DBPedia14 Dataset Checksum bug? | [
"Hi ! :)\r\n\r\nThis looks like the same issue as https://github.com/huggingface/datasets/issues/1856 \r\nBasically google drive has quota issues that makes it inconvenient for downloading files.\r\n\r\nIf the quota of a file is exceeded, you have to wait 24h for the quota to reset (which is painful).\r\n\r\nThe er... | Hi there!!!
I've been using successfully the DBPedia dataset (https://huggingface.co/datasets/dbpedia_14) with my codebase in the last couple of weeks, but in the last couple of days now I get this error:
```
Traceback (most recent call last):
File "./conditional_classification/basic_pipeline.py", line 178, in <module>
main()
File "./conditional_classification/basic_pipeline.py", line 128, in main
corpus.load_data(limit_train_examples_per_class=args.data_args.train_examples_per_class,
File "/home/fp/dev/conditional_classification/conditional_classification/datasets_base.py", line 83, in load_data
datasets = load_dataset(self.name, split=dataset_split)
File "/home/fp/anaconda3/envs/conditional/lib/python3.8/site-packages/datasets/load.py", line 609, in load_dataset
builder_instance.download_and_prepare(
File "/home/fp/anaconda3/envs/conditional/lib/python3.8/site-packages/datasets/builder.py", line 526, in download_and_prepare
self._download_and_prepare(
File "/home/fp/anaconda3/envs/conditional/lib/python3.8/site-packages/datasets/builder.py", line 586, in _download_and_prepare
verify_checksums(
File "/home/fp/anaconda3/envs/conditional/lib/python3.8/site-packages/datasets/utils/info_utils.py", line 39, in verify_checksums
raise NonMatchingChecksumError(error_msg + str(bad_urls))
datasets.utils.info_utils.NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://drive.google.com/uc?export=download&id=0Bz8a_Dbh9QhbQ2Vic1kxMmZZQ1k']
```
I've seen this has happened before in other datasets as reported in #537.
I've tried clearing my cache and call again `load_dataset` but still is not working. My same codebase is successfully downloading and using other datasets (e.g. AGNews) without any problem, so I guess something has happened specifically to the DBPedia dataset in the last few days.
Can you please check if there's a problem with the checksums?
Or this is related to any other stuff? I've seen that the path in the cache for the dataset is `/home/fp/.cache/huggingface/datasets/d_bpedia14/dbpedia_14/2.0.0/a70413e39e7a716afd0e90c9e53cb053691f56f9ef5fe317bd07f2c368e8e897...` and includes `d_bpedia14` instead maybe of `dbpedia_14`. Was this maybe a bug introduced recently?
Thanks! | 1,907 |
https://github.com/huggingface/datasets/issues/1906 | Feature Request: Support for Pandas `Categorical` | [
"We already have a ClassLabel type that does this kind of mapping between the label ids (integers) and actual label values (strings).\r\n\r\nI wonder if actually we should use the DictionaryType from Arrow and the Categorical type from pandas for the `datasets` ClassLabel feature type.\r\nCurrently ClassLabel corre... | ```
from datasets import Dataset
import pandas as pd
import pyarrow
df = pd.DataFrame(pd.Series(["a", "b", "c", "a"], dtype="category"))
pyarrow.Table.from_pandas(df)
Dataset.from_pandas(df)
# Throws NotImplementedError
# TODO(thom) this will need access to the dictionary as well (for labels). I.e. to the py_table
```
I'm curious if https://github.com/huggingface/datasets/blob/master/src/datasets/features.py#L796 could be built out in a way similar to `Sequence`?
