id
int64 599M
3.26B
| number
int64 1
7.7k
| title
stringlengths 1
290
| body
stringlengths 0
228k
⌀ | state
stringclasses 2
values | html_url
stringlengths 46
51
| created_at
timestamp[s]date 2020-04-14 10:18:02
2025-07-23 08:04:53
| updated_at
timestamp[s]date 2020-04-27 16:04:17
2025-07-23 18:53:44
| closed_at
timestamp[s]date 2020-04-14 12:01:40
2025-07-23 16:44:42
⌀ | user
dict | labels
listlengths 0
4
| is_pull_request
bool 2
classes | comments
listlengths 0
0
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
2,635,813,932 | 7,279 |
Feature proposal: Stacking, potentially heterogeneous, datasets
|
### Introduction
Hello there,
I noticed that there are two ways to combine multiple datasets: Either through `datasets.concatenate_datasets` or `datasets.interleave_datasets`. However, to my knowledge (please correct me if I am wrong) both approaches require the datasets that are combined to have the same features.
I think it would be a great idea to add support for combining multiple datasets that might not follow the same schema (i.e. have different features), for example an image and text dataset. That is why I propose a third function of the `datasets.combine` module called `stack_datasets`, which can be used to combine a list of datasets with (potentially) different features. This would look as follows:
```python
>>> from datasets import stack_datasets
>>> image_dataset = ...
>>> next(iter(image_dataset))
{'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=555x416 at 0x313E79CD0> }
>>> text_dataset = ...
>>> next(iter(text_dataset))
{'text': "This is a test."}
>>> stacked = stack_datasets(datasets={'i_ds': image_dataset, 't_ds': text_dataset}, stopping_strategy='all_exhausted')
>>> next(iter(stacked))
{
'i_ds': {'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=555x416 at 0x313E79CD0> }
't_ds': {'text': "This is a test."}
}
```
<br />
### Motivation
I motivate this by:
**A**: The fact that Pytorch offers a similar functionality under `torch.utils.data.StackDataset` ([link](https://pytorch.org/docs/stable/data.html#torch.utils.data.StackDataset)).
**B**: In settings where one would like to e.g. train a Vision-Language model using an image-text dataset, an image dataset, and a text dataset, this functionality would offer a clean and intuitive solution to create multimodal datasets. I am aware that the aforementioned is also feasible without my proposed function, but I believe this offers a nice approach that aligns with existing functionality and is directly provided within the `datasets` package.
### API
`stack_datasets` has two arguments: `datasets` and `stopping_strategy `.
<br />
`datasets` is a dictionary of either type `Dict[str, Dataset]` or `Dict[str, IterableDatasets]`, a mixture is not allowed. It contains the names of the datasets (the keys) and the datasets themselves (the values) that should be stacked. Each item returned is a dictionary with one key-value pair for each dataset. The keys are the names of the datasets as provided in the argument `datasets`, and the values are the respective examples from the datasets.
<br />
`stopping_strategy` is the same as for `interleave_datasets`. If it is `first_exhausted` we stop if the smallest dataset runs out of examples, if it is `all_exhausted` we stop if all datasets ran out of examples at least once. For `all_exhausted` that means that we may visit examples from datasets multiple times.
### Docs
I saw that there are multiple documentations and guides on the HuggingFace website that introduce `concatenate_datasets` and `interleave_datasets`, for example [here](https://huggingface.co/docs/datasets/process#concatenate). If this request is merged I would be willing to add the new functionality at the appropriate points in the documentation (if desired).
### Tests
I also added some tests to ensure correctness. Some tests I wrote in [tests/test_iterable_dataset.py](https://github.com/TimCares/datasets/blob/fadc1159debf2a65d44e40cbf7758f2bd2cc8b08/tests/test_iterable_dataset.py#L2169)
run for both `Dataset` and `IterableDataset` even though tests for `Dataset` technically do not belong in this script, but I found that this was a nice way to cover more cases with mostly the same code.
### Additional information
I tried to write the code in a way so that it is similar to that of `concatenate_datasets` and `interleave_datasets`.
I’m open to feedback and willing to make adjustments based on your suggestions, so feel free to give me your take. :)
|
open
|
https://github.com/huggingface/datasets/pull/7279
| 2024-11-05T15:40:50 | 2024-11-05T15:40:50 | null |
{
"login": "TimCares",
"id": 96243987,
"type": "User"
}
|
[] | true |
[] |
2,633,436,151 | 7,278 |
Let soundfile directly read local audio files
|
- [x] Fixes #7276
|
open
|
https://github.com/huggingface/datasets/pull/7278
| 2024-11-04T17:41:13 | 2024-11-18T14:01:25 | null |
{
"login": "fawazahmed0",
"id": 20347013,
"type": "User"
}
|
[] | true |
[] |
2,632,459,184 | 7,277 |
Add link to video dataset
|
This PR updates https://huggingface.co/docs/datasets/loading to also link to the new video loading docs.
cc @mfarre
|
closed
|
https://github.com/huggingface/datasets/pull/7277
| 2024-11-04T10:45:12 | 2024-11-04T17:05:06 | 2024-11-04T17:05:06 |
{
"login": "NielsRogge",
"id": 48327001,
"type": "User"
}
|
[] | true |
[] |
2,631,917,431 | 7,276 |
Accessing audio dataset value throws Format not recognised error
|
### Describe the bug
Accessing audio dataset value throws `Format not recognised error`
### Steps to reproduce the bug
**code:**
```py
from datasets import load_dataset
dataset = load_dataset("fawazahmed0/bug-audio")
for data in dataset["train"]:
print(data)
```
**output:**
```bash
(mypy) C:\Users\Nawaz-Server\Documents\ml>python myest.py
[C:\vcpkg\buildtrees\mpg123\src\0d8db63f9b-3db975bc05.clean\src\libmpg123\layer3.c:INT123_do_layer3():1801] error: dequantization failed!
{'audio': {'path': 'C:\\Users\\Nawaz-Server\\.cache\\huggingface\\hub\\datasets--fawazahmed0--bug-audio\\snapshots\\fab1398431fed1c0a2a7bff0945465bab8b5daef\\data\\Ghamadi\\037135.mp3', 'array': array([ 0.00000000e+00, -2.86519935e-22, -2.56504911e-21, ...,
-1.94239747e-02, -2.42924765e-02, -2.99104657e-02]), 'sampling_rate': 22050}, 'reciter': 'Ghamadi', 'transcription': 'الا عجوز ا في الغبرين', 'line': 3923, 'chapter': 37, 'verse': 135, 'text': 'إِلَّا عَجُوزࣰ ا فِي ٱلۡغَٰبِرِينَ'}
Traceback (most recent call last):
File "C:\Users\Nawaz-Server\Documents\ml\myest.py", line 5, in <module>
for data in dataset["train"]:
~~~~~~~^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\arrow_dataset.py", line 2372, in __iter__
formatted_output = format_table(
^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\formatting\formatting.py", line 639, in format_table
return formatter(pa_table, query_type=query_type)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\formatting\formatting.py", line 403, in __call__
return self.format_row(pa_table)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\formatting\formatting.py", line 444, in format_row
row = self.python_features_decoder.decode_row(row)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\formatting\formatting.py", line 222, in decode_row
return self.features.decode_example(row) if self.features else row
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\features\features.py", line 2042, in decode_example
column_name: decode_nested_example(feature, value, token_per_repo_id=token_per_repo_id)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\features\features.py", line 1403, in decode_nested_example
return schema.decode_example(obj, token_per_repo_id=token_per_repo_id)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\datasets\features\audio.py", line 184, in decode_example
array, sampling_rate = sf.read(f)
^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\soundfile.py", line 285, in read
with SoundFile(file, 'r', samplerate, channels,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\soundfile.py", line 658, in __init__
self._file = self._open(file, mode_int, closefd)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Nawaz-Server\.conda\envs\mypy\Lib\site-packages\soundfile.py", line 1216, in _open
raise LibsndfileError(err, prefix="Error opening {0!r}: ".format(self.name))
soundfile.LibsndfileError: Error opening <_io.BufferedReader name='C:\\Users\\Nawaz-Server\\.cache\\huggingface\\hub\\datasets--fawazahmed0--bug-audio\\snapshots\\fab1398431fed1c0a2a7bff0945465bab8b5daef\\data\\Ghamadi\\037136.mp3'>: Format not recognised.
```
### Expected behavior
Everything should work fine, as loading the problematic audio file directly with soundfile package works fine
**code:**
```
import soundfile as sf
print(sf.read('C:\\Users\\Nawaz-Server\\.cache\\huggingface\\hub\\datasets--fawazahmed0--bug-audio\\snapshots\\fab1398431fed1c0a2a7bff0945465bab8b5daef\\data\\Ghamadi\\037136.mp3'))
```
**output:**
```bash
(mypy) C:\Users\Nawaz-Server\Documents\ml>python myest.py
[C:\vcpkg\buildtrees\mpg123\src\0d8db63f9b-3db975bc05.clean\src\libmpg123\layer3.c:INT123_do_layer3():1801] error: dequantization failed!