e.g. a `Map` class (or whatever name the maintainers might prefer) that can accept:
```
index_type = generate_from_arrow_type(pa_type.index_type)
value_type = generate_from_arrow_type(pa_type.value_type)
```
and then additional code points to modify:
- FeatureType: https://github.com/huggingface/datasets/blob/master/src/datasets/features.py#L694
- A branch to handle Map in get_nested_type: https://github.com/huggingface/datasets/blob/master/src/datasets/features.py#L719
- I don't quite understand what `encode_nested_example` does but perhaps a branch there? https://github.com/huggingface/datasets/blob/master/src/datasets/features.py#L755
- Similarly, I don't quite understand why `Sequence` is used this way in `generate_from_dict`, but perhaps a branch here? https://github.com/huggingface/datasets/blob/master/src/datasets/features.py#L775
I couldn't find other usages of `Sequence` outside of defining specific datasets, so I'm not sure if that's a comprehensive set of touchpoints. | 1,906 |
https://github.com/huggingface/datasets/issues/1898 | ALT dataset has repeating instances in all splits | [
"Thanks for reporting. This looks like a very bad issue. I'm looking into it",
"I just merged a fix, we'll do a patch release soon. Thanks again for reporting, and sorry for the inconvenience.\r\nIn the meantime you can load `ALT` using `datasets` from the master branch",
"Thanks!!! works perfectly in the blead... | The [ALT](https://huggingface.co/datasets/alt) dataset has all the same instances within each split :/
Seemed like a great dataset for some experiments I wanted to carry out, especially since its medium-sized, and has all splits.
Would be great if this could be fixed :)
Added a snapshot of the contents from `explore-datset` feature, for quick reference.
| 1,898 |
https://github.com/huggingface/datasets/issues/1895 | Bug Report: timestamp[ns] not recognized | [
"Thanks for reporting !\r\n\r\nYou're right, `string_to_arrow` should be able to take `\"timestamp[ns]\"` as input and return the right pyarrow timestamp type.\r\nFeel free to suggest a fix for `string_to_arrow` and open a PR if you want to contribute ! This would be very appreciated :)\r\n\r\nTo give you more cont... | Repro:
```
from datasets import Dataset
import pandas as pd
import pyarrow
df = pd.DataFrame(pd.date_range("2018-01-01", periods=3, freq="H"))
pyarrow.Table.from_pandas(df)
Dataset.from_pandas(df)
# Throws ValueError: Neither timestamp[ns] nor timestamp[ns]_ seems to be a pyarrow data type.
```
The factory function seems to be just "timestamp": https://arrow.apache.org/docs/python/generated/pyarrow.timestamp.html#pyarrow.timestamp
It seems like https://github.com/huggingface/datasets/blob/master/src/datasets/features.py#L36-L43 could have a little bit of additional structure for handling these cases? I'd be happy to take a shot at opening a PR if I could receive some guidance on whether parsing something like `timestamp[ns]` and resolving it to timestamp('ns') is the goal of this method.
Alternatively, if I'm using this incorrectly (e.g. is the expectation that we always provide a schema when timestamps are involved?), that would be very helpful to know as well!
```
$ pip list # only the relevant libraries/versions
datasets 1.2.1
pandas 1.0.3
pyarrow 3.0.0
``` | 1,895 |
https://github.com/huggingface/datasets/issues/1894 | benchmarking against MMapIndexedDataset | [
"Hi sam !\r\nIndeed we can expect the performances to be very close since both MMapIndexedDataset and the `datasets` implem use memory mapping. With memory mapping what determines the I/O performance is the speed of your hard drive/SSD.\r\n\r\nIn terms of performance we're pretty close to the optimal speed for read... | I am trying to benchmark my datasets based implementation against fairseq's [`MMapIndexedDataset`](https://github.com/pytorch/fairseq/blob/master/fairseq/data/indexed_dataset.py#L365) and finding that, according to psrecord, my `datasets` implem uses about 3% more CPU memory and runs 1% slower for `wikitext103` (~1GB of tokens).
Questions:
1) Is this (basically identical) performance expected?
2) Is there a scenario where this library will outperform `MMapIndexedDataset`? (maybe more examples/larger examples?)
3) Should I be using different benchmarking tools than `psrecord`/how do you guys do benchmarks?
Thanks in advance! Sam | 1,894 |
https://github.com/huggingface/datasets/issues/1893 | wmt19 is broken | [
"This was also mentioned in https://github.com/huggingface/datasets/issues/488 \r\n\r\nThe bucket where is data was stored seems to be unavailable now. Maybe we can change the URL to the ones in https://conferences.unite.un.org/uncorpus/en/downloadoverview ?",
"Closing since this has been fixed by #1912"
] | 1. Check which lang pairs we have: `--dataset_name wmt19`:
Please pick one among the available configs: ['cs-en', 'de-en', 'fi-en', 'gu-en', 'kk-en', 'lt-en', 'ru-en', 'zh-en', 'fr-de']
2. OK, let's pick `ru-en`:
`--dataset_name wmt19 --dataset_config "ru-en"`
no cookies:
```
Traceback (most recent call last):
File "./run_seq2seq.py", line 661, in <module>
main()
File "./run_seq2seq.py", line 317, in main
datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/load.py", line 740, in load_dataset
builder_instance.download_and_prepare(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/builder.py", line 572, in download_and_prepare
self._download_and_prepare(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/builder.py", line 628, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/home/stas/.cache/huggingface/modules/datasets_modules/datasets/wmt19/436092de5f3faaf0fc28bc84875475b384e90a5470fa6afaee11039ceddc5052/wmt_utils.py", line 755, in _split_generators
downloaded_files = dl_manager.download_and_extract(urls_to_download)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/download_manager.py", line 276, in download_and_extract
return self.extract(self.download(url_or_urls))
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/download_manager.py", line 191, in download
downloaded_path_or_paths = map_nested(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/py_utils.py", line 233, in map_nested
mapped = [
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/py_utils.py", line 234, in <listcomp>
_single_map_nested((function, obj, types, None, True)) for obj in tqdm(iterable, disable=disable_tqdm)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/py_utils.py", line 190, in _single_map_nested
mapped = [_single_map_nested((function, v, types, None, True)) for v in pbar]
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/py_utils.py", line 190, in <listcomp>
mapped = [_single_map_nested((function, v, types, None, True)) for v in pbar]
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/py_utils.py", line 172, in _single_map_nested
return function(data_struct)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/download_manager.py", line 211, in _download
return cached_path(url_or_filename, download_config=download_config)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/file_utils.py", line 274, in cached_path
output_path = get_from_cache(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/file_utils.py", line 584, in get_from_cache
raise FileNotFoundError("Couldn't find file at {}".format(url))
FileNotFoundError: Couldn't find file at https://storage.googleapis.com/tfdataset-data/downloadataset/uncorpus/UNv1.0.en-ru.tar.gz
``` | 1,893 |
https://github.com/huggingface/datasets/issues/1892 | request to mirror wmt datasets, as they are really slow to download | [
"Yes that would be awesome. Not only the download speeds are awful, but also some files are missing.\r\nWe list all the URLs in the datasets/wmt19/wmt_utils.py so we can make a script to download them all and host on S3.\r\nAlso I think most of the materials are under the CC BY-NC-SA 3.0 license (must double check)... | Would it be possible to mirror the wmt data files under hf? Some of them take hours to download and not because of the local speed. They are all quite small datasets, just extremely slow to download.
Thank you! | 1,892 |
https://github.com/huggingface/datasets/issues/1891 | suggestion to improve a missing dataset error | [
"This is the current error thrown for missing datasets:\r\n```\r\nFileNotFoundError: Couldn't find a dataset script at C:\\Users\\Mario\\Desktop\\projects\\datasets\\missing_dataset\\missing_dataset.py or any data file in the same directory. Couldn't find 'missing_dataset' on the Hugging Face Hub either: FileNotFou... | I was using `--dataset_name wmt19` all was good. Then thought perhaps wmt20 is out, so I tried to use `--dataset_name wmt20`, got 3 different errors (1 repeated twice), none telling me the real issue - that `wmt20` isn't in the `datasets`:
```
True, predict_with_generate=True)
Traceback (most recent call last):
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/load.py", line 323, in prepare_module
local_path = cached_path(file_path, download_config=download_config)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/file_utils.py", line 274, in cached_path
output_path = get_from_cache(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/file_utils.py", line 584, in get_from_cache
raise FileNotFoundError("Couldn't find file at {}".format(url))
FileNotFoundError: Couldn't find file at https://raw.githubusercontent.com/huggingface/datasets/master/datasets/wmt20/wmt20.py
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/load.py", line 335, in prepare_module
local_path = cached_path(file_path, download_config=download_config)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/file_utils.py", line 274, in cached_path
output_path = get_from_cache(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/utils/file_utils.py", line 584, in get_from_cache
raise FileNotFoundError("Couldn't find file at {}".format(url))
FileNotFoundError: Couldn't find file at https://raw.githubusercontent.com/huggingface/datasets/master/datasets/wmt20/wmt20.py
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "./run_seq2seq.py", line 661, in <module>
main()
File "./run_seq2seq.py", line 317, in main
datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/load.py", line 706, in load_dataset
module_path, hash, resolved_file_path = prepare_module(
File "/mnt/nvme1/code/huggingface/datasets-master/src/datasets/load.py", line 343, in prepare_module
raise FileNotFoundError(
FileNotFoundError: Couldn't find file locally at wmt20/wmt20.py, or remotely at https://raw.githubusercontent.com/huggingface/datasets/master/datasets/wmt20/wmt20.py.