(array([ 0.00000000e+00, -8.43723821e-22, -2.45370628e-22, ...,
-7.71464454e-03, -6.90496899e-03, -8.63333419e-03]), 22050)
```
### Environment info
- `datasets` version: 3.0.2
- Platform: Windows-11-10.0.22621-SP0
- Python version: 3.12.7
- `huggingface_hub` version: 0.26.2
- PyArrow version: 17.0.0
- Pandas version: 2.2.3
- `fsspec` version: 2024.10.0
- soundfile: 0.12.1
|
open
|
https://github.com/huggingface/datasets/issues/7276
| 2024-11-04T05:59:13 | 2024-11-09T18:51:52 | null |
{
"login": "fawazahmed0",
"id": 20347013,
"type": "User"
}
|
[] | false |
[] |
2,631,713,397 | 7,275 |
load_dataset
|
### Describe the bug
I am performing two operations I see on a hugging face tutorial (Fine-tune a language model), and I am defining every aspect inside the mapped functions, also some imports of the library because it doesnt identify anything not defined outside that function where the dataset elements are being mapped:
https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb#scrollTo=iaAJy5Hu3l_B
`- lm_datasets = tokenized_datasets.map(
group_texts,
batched=True,
batch_size=batch_size,
num_proc=4,
)
- tokenized_datasets = datasets.map(tokenize_function, batched=True, num_proc=4, remove_columns=["text"])
def tokenize_function(examples):
model_checkpoint = 'gpt2'
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint, use_fast=True)
return tokenizer(examples["text"])`
### Steps to reproduce the bug
Currently handle all the imports inside the function
### Expected behavior
The code must work es expected in the notebook, but currently this is not happening.
https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb#scrollTo=iaAJy5Hu3l_B
### Environment info
print(transformers.__version__)
4.46.1
|
open
|
https://github.com/huggingface/datasets/issues/7275
| 2024-11-04T03:01:44 | 2024-11-04T03:01:44 | null |
{
"login": "santiagobp99",
"id": 46941974,
"type": "User"
}
|
[] | false |
[] |
2,629,882,821 | 7,274 |
[MINOR:TYPO] Fix typo in exception text
| null |
closed
|
https://github.com/huggingface/datasets/pull/7274
| 2024-11-01T21:15:29 | 2025-05-21T13:17:20 | 2025-05-21T13:17:20 |
{
"login": "cakiki",
"id": 3664563,
"type": "User"
}
|
[] | true |
[] |
2,628,896,492 | 7,273 |
Raise error for incorrect JSON serialization
|
Raise error when `lines = False` and `batch_size < Dataset.num_rows` in `Dataset.to_json()`.
Issue: #7037
Related PRs:
#7039 #7181
|
closed
|
https://github.com/huggingface/datasets/pull/7273
| 2024-11-01T11:54:35 | 2024-11-18T11:25:01 | 2024-11-18T11:25:01 |
{
"login": "varadhbhatnagar",
"id": 20443618,
"type": "User"
}
|
[] | true |
[] |
2,627,223,390 | 7,272 |
fix conda release worlflow
| null |
closed
|
https://github.com/huggingface/datasets/pull/7272
| 2024-10-31T15:56:19 | 2024-10-31T15:58:35 | 2024-10-31T15:57:29 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,627,135,540 | 7,271 |
Set dev version
| null |
closed
|
https://github.com/huggingface/datasets/pull/7271
| 2024-10-31T15:22:51 | 2024-10-31T15:25:27 | 2024-10-31T15:22:59 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,627,107,016 | 7,270 |
Release: 3.1.0
| null |
closed
|
https://github.com/huggingface/datasets/pull/7270
| 2024-10-31T15:10:01 | 2024-10-31T15:14:23 | 2024-10-31T15:14:20 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,626,873,843 | 7,269 |
Memory leak when streaming
|
### Describe the bug
I try to use a dataset with streaming=True, the issue I have is that the RAM usage becomes higher and higher until it is no longer sustainable.
I understand that huggingface store data in ram during the streaming, and more worker in dataloader there are, more a lot of shard will be stored in ram, but the issue I have is that the ram usage is not constant. So after each new shard loaded, the ram usage will be higher and higher.
### Steps to reproduce the bug
You can run this code and see you ram usage, after each shard of 255 examples, your ram usage will be extended.
```py
from datasets import load_dataset
from torch.utils.data import DataLoader
dataset = load_dataset("WaveGenAI/dataset", streaming=True)
dataloader = DataLoader(dataset["train"], num_workers=3)
for i, data in enumerate(dataloader):
print(i, end="\r")
```
### Expected behavior
The Ram usage should be always the same (just 3 shards loaded in the ram).
### Environment info
- `datasets` version: 3.0.1
- Platform: Linux-6.10.5-arch1-1-x86_64-with-glibc2.40
- Python version: 3.12.4
- `huggingface_hub` version: 0.26.0
- PyArrow version: 17.0.0
- Pandas version: 2.2.3
- `fsspec` version: 2024.6.1
|
open
|
https://github.com/huggingface/datasets/issues/7269
| 2024-10-31T13:33:52 | 2024-11-18T11:46:07 | null |
{
"login": "Jourdelune",
"id": 64205064,
"type": "User"
}
|
[] | false |
[] |
2,626,664,687 | 7,268 |
load_from_disk
|
### Describe the bug
I have data saved with save_to_disk. The data is big (700Gb). When I try loading it, the only option is load_from_disk, and this function copies the data to a tmp directory, causing me to run out of disk space. Is there an alternative solution to that?
### Steps to reproduce the bug
when trying to load data using load_From_disk after being saved using save_to_disk
### Expected behavior
run out of disk space
### Environment info
lateest version
|
open
|
https://github.com/huggingface/datasets/issues/7268
| 2024-10-31T11:51:56 | 2025-07-01T08:42:17 | null |
{
"login": "ghaith-mq",
"id": 71670961,
"type": "User"
}
|
[] | false |
[] |
2,626,490,029 | 7,267 |
Source installation fails on Macintosh with python 3.10
|
### Describe the bug
Hi,
Decord is a dev dependency not maintained since couple years.
It does not have an ARM package available rendering it uninstallable on non-intel based macs
Suggestion is to move to eva-decord (https://github.com/georgia-tech-db/eva-decord) which doesnt have this problem.
Happy to raise a PR
### Steps to reproduce the bug
Source installation as mentioned in contributinog.md
### Expected behavior
Installation without decord failing to be installed.
### Environment info
python=3.10, M3 Mac
|
open
|
https://github.com/huggingface/datasets/issues/7267
| 2024-10-31T10:18:45 | 2024-11-04T22:18:06 | null |
{
"login": "mayankagarwals",
"id": 39498938,
"type": "User"
}
|
[] | false |
[] |
2,624,666,087 | 7,266 |
The dataset viewer should be available soon. Please retry later.
|
### Describe the bug
After waiting for 2 hours, it still presents ``The dataset viewer should be available soon. Please retry later.''
### Steps to reproduce the bug
dataset link: https://huggingface.co/datasets/BryanW/HI_EDIT
### Expected behavior
Present the dataset viewer.
### Environment info
NA
|
closed
|
https://github.com/huggingface/datasets/issues/7266
| 2024-10-30T16:32:00 | 2024-10-31T03:48:11 | 2024-10-31T03:48:10 |
{
"login": "viiika",
"id": 39821659,
"type": "User"
}
|
[] | false |
[] |
2,624,090,418 | 7,265 |
Disallow video push_to_hub
| null |
closed
|
https://github.com/huggingface/datasets/pull/7265
| 2024-10-30T13:21:55 | 2024-10-30T13:36:05 | 2024-10-30T13:36:02 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,624,047,640 | 7,264 |
fix docs relative links
| null |
closed
|
https://github.com/huggingface/datasets/pull/7264
| 2024-10-30T13:07:34 | 2024-10-30T13:10:13 | 2024-10-30T13:09:02 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,621,844,054 | 7,263 |
Small addition to video docs
| null |
closed
|
https://github.com/huggingface/datasets/pull/7263
| 2024-10-29T16:58:37 | 2024-10-29T17:01:05 | 2024-10-29T16:59:10 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,620,879,059 | 7,262 |
Allow video with disabeld decoding without decord
|
for the viewer, this way it can use Video(decode=False) and doesn't need decord (which causes segfaults)
|
closed
|
https://github.com/huggingface/datasets/pull/7262
| 2024-10-29T10:54:04 | 2024-10-29T10:56:19 | 2024-10-29T10:55:37 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,620,510,840 | 7,261 |
Cannot load the cache when mapping the dataset
|
### Describe the bug
I'm training the flux controlnet. The train_dataset.map() takes long time to finish. However, when I killed one training process and want to restart a new training with the same dataset. I can't reuse the mapped result even I defined the cache dir for the dataset.
with accelerator.main_process_first():
from datasets.fingerprint import Hasher
# fingerprint used by the cache for the other processes to load the result
# details: https://github.com/huggingface/diffusers/pull/4038#discussion_r1266078401
new_fingerprint = Hasher.hash(args)
train_dataset = train_dataset.map(
compute_embeddings_fn, batched=True, new_fingerprint=new_fingerprint, batch_size=10,
)
### Steps to reproduce the bug
train flux controlnet and start again
### Expected behavior
will not map again
### Environment info
latest diffusers
|
open
|
https://github.com/huggingface/datasets/issues/7261
| 2024-10-29T08:29:40 | 2025-03-24T13:27:55 | null |
{
"login": "zhangn77",
"id": 43033959,
"type": "User"
}
|
[] | false |
[] |
2,620,014,285 | 7,260 |
cache can't cleaned or disabled
|
### Describe the bug
I tried following ways, the cache can't be disabled.