The file is also not present on the master branch on github.
```
Suggestion: if it is not in a local path, check that there is an actual `https://github.com/huggingface/datasets/tree/master/datasets/wmt20` first and assert "dataset `wmt20` doesn't exist in datasets", rather than trying to find a load script - since the whole repo is not there.
The error occured when running:
```
cd examples/seq2seq
export BS=16; rm -r output_dir; PYTHONPATH=../../src USE_TF=0 CUDA_VISIBLE_DEVICES=0 python ./run_seq2seq.py --model_name_or_path t5-small --output_dir output_dir --adam_eps 1e-06 --do_eval --evaluation_strategy=steps --label_smoothing 0.1 --learning_rate 3e-5 --logging_first_step --logging_steps 1000 --max_source_length 128 --max_target_length 128 --num_train_epochs 1 --overwrite_output_dir --per_device_eval_batch_size $BS --predict_with_generate --eval_steps 25000 --sortish_sampler --task translation_en_to_ro --val_max_target_length 128 --warmup_steps 500 --max_val_samples 500 --dataset_name wmt20 --dataset_config "ro-en" --source_prefix "translate English to Romanian: "
```
Thanks. | 1,891 |
https://github.com/huggingface/datasets/issues/1877 | Allow concatenation of both in-memory and on-disk datasets | [
"I started working on this. My idea is to first add the pyarrow Table wrappers InMemoryTable and MemoryMappedTable that both implement what's necessary regarding copy/pickle. Then have another wrapper that takes the concatenation of InMemoryTable/MemoryMappedTable objects.\r\n\r\nWhat's important here is that conca... | This is a prerequisite for the addition of the `add_item` feature (see #1870).
Currently there is one assumption that we would need to change: a dataset is either fully in memory (dataset._data_files is empty), or the dataset can be reloaded from disk (using the dataset._data_files).
This assumption is used for pickling for example:
- in-memory dataset can just be pickled/unpickled in-memory
- on-disk dataset can be unloaded to only keep the filepaths when pickling, and then reloaded from the disk when unpickling
Maybe let's have a design that allows a Dataset to have a Table that can be rebuilt from heterogenous sources like in-memory tables or on-disk tables ? This could also be further extended in the future
One idea would be to define a list of sources and each source implements a way to reload its corresponding pyarrow Table.
Then the dataset would be the concatenation of all these tables.
Depending on the source type, the serialization using pickle would be different. In-memory data would be copied while on-disk data would simply be replaced by the path to these data.
If you have some ideas you would like to share about the design/API feel free to do so :)
cc @albertvillanova | 1,877 |
https://github.com/huggingface/datasets/issues/1876 | load_dataset("multi_woz_v22") NonMatchingChecksumError | [
"Thanks for reporting !\r\nThis is due to the changes made in the data files in the multiwoz repo: https://github.com/budzianowski/multiwoz/pull/59\r\nI'm opening a PR to update the checksums of the data files.",
"I just merged the fix. It will be available in the new release of `datasets` later today.\r\nYou'll ... | Hi, it seems that loading the multi_woz_v22 dataset gives a NonMatchingChecksumError.
To reproduce:
`dataset = load_dataset('multi_woz_v22','v2.2_active_only',split='train')`
This will give the following error:
```
raise NonMatchingChecksumError(error_msg + str(bad_urls))
datasets.utils.info_utils.NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/dialog_acts.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_001.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_003.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_004.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_005.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_006.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_007.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_008.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_009.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_010.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_012.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_013.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_014.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_015.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_016.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/train/dialogues_017.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/dev/dialogues_001.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/dev/dialogues_002.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/test/dialogues_001.json', 'https://github.com/budzianowski/multiwoz/raw/master/data/MultiWOZ_2.2/test/dialogues_002.json']
```
| 1,876 |
https://github.com/huggingface/datasets/issues/1872 | Adding a new column to the dataset after set_format was called | [
"Hi ! Indeed if you add a column to a formatted dataset, then the new dataset gets a new formatting in which:\r\n```\r\nnew formatted columns = (all columns - previously unformatted columns)\r\n```\r\nTherefore the new column is going to be formatted using the `torch` formatting.\r\n\r\nIf you want your new column ... | Hi,
thanks for the nice library. I'm in the process of creating a custom dataset, which has a mix of tensors and lists of strings. I stumbled upon an error and want to know if its a problem on my side.