I got 2T data, but I also got more than 2T cache file. I got pressure on storage. I need to diable cache or cleaned immediately after processed. Following ways are all not working, please give some help!
```python
from datasets import disable_caching
from transformers import AutoTokenizer
disable_caching()
tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_path)
def tokenization_fn(examples):
column_name = 'text' if 'text' in examples else 'data'
tokenized_inputs = tokenizer(
examples[column_name], return_special_tokens_mask=True, truncation=False,
max_length=tokenizer.model_max_length
)
return tokenized_inputs
data = load_dataset('json', data_files=save_local_path, split='train', cache_dir=None)
data.cleanup_cache_files()
updated_dataset = data.map(tokenization_fn, load_from_cache_file=False)
updated_dataset .cleanup_cache_files()
```
### Expected behavior
no cache file generated
### Environment info
Ubuntu 20.04.6 LTS
datasets 3.0.2
|
open
|
https://github.com/huggingface/datasets/issues/7260
| 2024-10-29T03:15:28 | 2024-12-11T09:04:52 | null |
{
"login": "charliedream1",
"id": 15007828,
"type": "User"
}
|
[] | false |
[] |
2,618,909,241 | 7,259 |
Don't embed videos
|
don't include video bytes when running download_and_prepare(format="parquet")
this also affects push_to_hub which will just upload the local paths of the videos though
|
closed
|
https://github.com/huggingface/datasets/pull/7259
| 2024-10-28T16:25:10 | 2024-10-28T16:27:34 | 2024-10-28T16:26:01 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,618,758,399 | 7,258 |
Always set non-null writer batch size
|
bug introduced in #7230, it was preventing the Viewer limit writes to work
|
closed
|
https://github.com/huggingface/datasets/pull/7258
| 2024-10-28T15:26:14 | 2024-10-28T15:28:41 | 2024-10-28T15:26:29 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,618,602,173 | 7,257 |
fix ci for pyarrow 18
| null |
closed
|
https://github.com/huggingface/datasets/pull/7257
| 2024-10-28T14:31:34 | 2024-10-28T14:34:05 | 2024-10-28T14:31:44 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,618,580,188 | 7,256 |
Retry all requests timeouts
|
as reported in https://github.com/huggingface/datasets/issues/6843
|
closed
|
https://github.com/huggingface/datasets/pull/7256
| 2024-10-28T14:23:16 | 2024-10-28T14:56:28 | 2024-10-28T14:56:26 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,618,540,355 | 7,255 |
fix decord import
|
delay the import until Video() is instantiated + also import duckdb first (otherwise importing duckdb later causes a segfault)
|
closed
|
https://github.com/huggingface/datasets/pull/7255
| 2024-10-28T14:08:19 | 2024-10-28T14:10:43 | 2024-10-28T14:09:14 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,616,174,996 | 7,254 |
mismatch for datatypes when providing `Features` with `Array2D` and user specified `dtype` and using with_format("numpy")
|
### Describe the bug
If the user provides a `Features` type value to `datasets.Dataset` with members having `Array2D` with a value for `dtype`, it is not respected during `with_format("numpy")` which should return a `np.array` with `dtype` that the user provided for `Array2D`. It seems for floats, it will be set to `float32` and for ints it will be set to `int64`
### Steps to reproduce the bug
```python
import numpy as np
import datasets
from datasets import Dataset, Features, Array2D
print(f"datasets version: {datasets.__version__}")
data_info = {
"arr_float" : "float64",
"arr_int" : "int32"
}
sample = {key : [np.zeros([4, 5], dtype=dtype)] for key, dtype in data_info.items()}
features = {key : Array2D(shape=(None, 5), dtype=dtype) for key, dtype in data_info.items()}
features = Features(features)
dataset = Dataset.from_dict(sample, features=features)
ds = dataset.with_format("numpy")
for key in features:
print(f"{key} feature dtype: ", ds.features[key].dtype)
print(f"{key} dtype:", ds[key].dtype)
```
Output:
```bash
datasets version: 3.0.2
arr_float feature dtype: float64
arr_float dtype: float32
arr_int feature dtype: int32
arr_int dtype: int64
```
### Expected behavior
It should return a `np.array` with `dtype` that the user provided for the corresponding member in the `Features` type value
### Environment info
- `datasets` version: 3.0.2
- Platform: Linux-6.11.5-arch1-1-x86_64-with-glibc2.40
- Python version: 3.12.7
- `huggingface_hub` version: 0.26.1
- PyArrow version: 16.1.0
- Pandas version: 2.2.2
- `fsspec` version: 2024.5.0
|
open
|
https://github.com/huggingface/datasets/issues/7254
| 2024-10-26T22:06:27 | 2024-10-26T22:07:37 | null |
{
"login": "Akhil-CM",
"id": 97193607,
"type": "User"
}
|
[] | false |
[] |
2,615,862,202 | 7,253 |
Unable to upload a large dataset zip either from command line or UI
|
### Describe the bug
Unable to upload a large dataset zip from command line or UI. UI simply says error. I am trying to a upload a tar.gz file of 17GB.
<img width="550" alt="image" src="https://github.com/user-attachments/assets/f9d29024-06c8-49c4-a109-0492cff79d34">
<img width="755" alt="image" src="https://github.com/user-attachments/assets/a8d4acda-7f02-4279-9c2d-b2e0282b4faa">
### Steps to reproduce the bug
Upload a large file
### Expected behavior
The file should upload without any issue.
### Environment info
None
|
open
|
https://github.com/huggingface/datasets/issues/7253
| 2024-10-26T13:17:06 | 2024-10-26T13:17:06 | null |
{
"login": "vakyansh",
"id": 159609047,
"type": "User"
}
|
[] | false |
[] |
2,613,795,544 | 7,252 |
Add IterableDataset.shard()
|
Will be useful to distribute a dataset across workers (other than pytorch) like spark
I also renamed `.n_shards` -> `.num_shards` for consistency and kept the old name for backward compatibility. And a few changes in internal functions for consistency as well (rank, world_size -> num_shards, index)
Breaking change: the new default for `contiguous` in `Dataset.shard()` is `True`, but imo not a big deal since I couldn't find any usage of `contiguous=False` internally (we always do contiguous=True for map-style datasets since its more optimized) or in the wild
|
closed
|
https://github.com/huggingface/datasets/pull/7252
| 2024-10-25T11:07:12 | 2025-03-21T03:58:43 | 2024-10-25T15:45:22 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,612,097,435 | 7,251 |
Missing video docs
| null |
closed
|
https://github.com/huggingface/datasets/pull/7251
| 2024-10-24T16:45:12 | 2024-10-24T16:48:29 | 2024-10-24T16:48:27 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,612,041,969 | 7,250 |
Basic XML support (mostly copy pasted from text)
|
enable the viewer for datasets like https://huggingface.co/datasets/FrancophonIA/e-calm (there will be more and more apparently)
|
closed
|
https://github.com/huggingface/datasets/pull/7250
| 2024-10-24T16:14:50 | 2024-10-24T16:19:18 | 2024-10-24T16:19:16 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,610,136,636 | 7,249 |
How to debugging
|
### Describe the bug
I wanted to use my own script to handle the processing, and followed the tutorial documentation by rewriting the MyDatasetConfig and MyDatasetBuilder (which contains the _info,_split_generators and _generate_examples methods) classes. Testing with simple data was able to output the results of the processing, but when I wished to do more complex processing, I found that I was unable to debug (even the simple samples were inaccessible). There are no errors reported, and I am able to print the _info,_split_generators and _generate_examples messages, but I am unable to access the breakpoints.