I load some lists of strings and integers, then call `data.set_format("torch", columns=["some_integer_column1", "some_integer_column2"], output_all_columns=True)`. This converts the integer columns into tensors, but keeps the lists of strings as they are. I then call `map` to add a new column to my dataset, which is a **list of strings**. Once I iterate through my dataset, I get an error that the new column can't be converted into a tensor (which is probably caused by `set_format`).
Below some pseudo code:
```python
def augment_func(sample: Dict) -> Dict:
# do something
return {
"some_integer_column1" : augmented_data["some_integer_column1"], # <-- tensor
"some_integer_column2" : augmented_data["some_integer_column2"], # <-- tensor
"NEW_COLUMN": targets, # <-- list of strings
}
data = datasets.load_dataset(__file__, data_dir="...", split="train")
data.set_format("torch", columns=["some_integer_column1", "some_integer_column2"], output_all_columns=True)
augmented_dataset = data.map(augment_func, batched=False)
for sample in augmented_dataset:
print(sample) # fails
```
and the exception:
```python
Traceback (most recent call last):
File "dataset.py", line 487, in <module>
main()
File "dataset.py", line 471, in main
for sample in augmented_dataset:
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 697, in __iter__
yield self._getitem(
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 1069, in _getitem
outputs = self._convert_outputs(
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 890, in _convert_outputs
v = map_nested(command, v, **map_nested_kwargs)
File "lib/python3.8/site-packages/datasets/utils/py_utils.py", line 225, in map_nested
return function(data_struct)
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 850, in command
return [map_nested(command, i, **map_nested_kwargs) for i in x]
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 850, in <listcomp>
return [map_nested(command, i, **map_nested_kwargs) for i in x]
File "lib/python3.8/site-packages/datasets/utils/py_utils.py", line 225, in map_nested
return function(data_struct)
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 850, in command
return [map_nested(command, i, **map_nested_kwargs) for i in x]
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 850, in <listcomp>
return [map_nested(command, i, **map_nested_kwargs) for i in x]
File "lib/python3.8/site-packages/datasets/utils/py_utils.py", line 225, in map_nested
return function(data_struct)
File "lib/python3.8/site-packages/datasets/arrow_dataset.py", line 851, in command
return torch.tensor(x, **format_kwargs)
TypeError: new(): invalid data type 'str'
```
Thanks!
| 1,872 |
https://github.com/huggingface/datasets/issues/1867 | ERROR WHEN USING SET_TRANSFORM() | [
"Hi @alejandrocros it looks like an incompatibility with the current Trainer @sgugger \r\nIndeed currently the Trainer of `transformers` doesn't support a dataset with a transform\r\n\r\nIt looks like it comes from this line: https://github.com/huggingface/transformers/blob/f51188cbe74195c14c5b3e2e8f10c2f435f9751a/... | Hi, I'm trying to use dataset.set_transform(encode) as @lhoestq told me in this issue: https://github.com/huggingface/datasets/issues/1825#issuecomment-774202797
However, when I try to use Trainer from transformers with such dataset, it throws an error:
```
TypeError: __init__() missing 1 required positional argument: 'transform'
[INFO|trainer.py:357] 2021-02-12 10:18:09,893 >> The following columns in the training set don't have a corresponding argument in `AlbertForMaskedLM.forward` and have been ignored: text.