### Steps to reproduce the bug
# my_dataset.py
import json
import datasets
class MyDatasetConfig(datasets.BuilderConfig):
def __init__(self, **kwargs):
super(MyDatasetConfig, self).__init__(**kwargs)
class MyDataset(datasets.GeneratorBasedBuilder):
VERSION = datasets.Version("1.0.0")
BUILDER_CONFIGS = [
MyDatasetConfig(
name="default",
version=VERSION,
description="myDATASET"
),
]
def _info(self):
print("info") # breakpoints
return datasets.DatasetInfo(
description="myDATASET",
features=datasets.Features(
{
"id": datasets.Value("int32"),
"text": datasets.Value("string"),
"label": datasets.ClassLabel(names=["negative", "positive"]),
}
),
supervised_keys=("text", "label"),
)
def _split_generators(self, dl_manager):
print("generate") # breakpoints
data_file = "data.json"
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_file}
),
]
def _generate_examples(self, filepath):
print("example") # breakpoints
with open(filepath, encoding="utf-8") as f:
data = json.load(f)
for idx, sample in enumerate(data):
yield idx, {
"id": sample["id"],
"text": sample["text"],
"label": sample["label"],
}
#main.py
import os
os.environ["TRANSFORMERS_NO_MULTIPROCESSING"] = "1"
from datasets import load_dataset
dataset = load_dataset("my_dataset.py", split="train", cache_dir=None)
print(dataset[:5])
### Expected behavior
Pause at breakpoints while running debugging
### Environment info
pycharm
|
open
|
https://github.com/huggingface/datasets/issues/7249
| 2024-10-24T01:03:51 | 2024-10-24T01:03:51 | null |
{
"login": "ShDdu",
"id": 49576595,
"type": "User"
}
|
[] | false |
[] |
2,609,926,089 | 7,248 |
ModuleNotFoundError: No module named 'datasets.tasks'
|
### Describe the bug
---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
[<ipython-input-9-13b5f31bd391>](https://bcb6shpazyu-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20241022-060119_RC00_688494744#) in <cell line: 1>()
----> 1 dataset = load_dataset('knowledgator/events_classification_biotech')
11 frames
[/usr/local/lib/python3.10/dist-packages/datasets/load.py](https://bcb6shpazyu-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20241022-060119_RC00_688494744#) in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, verification_mode, keep_in_memory, save_infos, revision, token, streaming, num_proc, storage_options, trust_remote_code, **config_kwargs)
2130
2131 # Create a dataset builder
-> 2132 builder_instance = load_dataset_builder(
2133 path=path,
2134 name=name,
[/usr/local/lib/python3.10/dist-packages/datasets/load.py](https://bcb6shpazyu-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20241022-060119_RC00_688494744#) in load_dataset_builder(path, name, data_dir, data_files, cache_dir, features, download_config, download_mode, revision, token, storage_options, trust_remote_code, _require_default_config_name, **config_kwargs)
1886 raise ValueError(error_msg)
1887
-> 1888 builder_cls = get_dataset_builder_class(dataset_module, dataset_name=dataset_name)
1889 # Instantiate the dataset builder
1890 builder_instance: DatasetBuilder = builder_cls(
[/usr/local/lib/python3.10/dist-packages/datasets/load.py](https://bcb6shpazyu-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20241022-060119_RC00_688494744#) in get_dataset_builder_class(dataset_module, dataset_name)
246 dataset_module.importable_file_path
247 ) if dataset_module.importable_file_path else nullcontext():
--> 248 builder_cls = import_main_class(dataset_module.module_path)
249 if dataset_module.builder_configs_parameters.builder_configs:
250 dataset_name = dataset_name or dataset_module.builder_kwargs.get("dataset_name")
[/usr/local/lib/python3.10/dist-packages/datasets/load.py](https://bcb6shpazyu-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20241022-060119_RC00_688494744#) in import_main_class(module_path)
167 def import_main_class(module_path) -> Optional[Type[DatasetBuilder]]:
168 """Import a module at module_path and return its main class: a DatasetBuilder"""
--> 169 module = importlib.import_module(module_path)
170 # Find the main class in our imported module
171 module_main_cls = None
[/usr/lib/python3.10/importlib/__init__.py](https://bcb6shpazyu-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20241022-060119_RC00_688494744#) in import_module(name, package)
124 break
125 level += 1
--> 126 return _bootstrap._gcd_import(name[level:], package, level)
127
128
/usr/lib/python3.10/importlib/_bootstrap.py in _gcd_import(name, package, level)
/usr/lib/python3.10/importlib/_bootstrap.py in _find_and_load(name, import_)
/usr/lib/python3.10/importlib/_bootstrap.py in _find_and_load_unlocked(name, import_)
/usr/lib/python3.10/importlib/_bootstrap.py in _load_unlocked(spec)
/usr/lib/python3.10/importlib/_bootstrap_external.py in exec_module(self, module)
/usr/lib/python3.10/importlib/_bootstrap.py in _call_with_frames_removed(f, *args, **kwds)
[~/.cache/huggingface/modules/datasets_modules/datasets/knowledgator--events_classification_biotech/9c8086d498c3104de3a3c5b6640837e18ccd829dcaca49f1cdffe3eb5c4a6361/events_classification_biotech.py](https://bcb6shpazyu-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20241022-060119_RC00_688494744#) in <module>
1 import datasets
2 from datasets import load_dataset
----> 3 from datasets.tasks import TextClassification
4
5 DESCRIPTION = """
ModuleNotFoundError: No module named 'datasets.tasks'
---------------------------------------------------------------------------
NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.
To view examples of installing some common dependencies, click the
"Open Examples" button below.
---------------------------------------------------------------------------
### Steps to reproduce the bug
!pip install datasets
from datasets import load_dataset
dataset = load_dataset('knowledgator/events_classification_biotech')
### Expected behavior
no ModuleNotFoundError
### Environment info
google colab
|
open
|
https://github.com/huggingface/datasets/issues/7248
| 2024-10-23T21:58:25 | 2024-10-24T17:00:19 | null |
{
"login": "shoowadoo",
"id": 93593941,
"type": "User"
}
|
[] | false |
[] |
2,606,230,029 | 7,247 |
Adding column with dict struction when mapping lead to wrong order
|
### Describe the bug
in `map()` function, I want to add a new column with a dict structure.
```
def map_fn(example):
example['text'] = {'user': ..., 'assistant': ...}
return example
```
However this leads to a wrong order `{'assistant':..., 'user':...}` in the dataset.
Thus I can't concatenate two datasets due to the different feature structures.
[Here](https://colab.research.google.com/drive/1zeaWq9Ith4DKWP_EiBNyLfc8S8I68LyY?usp=sharing) is a minimal reproducible example
This seems an issue in low level pyarrow library instead of datasets, however, I think datasets should allow concatenate two datasets actually in the same structure.
### Steps to reproduce the bug
[Here](https://colab.research.google.com/drive/1zeaWq9Ith4DKWP_EiBNyLfc8S8I68LyY?usp=sharing) is a minimal reproducible example
### Expected behavior
two datasets could be concatenated.
### Environment info
N/A
|
open
|
https://github.com/huggingface/datasets/issues/7247
| 2024-10-22T18:55:11 | 2024-10-22T18:55:23 | null |
{
"login": "chchch0109",
"id": 114604968,
"type": "User"
}
|
[] | false |
[] |
2,605,734,447 | 7,246 |
Set dev version
| null |
closed
|
https://github.com/huggingface/datasets/pull/7246
| 2024-10-22T15:04:47 | 2024-10-22T15:07:31 | 2024-10-22T15:04:58 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,605,701,235 | 7,245 |
Release: 3.0.2
| null |
closed
|
https://github.com/huggingface/datasets/pull/7245
| 2024-10-22T14:53:34 | 2024-10-22T15:01:50 | 2024-10-22T15:01:47 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,605,461,515 | 7,244 |
use huggingface_hub offline mode
|
and better handling of LocalEntryNotfoundError cc @Wauplin
follow up to #7234
|
closed
|
https://github.com/huggingface/datasets/pull/7244
| 2024-10-22T13:27:16 | 2024-10-22T14:10:45 | 2024-10-22T14:10:20 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,602,853,172 | 7,243 |
ArrayXD with None as leading dim incompatible with DatasetCardData
|
### Describe the bug
Creating a dataset with ArrayXD features leads to errors when downloading from hub due to DatasetCardData removing the Nones
@lhoestq
### Steps to reproduce the bug
```python
import numpy as np
from datasets import Array2D, Dataset, Features, load_dataset
def examples_generator():
for i in range(4):
yield {
"array_1d": np.zeros((10,1), dtype="uint16"),
"array_2d": np.zeros((10, 1), dtype="uint16"),
}
features = Features(array_1d=Array2D((None,1), "uint16"), array_2d=Array2D((None, 1), "uint16"))
dataset = Dataset.from_generator(examples_generator, features=features)
dataset.push_to_hub("alex-hh/test_array_1d2d")
ds = load_dataset("alex-hh/test_array_1d2d")
```
Source of error appears to be DatasetCardData.to_dict invoking DatasetCardData._remove_none
```python
from huggingface_hub import DatasetCardData
from datasets.info import DatasetInfosDict
dataset_card_data = DatasetCardData()
DatasetInfosDict({"default": dataset.info.copy()}).to_dataset_card_data(dataset_card_data)
print(dataset_card_data.to_dict()) # removes Nones in shape
```
### Expected behavior
Should be possible to load datasets saved with shape None in leading dimension
### Environment info
3.0.2 and latest huggingface_hub
|
open
|
https://github.com/huggingface/datasets/issues/7243
| 2024-10-21T15:08:13 | 2024-10-22T14:18:10 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,599,899,156 | 7,241 |
`push_to_hub` overwrite argument
|
### Feature request
Add an `overwrite` argument to the `push_to_hub` method.
### Motivation
I want to overwrite a repo without deleting it on Hugging Face. Is this possible? I couldn't find anything in the documentation or tutorials.
### Your contribution
I can create a PR.
|
closed
|
https://github.com/huggingface/datasets/issues/7241
| 2024-10-20T03:23:26 | 2024-10-24T17:39:08 | 2024-10-24T17:39:08 |
{
"login": "ceferisbarov",
"id": 60838378,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,598,980,027 | 7,240 |
Feature Request: Add functionality to pass split types like train, test in DatasetDict.map
|
Hello datasets!
We often encounter situations where we need to preprocess data differently depending on split types such as train, valid, and test.
However, while DatasetDict.map has features to pass rank or index, there's no functionality to pass split types.
Therefore, I propose adding a 'with_splits' parameter to DatasetDict, which would allow passing the split type through fn_kwargs.
|
closed
|
https://github.com/huggingface/datasets/pull/7240
| 2024-10-19T09:59:12 | 2025-01-06T08:04:08 | 2025-01-06T08:04:08 |
{
"login": "jp1924",
"id": 93233241,
"type": "User"
}
|
[] | true |
[] |
2,598,409,993 | 7,238 |
incompatibily issue when using load_dataset with datasets==3.0.1
|
### Describe the bug
There is a bug when using load_dataset with dataset version at 3.0.1 .
Please see below in the "steps to reproduce the bug".
To resolve the bug, I had to downgrade to version 2.21.0
OS: Ubuntu 24 (AWS instance)
Python: same bug under 3.12 and 3.10
The error I had was:
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/ubuntu/miniconda3/envs/maxence_env/lib/python3.10/site-packages/datasets/load.py", line 2096, in load_dataset
builder_instance.download_and_prepare(
File "/home/ubuntu/miniconda3/envs/maxence_env/lib/python3.10/site-packages/datasets/builder.py", line 924, in download_and_prepare
self._download_and_prepare(
File "/home/ubuntu/miniconda3/envs/maxence_env/lib/python3.10/site-packages/datasets/builder.py", line 1647, in _download_and_prepare
super()._download_and_prepare(
File "/home/ubuntu/miniconda3/envs/maxence_env/lib/python3.10/site-packages/datasets/builder.py", line 977, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/home/ubuntu/.cache/huggingface/modules/datasets_modules/datasets/mozilla-foundation--common_voice_6_0/cb17afd34f5799f97e8f48398748f83006335b702bd785f9880797838d541b81/common_voice_6_0.py", line 159, in _split_generators
archive_path = dl_manager.download(self._get_bundle_url(self.config.name, bundle_url_template))
File "/home/ubuntu/miniconda3/envs/maxence_env/lib/python3.10/site-packages/datasets/download/download_manager.py", line 150, in download
download_config = self.download_config.copy()
File "/home/ubuntu/miniconda3/envs/maxence_env/lib/python3.10/site-packages/datasets/download/download_config.py", line 73, in copy
return self.__class__(**{k: copy.deepcopy(v) for k, v in self.__dict__.items()})
TypeError: DownloadConfig.__init__() got an unexpected keyword argument 'ignore_url_params'
### Steps to reproduce the bug
1. install dataset with ```pip install datasets --upgrade```
2. launch python; from datasets import loaad_dataset
3. run load_dataset("mozilla-foundation/common_voice_6_0")
4. exit python
5. uninstall datasets; then ```pip install datasets==2.21.0```
6. launch python; from datasets import loaad_dataset
7. run load_dataset("mozilla-foundation/common_voice_6_0")
8. Everything runs great now
### Expected behavior
Be able to download a dataset without error
### Environment info
Copy-and-paste the text below in your GitHub issue.
- `datasets` version: 3.0.1
- Platform: Linux-6.8.0-1017-aws-x86_64-with-glibc2.39
- Python version: 3.12.4
- `huggingface_hub` version: 0.26.0
- PyArrow version: 17.0.0
- Pandas version: 2.2.3
- `fsspec` version: 2024.6.1
|
open
|
https://github.com/huggingface/datasets/issues/7238
| 2024-10-18T21:25:23 | 2024-12-09T09:49:32 | null |
{
"login": "jupiterMJM",
"id": 74985234,
"type": "User"
}
|
[] | false |
[] |
2,597,358,525 | 7,236 |
[MINOR:TYPO] Update arrow_dataset.py
|
Fix wrong link.
csv kwargs docstring link was pointing to pandas json docs.
|
closed
|
https://github.com/huggingface/datasets/pull/7236
| 2024-10-18T12:10:03 | 2024-10-24T15:06:43 | 2024-10-24T15:06:43 |
{
"login": "cakiki",
"id": 3664563,
"type": "User"
}
|
[] | true |
[] |
2,594,220,624 | 7,234 |
No need for dataset_info
|
save a useless call to /api/datasets/repo_id
|
closed
|
https://github.com/huggingface/datasets/pull/7234
| 2024-10-17T09:54:03 | 2024-10-22T12:30:40 | 2024-10-21T16:44:34 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,593,903,113 | 7,233 |
数据集数量问题
|
### Describe the bug
这里我进行大模型微调，当数据集数量为718时，模型可以正常微调，但是当我添加一个在前718个数据集中的数据或者新增一个数据就会报错
### Steps to reproduce the bug
1.
这里我的数据集可以微调的最后两个数据集是：
{
"messages": [
{
"role": "user",
"content": "完成校正装置设计后需要进行哪些工作？"
},
{
"role": "assistant",
"content": "一旦完成校正装置设计后，需要进行系统实际调校工作，或者进行MATLAB仿真以检查系统的时间响应特性。这时，需将系统建模时省略的部分尽可能加入系统，以保证仿真结果的逼真度。如果由于系统各种固有非线性因素影响，或者由于系统噪声和负载效应等因素的影响，使已校正系统不能满足全部性能指标要求，则需要适当调整校正装置的形式或参数，直到已校正系统满足全部性能指标为止。"
}
]
}
{
"messages": [
{
"role": "user",
"content": "为什么闭环带宽要求会影响串联超前校正的效果？"
},
{
"role": "assistant",
"content": "如果待校正系统不稳定，为了满足相角裕度的要求，超前网络需要提供很大的相角超前量。这会导致超前网络的a值必须选得很大，从而使已校正系统的带宽过大。带宽过大会使通过系统的高频噪声电平变高，可能导致系统失控。因此，闭环带宽的要求会影响串联超前校正的效果。"
}
]
}
2.
当我把倒数第二个数据集再次添加到数据集中时，此时我的最后三个数据集是
{
"messages": [
{
"role": "user",
"content": "完成校正装置设计后需要进行哪些工作？"
},
{
"role": "assistant",
"content": "一旦完成校正装置设计后，需要进行系统实际调校工作，或者进行MATLAB仿真以检查系统的时间响应特性。这时，需将系统建模时省略的部分尽可能加入系统，以保证仿真结果的逼真度。如果由于系统各种固有非线性因素影响，或者由于系统噪声和负载效应等因素的影响，使已校正系统不能满足全部性能指标要求，则需要适当调整校正装置的形式或参数，直到已校正系统满足全部性能指标为止。"
}
]
}
{
"messages": [
{
"role": "user",
"content": "为什么闭环带宽要求会影响串联超前校正的效果？"
},
{
"role": "assistant",
"content": "如果待校正系统不稳定，为了满足相角裕度的要求，超前网络需要提供很大的相角超前量。这会导致超前网络的a值必须选得很大，从而使已校正系统的带宽过大。带宽过大会使通过系统的高频噪声电平变高，可能导致系统失控。因此，闭环带宽的要求会影响串联超前校正的效果。"
}
]
}
{
"messages": [
{
"role": "user",
"content": "完成校正装置设计后需要进行哪些工作？"
},
{
"role": "assistant",
"content": "一旦完成校正装置设计后，需要进行系统实际调校工作，或者进行MATLAB仿真以检查系统的时间响应特性。这时，需将系统建模时省略的部分尽可能加入系统，以保证仿真结果的逼真度。如果由于系统各种固有非线性因素影响，或者由于系统噪声和负载效应等因素的影响，使已校正系统不能满足全部性能指标要求，则需要适当调整校正装置的形式或参数，直到已校正系统满足全部性能指标为止。"
}
]
}
这时系统会显示bug：
root@autodl-container-027f4cad3d-6baf4e64:~/autodl-tmp# python GLM-4/finetune_demo/finetune.py datasets/ ZhipuAI/glm-4-9b-chat GLM-4/finetune_demo/configs/lora.yaml
Loading checkpoint shards: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:02<00:00, 4.04it/s]
The installed version of bitsandbytes was compiled without GPU support. 8-bit optimizers, 8-bit multiplication, and GPU quantization are unavailable.