Exception in device=TPU:0: __init__() missing 1 required positional argument: 'transform'
Traceback (most recent call last):
File "/anaconda3/envs/torch-xla-1.7/lib/python3.6/site-packages/torch_xla/distributed/xla_multiprocessing.py", line 330, in _mp_start_fn
_start_fn(index, pf_cfg, fn, args)
File "/anaconda3/envs/torch-xla-1.7/lib/python3.6/site-packages/torch_xla/distributed/xla_multiprocessing.py", line 324, in _start_fn
fn(gindex, *args)
File "/home/alejandro_vaca/transformers/examples/language-modeling/run_mlm_wwm.py", line 368, in _mp_fn
main()
File "/home/alejandro_vaca/transformers/examples/language-modeling/run_mlm_wwm.py", line 332, in main
data_collator=data_collator,
File "/anaconda3/envs/torch-xla-1.7/lib/python3.6/site-packages/transformers/trainer.py", line 286, in __init__
self._remove_unused_columns(self.train_dataset, description="training")
File "/anaconda3/envs/torch-xla-1.7/lib/python3.6/site-packages/transformers/trainer.py", line 359, in _remove_unused_columns
dataset.set_format(type=dataset.format["type"], columns=columns)
File "/home/alejandro_vaca/datasets/src/datasets/fingerprint.py", line 312, in wrapper
out = func(self, *args, **kwargs)
File "/home/alejandro_vaca/datasets/src/datasets/arrow_dataset.py", line 818, in set_format
_ = get_formatter(type, **format_kwargs)
File "/home/alejandro_vaca/datasets/src/datasets/formatting/__init__.py", line 112, in get_formatter
return _FORMAT_TYPES[format_type](**format_kwargs)
TypeError: __init__() missing 1 required positional argument: 'transform'
```
The code I'm using:
```{python}
def tokenize_function(examples):
# Remove empty lines
examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
return tokenizer(examples["text"], padding=padding, truncation=True, max_length=data_args.max_seq_length)
datasets.set_transform(tokenize_function)
data_collator = DataCollatorForWholeWordMask(tokenizer=tokenizer, mlm_probability=data_args.mlm_probability)
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=datasets["train"] if training_args.do_train else None,
eval_dataset=datasets["val"] if training_args.do_eval else None,
tokenizer=tokenizer,
data_collator=data_collator,
)
```
I've installed from source, master branch.
| 1,867 |
https://github.com/huggingface/datasets/issues/1864 | Add Winogender Schemas | [
"Nevermind, this one is already available on the hub under the name `'wino_bias'`: https://huggingface.co/datasets/wino_bias"
] | ## Adding a Dataset
- **Name:** Winogender Schemas
- **Description:** Winogender Schemas (inspired by Winograd Schemas) are minimal pairs of sentences that differ only by the gender of one pronoun in the sentence, designed to test for the presence of gender bias in automated coreference resolution systems.
- **Paper:** https://arxiv.org/abs/1804.09301
- **Data:** https://github.com/rudinger/winogender-schemas (see data directory)
- **Motivation:** Testing gender bias in automated coreference resolution systems, improve coreference resolution in general.
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
| 1,864 |
https://github.com/huggingface/datasets/issues/1863 | Add WikiCREM | [
"Hi @NielsRogge I would like to work on this dataset.\r\n\r\nThanks!",
"Hi @udapy, are you working on this?"
] | ## Adding a Dataset
- **Name:** WikiCREM
- **Description:** A large unsupervised corpus for coreference resolution.
- **Paper:** https://arxiv.org/abs/1905.06290
- **Github repo:**: https://github.com/vid-koci/bert-commonsense
- **Data:** https://ora.ox.ac.uk/objects/uuid:c83e94bb-7584-41a1-aef9-85b0e764d9e3
- **Motivation:** Coreference resolution, common sense reasoning
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
| 1,863 |
https://github.com/huggingface/datasets/issues/1859 | Error "in void don't know how to serialize this type of index" when saving index to disk when device=0 (GPU) | [
"Hi @corticalstack ! Thanks for reporting. Indeed in the recent versions of Faiss we must use `getDevice` to check if the index in on GPU.\r\n\r\nI'm opening a PR",
"I fixed this issue. It should work fine now.\r\nFeel free to try it out by installing `datasets` from source.\r\nOtherwise you can wait for the next... | Error serializing faiss index. Error as follows:
`Error in void faiss::write_index(const faiss::Index*, faiss::IOWriter*) at /home/conda/feedstock_root/build_artifacts/faiss-split_1612472484670/work/faiss/impl/index_write.cpp:453: don't know how to serialize this type of index`
Note:
`torch.cuda.is_available()` reports:
```
Cuda is available
cuda:0
```
Adding index, device=0 for GPU.