trainable params: 2,785,280 || all params: 9,402,736,640 || trainable%: 0.0296
Generating train split: 0 examples [00:00, ? examples/s]Failed to load JSON from file '/root/autodl-tmp/datasets/train.jsonl' with error <class 'pyarrow.lib.ArrowInvalid'>: JSON parse error: Missing a name for object member. in row 718
Generating train split: 0 examples [00:00, ? examples/s]
╭──────────────────────────────────────────────────────────────────────────────────────────────────────── Traceback (most recent call last) ─────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ /root/miniconda3/lib/python3.10/site-packages/datasets/packaged_modules/json/json.py:153 in _generate_tables │
│ │
│ 150 │ │ │ │ │ │ │ │ with open( │
│ 151 │ │ │ │ │ │ │ │ │ file, encoding=self.config.encoding, errors=self.con │
│ 152 │ │ │ │ │ │ │ │ ) as f: │
│ ❱ 153 │ │ │ │ │ │ │ │ │ df = pd.read_json(f, dtype_backend="pyarrow") │
│ 154 │ │ │ │ │ │ │ except ValueError: │
│ 155 │ │ │ │ │ │ │ │ logger.error(f"Failed to load JSON from file '{file}' wi │
│ 156 │ │ │ │ │ │ │ │ raise e │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/pandas/io/json/_json.py:815 in read_json │
│ │
│ 812 │ if chunksize: │
│ 813 │ │ return json_reader │
│ 814 │ else: │
│ ❱ 815 │ │ return json_reader.read() │
│ 816 │
│ 817 │
│ 818 class JsonReader(abc.Iterator, Generic[FrameSeriesStrT]): │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/pandas/io/json/_json.py:1025 in read │
│ │
│ 1022 │ │ │ │ │ │ data_lines = data.split("\n") │
│ 1023 │ │ │ │ │ │ obj = self._get_object_parser(self._combine_lines(data_lines)) │
│ 1024 │ │ │ │ else: │
│ ❱ 1025 │ │ │ │ │ obj = self._get_object_parser(self.data) │
│ 1026 │ │ │ │ if self.dtype_backend is not lib.no_default: │
│ 1027 │ │ │ │ │ return obj.convert_dtypes( │
│ 1028 │ │ │ │ │ │ infer_objects=False, dtype_backend=self.dtype_backend │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/pandas/io/json/_json.py:1051 in _get_object_parser │
│ │
│ 1048 │ │ } │
│ 1049 │ │ obj = None │
│ 1050 │ │ if typ == "frame": │
│ ❱ 1051 │ │ │ obj = FrameParser(json, **kwargs).parse() │
│ 1052 │ │ │
│ 1053 │ │ if typ == "series" or obj is None: │
│ 1054 │ │ │ if not isinstance(dtype, bool): │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/pandas/io/json/_json.py:1187 in parse │
│ │
│ 1184 │ │
│ 1185 │ @final │
│ 1186 │ def parse(self): │
│ ❱ 1187 │ │ self._parse() │
│ 1188 │ │ │
│ 1189 │ │ if self.obj is None: │
│ 1190 │ │ │ return None │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/pandas/io/json/_json.py:1403 in _parse │
│ │
│ 1400 │ │ │
│ 1401 │ │ if orient == "columns": │
│ 1402 │ │ │ self.obj = DataFrame( │
│ ❱ 1403 │ │ │ │ ujson_loads(json, precise_float=self.precise_float), dtype=None │
│ 1404 │ │ │ ) │
│ 1405 │ │ elif orient == "split": │
│ 1406 │ │ │ decoded = { │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
ValueError: Trailing data
During handling of the above exception, another exception occurred:
╭──────────────────────────────────────────────────────────────────────────────────────────────────────── Traceback (most recent call last) ─────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ /root/miniconda3/lib/python3.10/site-packages/datasets/builder.py:1997 in _prepare_split_single │
│ │
│ 1994 │ │ │ ) │
│ 1995 │ │ │ try: │
│ 1996 │ │ │ │ _time = time.time() │
│ ❱ 1997 │ │ │ │ for _, table in generator: │
│ 1998 │ │ │ │ │ if max_shard_size is not None and writer._num_bytes > max_shard_size │
│ 1999 │ │ │ │ │ │ num_examples, num_bytes = writer.finalize() │
│ 2000 │ │ │ │ │ │ writer.close() │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/datasets/packaged_modules/json/json.py:156 in _generate_tables │
│ │
│ 153 │ │ │ │ │ │ │ │ │ df = pd.read_json(f, dtype_backend="pyarrow") │
│ 154 │ │ │ │ │ │ │ except ValueError: │
│ 155 │ │ │ │ │ │ │ │ logger.error(f"Failed to load JSON from file '{file}' wi │
│ ❱ 156 │ │ │ │ │ │ │ │ raise e │
│ 157 │ │ │ │ │ │ │ if df.columns.tolist() == [0]: │
│ 158 │ │ │ │ │ │ │ │ df.columns = list(self.config.features) if self.config.f │
│ 159 │ │ │ │ │ │ │ try: │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/datasets/packaged_modules/json/json.py:130 in _generate_tables │
│ │
│ 127 │ │ │ │ │ │ try: │
│ 128 │ │ │ │ │ │ │ while True: │
│ 129 │ │ │ │ │ │ │ │ try: │
│ ❱ 130 │ │ │ │ │ │ │ │ │ pa_table = paj.read_json( │
│ 131 │ │ │ │ │ │ │ │ │ │ io.BytesIO(batch), read_options=paj.ReadOptions( │
│ 132 │ │ │ │ │ │ │ │ │ ) │
│ 133 │ │ │ │ │ │ │ │ │ break │
│ │
│ in pyarrow._json.read_json:308 │
│ │
│ in pyarrow.lib.pyarrow_internal_check_status:154 │
│ │
│ in pyarrow.lib.check_status:91 │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
ArrowInvalid: JSON parse error: Missing a name for object member. in row 718
The above exception was the direct cause of the following exception:
╭──────────────────────────────────────────────────────────────────────────────────────────────────────── Traceback (most recent call last) ─────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ /root/autodl-tmp/GLM-4/finetune_demo/finetune.py:406 in main │
│ │
│ 403 ): │
│ 404 │ ft_config = FinetuningConfig.from_file(config_file) │
│ 405 │ tokenizer, model = load_tokenizer_and_model(model_dir, peft_config=ft_config.peft_co │
│ ❱ 406 │ data_manager = DataManager(data_dir, ft_config.data_config) │
│ 407 │ │
│ 408 │ train_dataset = data_manager.get_dataset( │
│ 409 │ │ Split.TRAIN, │
│ │
│ /root/autodl-tmp/GLM-4/finetune_demo/finetune.py:204 in __init__ │
│ │
│ 201 │ def __init__(self, data_dir: str, data_config: DataConfig): │
│ 202 │ │ self._num_proc = data_config.num_proc │
│ 203 │ │ │
│ ❱ 204 │ │ self._dataset_dct = _load_datasets( │
│ 205 │ │ │ data_dir, │
│ 206 │ │ │ data_config.data_format, │
│ 207 │ │ │ data_config.data_files, │
│ │
│ /root/autodl-tmp/GLM-4/finetune_demo/finetune.py:189 in _load_datasets │
│ │
│ 186 │ │ num_proc: Optional[int], │
│ 187 ) -> DatasetDict: │
│ 188 │ if data_format == '.jsonl': │
│ ❱ 189 │ │ dataset_dct = load_dataset( │
│ 190 │ │ │ data_dir, │
│ 191 │ │ │ data_files=data_files, │
│ 192 │ │ │ split=None, │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/datasets/load.py:2616 in load_dataset │
│ │
│ 2613 │ │ return builder_instance.as_streaming_dataset(split=split) │
│ 2614 │ │
│ 2615 │ # Download and prepare data │
│ ❱ 2616 │ builder_instance.download_and_prepare( │
│ 2617 │ │ download_config=download_config, │
│ 2618 │ │ download_mode=download_mode, │
│ 2619 │ │ verification_mode=verification_mode, │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/datasets/builder.py:1029 in download_and_prepare │
│ │
│ 1026 │ │ │ │ │ │ │ prepare_split_kwargs["max_shard_size"] = max_shard_size │
│ 1027 │ │ │ │ │ │ if num_proc is not None: │
│ 1028 │ │ │ │ │ │ │ prepare_split_kwargs["num_proc"] = num_proc │
│ ❱ 1029 │ │ │ │ │ │ self._download_and_prepare( │
│ 1030 │ │ │ │ │ │ │ dl_manager=dl_manager, │
│ 1031 │ │ │ │ │ │ │ verification_mode=verification_mode, │
│ 1032 │ │ │ │ │ │ │ **prepare_split_kwargs, │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/datasets/builder.py:1124 in _download_and_prepare │
│ │
│ 1121 │ │ │ │
│ 1122 │ │ │ try: │
│ 1123 │ │ │ │ # Prepare split will record examples associated to the split │
│ ❱ 1124 │ │ │ │ self._prepare_split(split_generator, **prepare_split_kwargs) │
│ 1125 │ │ │ except OSError as e: │
│ 1126 │ │ │ │ raise OSError( │
│ 1127 │ │ │ │ │ "Cannot find data file. " │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/datasets/builder.py:1884 in _prepare_split │
│ │
│ 1881 │ │ │ gen_kwargs = split_generator.gen_kwargs │
│ 1882 │ │ │ job_id = 0 │
│ 1883 │ │ │ with pbar: │
│ ❱ 1884 │ │ │ │ for job_id, done, content in self._prepare_split_single( │
│ 1885 │ │ │ │ │ gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args │
│ 1886 │ │ │ │ ): │
│ 1887 │ │ │ │ │ if done: │
│ │
│ /root/miniconda3/lib/python3.10/site-packages/datasets/builder.py:2040 in _prepare_split_single │
│ │
│ 2037 │ │ │ │ e = e.__context__ │
│ 2038 │ │ │ if isinstance(e, DatasetGenerationError): │
│ 2039 │ │ │ │ raise │
│ ❱ 2040 │ │ │ raise DatasetGenerationError("An error occurred while generating the dataset │
│ 2041 │ │ │
│ 2042 │ │ yield job_id, True, (total_num_examples, total_num_bytes, writer._features, num_ │
│ 2043 │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
DatasetGenerationError: An error occurred while generating the dataset
3.请问是否可以帮我解决
### Expected behavior
希望问题可以得到解决
### Environment info
Copy-and-paste the text below in your GitHub issue.