`dataset.add_faiss_index(column='embeddings', index_name='idx_embeddings', device=0)`
However, during a quick debug, self.faiss_index has no attr "device" when checked in` search.py, method save`, so fails to transform gpu index to cpu index. If I add index without device, index is saved OK.
```
def save(self, file: str):
"""Serialize the FaissIndex on disk"""
import faiss # noqa: F811
if (
hasattr(self.faiss_index, "device")
and self.faiss_index.device is not None
and self.faiss_index.device > -1
):
index = faiss.index_gpu_to_cpu(self.faiss_index)
else:
index = self.faiss_index
faiss.write_index(index, file)
```
| 1,859 |
https://github.com/huggingface/datasets/issues/1857 | Unable to upload "community provided" dataset - 400 Client Error | [
"Hi ! We're in the process of switching the community datasets to git repos, exactly like what we're doing for models.\r\nYou can find an example here:\r\nhttps://huggingface.co/datasets/lhoestq/custom_squad/tree/main\r\n\r\nWe'll update the CLI in the coming days and do a new release :)\r\n\r\nAlso cc @julien-c ma... | Hi,
i'm trying to a upload a dataset as described [here](https://huggingface.co/docs/datasets/v1.2.0/share_dataset.html#sharing-a-community-provided-dataset). This is what happens:
```
$ datasets-cli login
$ datasets-cli upload_dataset my_dataset
About to upload file /path/to/my_dataset/dataset_infos.json to S3 under filename my_dataset/dataset_infos.json and namespace username
About to upload file /path/to/my_dataset/my_dataset.py to S3 under filename my_dataset/my_dataset.py and namespace username
Proceed? [Y/n] Y
Uploading... This might take a while if files are large
400 Client Error: Bad Request for url: https://huggingface.co/api/datasets/presign
huggingface.co migrated to a new model hosting system.
You need to upgrade to transformers v3.5+ to upload new models.
More info at https://discuss.hugginface.co or https://twitter.com/julien_c. Thank you!
```
I'm using the latest releases of datasets and transformers. | 1,857 |
https://github.com/huggingface/datasets/issues/1856 | load_dataset("amazon_polarity") NonMatchingChecksumError | [
"Hi ! This issue may be related to #996 \r\nThis comes probably from the Quota Exceeded error from Google Drive.\r\nCan you try again tomorrow and see if you still have the error ?\r\n\r\nOn my side I didn't get any error today with `load_dataset(\"amazon_polarity\")`",
"+1 encountering this issue as well",
"@l... | Hi, it seems that loading the amazon_polarity dataset gives a NonMatchingChecksumError.
To reproduce:
```
load_dataset("amazon_polarity")
```
This will give the following error:
```
---------------------------------------------------------------------------
NonMatchingChecksumError Traceback (most recent call last)
<ipython-input-3-8559a03fe0f8> in <module>()
----> 1 dataset = load_dataset("amazon_polarity")
3 frames
/usr/local/lib/python3.6/dist-packages/datasets/utils/info_utils.py in verify_checksums(expected_checksums, recorded_checksums, verification_name)
37 if len(bad_urls) > 0:
38 error_msg = "Checksums didn't match" + for_verification_name + ":\n"
---> 39 raise NonMatchingChecksumError(error_msg + str(bad_urls))
40 logger.info("All the checksums matched successfully" + for_verification_name)
41
NonMatchingChecksumError: Checksums didn't match for dataset source files:
['https://drive.google.com/u/0/uc?id=0Bz8a_Dbh9QhbaW12WVVZS2drcnM&export=download']
``` | 1,856 |
https://github.com/huggingface/datasets/issues/1854 | Feature Request: Dataset.add_item | [
"Hi @sshleifer.\r\n\r\nI am not sure of understanding the need of the `add_item` approach...\r\n\r\nBy just reading your \"Desired API\" section, I would say you could (nearly) get it with a 1-column Dataset:\r\n```python\r\ndata = {\"input_ids\": [np.array([4,4,2]), np.array([8,6,5,5,2]), np.array([3,3,31,5])]}\r\... | I'm trying to integrate `huggingface/datasets` functionality into `fairseq`, which requires (afaict) being able to build a dataset through an `add_item` method, such as https://github.com/pytorch/fairseq/blob/master/fairseq/data/indexed_dataset.py#L318, as opposed to loading all the text into arrow, and then `dataset.map(binarizer)`.