- `datasets` version: 2.20.0
- Platform: Linux-4.19.90-2107.6.0.0192.8.oe1.bclinux.x86_64-x86_64-with-glibc2.35
- Python version: 3.10.8
- `huggingface_hub` version: 0.24.6
- PyArrow version: 16.1.0
- Pandas version: 2.2.2
- `fsspec` version: 2023.12.2
|
open
|
https://github.com/huggingface/datasets/issues/7233
| 2024-10-17T07:41:44 | 2024-10-17T07:41:44 | null |
{
"login": "want-well",
"id": 180297268,
"type": "User"
}
|
[] | false |
[] |
2,593,720,548 | 7,232 |
(Super tiny doc update) Mention to_polars
|
polars is also quite popular now, thus this tiny update can tell users polars is supported
|
closed
|
https://github.com/huggingface/datasets/pull/7232
| 2024-10-17T06:08:53 | 2024-10-24T23:11:05 | 2024-10-24T15:06:16 |
{
"login": "fzyzcjy",
"id": 5236035,
"type": "User"
}
|
[] | true |
[] |
2,592,011,737 | 7,231 |
Fix typo in image dataset docs
|
Fix typo in image dataset docs.
Typo reported by @datavistics.
|
closed
|
https://github.com/huggingface/datasets/pull/7231
| 2024-10-16T14:05:46 | 2024-10-16T17:06:21 | 2024-10-16T17:06:19 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true |
[] |
2,589,531,942 | 7,230 |
Video support
|
(wip and experimental)
adding the `Video` type based on `VideoReader` from `decord`
```python
>>>from datasets import load_dataset
>>> ds = load_dataset("path/to/videos", split="train").with_format("torch")
>>> print(ds[0]["video"])
<decord.video_reader.VideoReader object at 0x337a47910>
>>> print(ds[0]["video"][0])
tensor([[[73, 73, 73],
[73, 73, 73],
[73, 73, 73],
...,
[23, 23, 23],
[23, 23, 23],
[23, 23, 23]]], dtype=torch.uint8)
```
the storage is the same as for audio and images: `{"path": pa.string(), "bytes": pa.binary()}` and I did a small to keep the hf:// URL in the "path" field if possible, this way the viewer can link to fiels on the hub if possible
|
closed
|
https://github.com/huggingface/datasets/pull/7230
| 2024-10-15T18:17:29 | 2024-10-24T16:39:51 | 2024-10-24T16:39:50 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,588,847,398 | 7,229 |
handle config_name=None in push_to_hub
|
This caught me out - thought it might be better to explicitly handle None?
|
closed
|
https://github.com/huggingface/datasets/pull/7229
| 2024-10-15T13:48:57 | 2024-10-24T17:51:52 | 2024-10-24T17:51:52 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,587,310,094 | 7,228 |
Composite (multi-column) features
|
### Feature request
Structured data types (graphs etc.) might often be most efficiently stored as multiple columns, which then need to be combined during feature decoding
Although it is currently possible to nest features as structs, my impression is that in particular when dealing with e.g. a feature composed of multiple numpy array / ArrayXD's, it would be more efficient to store each ArrayXD as a separate column (though I'm not sure by how much)
Perhaps specification / implementation could be supported by something like:
```
features=Features(**{("feature0", "feature1")=Features(feature0=Array2D((None,10), dtype="float32"), feature1=Array2D((None,10), dtype="float32"))
```
### Motivation
Defining efficient composite feature types based on numpy arrays for representing data such as graphs with multiple node and edge attributes is currently challenging.
### Your contribution
Possibly able to contribute
|
open
|
https://github.com/huggingface/datasets/issues/7228
| 2024-10-14T23:59:19 | 2024-10-15T11:17:15 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,587,048,312 | 7,227 |
fast array extraction
|
Implements #7210 using method suggested in https://github.com/huggingface/datasets/pull/7207#issuecomment-2411789307
```python
import numpy as np
from datasets import Dataset, Features, Array3D
features=Features(**{"array0": Array3D((None, 10, 10), dtype="float32"), "array1": Array3D((None,10,10), dtype="float32")})
dataset = Dataset.from_dict({f"array{i}": [np.zeros((x,10,10), dtype=np.float32) for x in [2000,1000]*25] for i in range(2)}, features=features)
```
~0.02 s vs 0.9s on main
```python
ds = dataset.to_iterable_dataset()
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
```
< 0.01 s vs 1.3 s on main
@lhoestq I can see this breaks a bunch of array-related tests but can update the test cases if you would support making this change?
I also added an Array1D feature which will always be decoded into a numpy array and likewise improves extraction performance:
```python
from datasets import Dataset, Features, Array1D, Sequence, Value
array_features=Features(**{"array0": Array1D((None,), dtype="float32"), "array1": Array1D((None,), dtype="float32")})
sequence_features=Features(**{"array0": Sequence(feature=Value("float32"), length=-1), "array1": Sequence(feature=Value("float32"), length=-1)})
array_dataset = Dataset.from_dict({f"array{i}": [np.zeros((x,), dtype=np.float32) for x in [20000,10000]*25] for i in range(2)}, features=array_features)
sequence_dataset = Dataset.from_dict({f"array{i}": [np.zeros((x,), dtype=np.float32) for x in [20000,10000]*25] for i in range(2)}, features=sequence_features)
```python
t0 = time.time()
for ex in array_dataset.to_iterable_dataset():
pass
t1 = time.time()
```
< 0.01 s
```python
t0 = time.time()
for ex in sequence_dataset.to_iterable_dataset():
pass
t1 = time.time()
```
~1.1s
And also added support for extracting structs of arrays as dicts of numpy arrays:
```python
import numpy as np
from datasets import Dataset, Features, Array3D, Sequence
features=Features(struct={"array0": Array3D((None,10,10), dtype="float32"), "array1": Array3D((None,10,10), dtype="float32")}, _list=Sequence(feature=Array3D((None,10,10), dtype="float32")))
dataset = Dataset.from_dict({"struct": [{f"array{i}": np.zeros((x,10,10), dtype=np.float32) for i in range(2)} for x in [2000,1000]*25], "_list": [[np.zeros((x,10,10), dtype=np.float32) for i in range(2)] for x in [2000,1000]*25]}, features=features)
```
```python
t0 = time.time()
for ex in dataset.to_iterable_dataset():
pass
t1 = time.time()
assert isinstance(ex["struct"]["array0"], np.ndarray) and ex["struct"]["array0"].ndim == 3
```
~0.02 s and no exception vs ~7s with an exception on main
|
open
|
https://github.com/huggingface/datasets/pull/7227
| 2024-10-14T20:51:32 | 2025-01-28T09:39:26 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,586,920,351 | 7,226 |
Add R as a How to use from the Polars (R) Library as an option
|
### Feature request
The boiler plate code to access a dataset via the hugging face file system is very useful. Please addd
## Add Polars (R) option
The equivailent code works, because the [Polars-R](https://github.com/pola-rs/r-polars) wrapper has hugging faces funcitonaliy as well.
```r
library(polars)
df <- pl$read_parquet("hf://datasets/SALURBAL/core__admin_cube_public/core__admin_cube_public.parquet")
```
## Polars (python) option
## Libraries Currently
### Motivation
There are many data/analysis/research/statistics teams (particularly in academia and pharma) that use R as the default language. R has great integration with most of the newer data techs (arrow, parquet, polars) and having this included could really help in bringing this community into the hugging faces ecosystem.
**This is a small/low-hanging-fruit front end change but would make a big impact expanding the community**
### Your contribution
I am not sure which repositroy this should be in, but I have experience in R, Python and JS and happy to submit a PR in the appropriate repository.
|
open
|
https://github.com/huggingface/datasets/issues/7226
| 2024-10-14T19:56:07 | 2024-10-14T19:57:13 | null |
{
"login": "ran-codes",
"id": 45013044,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,586,229,216 | 7,225 |
Huggingface GIT returns null as Content-Type instead of application/x-git-receive-pack-result
|
### Describe the bug
We push changes to our datasets programmatically. Our git client jGit reports that the hf git server returns null as Content-Type after a push.