Is this possible at the moment? Is there an example? I'm happy to use raw `pa.Table` but not sure whether it will support uneven length entries.
### Desired API
```python
import numpy as np
tokenized: List[np.NDArray[np.int64]] = [np.array([4,4,2]), np.array([8,6,5,5,2]), np.array([3,3,31,5])
def build_dataset_from_tokenized(tokenized: List[np.NDArray[int]]) -> Dataset:
"""FIXME"""
dataset = EmptyDataset()
for t in tokenized: dataset.append(t)
return dataset
ds = build_dataset_from_tokenized(tokenized)
assert (ds[0] == np.array([4,4,2])).all()
```
### What I tried
grep, google for "add one entry at a time", "datasets.append"
### Current Code
This code achieves the same result but doesn't fit into the `add_item` abstraction.
```python
dataset = load_dataset('text', data_files={'train': 'train.txt'})
tokenizer = RobertaTokenizerFast.from_pretrained('roberta-base', max_length=4096)
def tokenize_function(examples):
ids = tokenizer(examples['text'], return_attention_mask=False)['input_ids']
return {'input_ids': [x[1:] for x in ids]}
ds = dataset.map(tokenize_function, batched=True, num_proc=4, remove_columns=['text'], load_from_cache_file=not overwrite_cache)
print(ds['train'][0]) => np array
```
Thanks in advance! | 1,854 |
https://github.com/huggingface/datasets/issues/1849 | Add TIMIT | [
"@patrickvonplaten Could you please help me with how the output text has to be represented in the data? TIMIT has Words, Phonemes and texts. Also has lot on info on the speaker and the dialect. Could you please help me? An example of how to arrange it would be super helpful!\r\n\r\n",
"Hey @vrindaprabhu - sure I'... | ## Adding a Dataset
- **Name:** *TIMIT*
- **Description:** *The TIMIT corpus of read speech has been designed to provide speech data for the acquisition of acoustic-phonetic knowledge and for the development and evaluation of automatic speech recognition systems*
- **Paper:** *Homepage*: http://groups.inf.ed.ac.uk/ami/corpus/ / *Wikipedia*: https://en.wikipedia.org/wiki/TIMIT
- **Data:** *https://deepai.org/dataset/timit*
- **Motivation:** Important speech dataset
If interested in tackling this issue, feel free to tag @patrickvonplaten
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
| 1,849 |
https://github.com/huggingface/datasets/issues/1844 | Update Open Subtitles corpus with original sentence IDs | [
"Hi ! You're right this can can useful.\r\nThis should be easy to add, so feel free to give it a try if you want to contribute :)\r\nI think we just need to add it to the _generate_examples method of the OpenSubtitles dataset builder [here](https://github.com/huggingface/datasets/blob/master/datasets/open_subtitles... | Hi! It would be great if you could add the original sentence ids to [Open Subtitles](https://huggingface.co/datasets/open_subtitles).
I can think of two reasons: first, it's possible to gather sentences for an entire document (the original ids contain media id, subtitle file id and sentence id), therefore somewhat allowing for document-level machine translation (and other document-level stuff which could be cool to have); second, it's possible to have parallel sentences in multiple languages, as they share the same ids across bitexts.
I think I should tag @abhishekkrthakur as he's the one who added it in the first place.
Thanks! | 1,844 |
https://github.com/huggingface/datasets/issues/1843 | MustC Speech Translation | [
"Hi @patrickvonplaten I would like to work on this dataset. \r\n\r\nThanks! ",
"That's awesome! Actually, I just noticed that this dataset might become a bit too big!\r\n\r\nMuST-C is the main dataset used for IWSLT19 and should probably be added as a standalone dataset. Would you be interested also in adding `d... | ## Adding a Dataset
- **Name:** *IWSLT19*
- **Description:** *The Speech Translation Task addresses the translation of English audio into German and Portuguese text.*
- **Hompage:** *https://sites.google.com/view/iwslt-evaluation-2019/speech-translation*
- **Data:** *https://sites.google.com/view/iwslt-evaluation-2019/speech-translation* - all data under "Allowed Training Data" and "Development and Evalutaion Data for TED/How2"
- **Motivation:** Important speech dataset
If interested in tackling this issue, feel free to tag @patrickvonplaten
Instructions to add a new dataset can be found [here](https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md).
| 1,843 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.