### Steps to reproduce the bug
A basic kotlin application:
```
val person = PersonIdent(
"padmalcom",
"padmalcom@sth.com"
)
val cp = UsernamePasswordCredentialsProvider(
"padmalcom",
"mysecrettoken"
)
val git =
KGit.cloneRepository {
setURI("https://huggingface.co/datasets/sth/images")
setTimeout(60)
setProgressMonitor(TextProgressMonitor())
setCredentialsProvider(cp)
}
FileOutputStream("./images/images.csv").apply { writeCsv(images) }
git.add {
addFilepattern("images.csv")
}
for (i in images) {
FileUtils.copyFile(
File("./files/${i.id}"),
File("./images/${i.id + File(i.fileName).extension }")
)
git.add {
addFilepattern("${i.id + File(i.fileName).extension }")
}
}
val revCommit = git.commit {
author = person
message = "Uploading images at " + LocalDateTime.now()
.format(DateTimeFormatter.ISO_DATE_TIME)
setCredentialsProvider(cp)
}
val push = git.push {
setCredentialsProvider(cp)
}
```
### Expected behavior
The git server is expected to return the Content-Type _application/x-git-receive-pack-result_.
### Environment info
It is independent from the datasets library.
|
open
|
https://github.com/huggingface/datasets/issues/7225
| 2024-10-14T14:33:06 | 2024-10-14T14:33:06 | null |
{
"login": "padmalcom",
"id": 3961950,
"type": "User"
}
|
[] | false |
[] |
2,583,233,980 | 7,224 |
fallback to default feature casting in case custom features not available during dataset loading
|
a fix for #7223 in case datasets is happy to support this kind of extensibility! seems cool / powerful for allowing sharing of datasets with potentially different feature types
|
open
|
https://github.com/huggingface/datasets/pull/7224
| 2024-10-12T16:13:56 | 2024-10-12T16:13:56 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,583,231,590 | 7,223 |
Fallback to arrow defaults when loading dataset with custom features that aren't registered locally
|
### Describe the bug
Datasets allows users to create and register custom features.
However if datasets are then pushed to the hub, this means that anyone calling load_dataset without registering the custom Features in the same way as the dataset creator will get an error message.
It would be nice to offer a fallback in this case.
### Steps to reproduce the bug
```python
load_dataset("alex-hh/custom-features-example")
```
(Dataset creation process - must be run in separate session so that NewFeature isn't registered in session in which download is attempted:)
```python
from dataclasses import dataclass, field
import pyarrow as pa
from datasets.features.features import register_feature
from datasets import Dataset, Features, Value, load_dataset
from datasets import Feature
@dataclass
class NewFeature(Feature):
_type: str = field(default="NewFeature", init=False, repr=False)
def __call__(self):
return pa.int32()
def examples_generator():
for i in range(5):
yield {"feature": i}
ds = Dataset.from_generator(examples_generator, features=Features(feature=NewFeature()))
ds.push_to_hub("alex-hh/custom-features-example")
register_feature(NewFeature, "NewFeature")
```
### Expected behavior
It would be nice, and offer greater extensibility, if there was some kind of graceful fallback mechanism in place for cases where user-defined features are stored in the dataset but not available locally.
### Environment info
3.0.2
|
open
|
https://github.com/huggingface/datasets/issues/7223
| 2024-10-12T16:08:20 | 2024-10-12T16:08:20 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,582,678,033 | 7,222 |
TypeError: Couldn't cast array of type string to null in long json
|
### Describe the bug
In general, changing the type from string to null is allowed within a dataset — there are even examples of this in the documentation.
However, if the dataset is large and unevenly distributed, this allowance stops working. The schema gets locked in after reading a chunk.
Consequently, if all values in the first chunk of a field are, for example, null, the field will be locked as type null, and if a string appears in that field in the second chunk, it will trigger this error:
<details>
<summary>Traceback </summary>
```
TypeError Traceback (most recent call last)
[/usr/local/lib/python3.10/dist-packages/datasets/builder.py](https://localhost:8080/#) in _prepare_split_single(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)
1868 try:
-> 1869 writer.write_table(table)
1870 except CastError as cast_error:
14 frames
[/usr/local/lib/python3.10/dist-packages/datasets/arrow_writer.py](https://localhost:8080/#) in write_table(self, pa_table, writer_batch_size)
579 pa_table = pa_table.combine_chunks()
--> 580 pa_table = table_cast(pa_table, self._schema)
581 if self.embed_local_files:
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in table_cast(table, schema)
2291 if table.schema != schema:
-> 2292 return cast_table_to_schema(table, schema)
2293 elif table.schema.metadata != schema.metadata:
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in cast_table_to_schema(table, schema)
2244 )
-> 2245 arrays = [
2246 cast_array_to_feature(
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in <listcomp>(.0)
2245 arrays = [
-> 2246 cast_array_to_feature(
2247 table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in wrapper(array, *args, **kwargs)
1794 if isinstance(array, pa.ChunkedArray):
-> 1795 return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
1796 else:
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in <listcomp>(.0)
1794 if isinstance(array, pa.ChunkedArray):
-> 1795 return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
1796 else:
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in cast_array_to_feature(array, feature, allow_primitive_to_str, allow_decimal_to_str)
2101 elif not isinstance(feature, (Sequence, dict, list, tuple)):
-> 2102 return array_cast(
2103 array,
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in wrapper(array, *args, **kwargs)
1796 else:
-> 1797 return func(array, *args, **kwargs)
1798
[/usr/local/lib/python3.10/dist-packages/datasets/table.py](https://localhost:8080/#) in array_cast(array, pa_type, allow_primitive_to_str, allow_decimal_to_str)
1947 if pa.types.is_null(pa_type) and not pa.types.is_null(array.type):
-> 1948 raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
1949 return array.cast(pa_type)
TypeError: Couldn't cast array of type string to null
The above exception was the direct cause of the following exception:
DatasetGenerationError Traceback (most recent call last)
[<ipython-input-353-e02f83980611>](https://localhost:8080/#) in <cell line: 1>()
----> 1 dd = load_dataset("json", data_files=["TEST.json"])
[/usr/local/lib/python3.10/dist-packages/datasets/load.py](https://localhost:8080/#) in load_dataset(path, name, data_dir, data_files, split, cache_dir, features, download_config, download_mode, verification_mode, keep_in_memory, save_infos, revision, token, streaming, num_proc, storage_options, trust_remote_code, **config_kwargs)
2094
2095 # Download and prepare data
-> 2096 builder_instance.download_and_prepare(
2097 download_config=download_config,
2098 download_mode=download_mode,
[/usr/local/lib/python3.10/dist-packages/datasets/builder.py](https://localhost:8080/#) in download_and_prepare(self, output_dir, download_config, download_mode, verification_mode, dl_manager, base_path, file_format, max_shard_size, num_proc, storage_options, **download_and_prepare_kwargs)
922 if num_proc is not None:
923 prepare_split_kwargs["num_proc"] = num_proc
--> 924 self._download_and_prepare(
925 dl_manager=dl_manager,
926 verification_mode=verification_mode,
[/usr/local/lib/python3.10/dist-packages/datasets/builder.py](https://localhost:8080/#) in _download_and_prepare(self, dl_manager, verification_mode, **prepare_split_kwargs)
997 try:
998 # Prepare split will record examples associated to the split
--> 999 self._prepare_split(split_generator, **prepare_split_kwargs)
1000 except OSError as e:
1001 raise OSError(
[/usr/local/lib/python3.10/dist-packages/datasets/builder.py](https://localhost:8080/#) in _prepare_split(self, split_generator, file_format, num_proc, max_shard_size)
1738 job_id = 0
1739 with pbar:
-> 1740 for job_id, done, content in self._prepare_split_single(
1741 gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
1742 ):
[/usr/local/lib/python3.10/dist-packages/datasets/builder.py](https://localhost:8080/#) in _prepare_split_single(self, gen_kwargs, fpath, file_format, max_shard_size, job_id)
1894 if isinstance(e, DatasetGenerationError):
1895 raise
-> 1896 raise DatasetGenerationError("An error occurred while generating the dataset") from e
1897
1898 yield job_id, True, (total_num_examples, total_num_bytes, writer._features, num_shards, shard_lengths)
DatasetGenerationError: An error occurred while generating the dataset
```
</details>
### Steps to reproduce the bug
```python
import json
from datasets import load_dataset
with open("TEST.json", "w") as f:
row = {"ballast": "qwerty" * 1000, "b": None}
row_str = json.dumps(row) + "\n"
line_size = len(row_str)
chunk_size = 10 << 20
lines_in_chunk = chunk_size // line_size + 1
print(f"Writing {lines_in_chunk} lines")
for i in range(lines_in_chunk):
f.write(row_str)
null_row = {"ballast": "Gotcha", "b": "Not Null"}
f.write(json.dumps(null_row) + "\n")
load_dataset("json", data_files=["TEST.json"])
```
### Expected behavior
Concatenation of the chunks without errors
### Environment info
- `datasets` version: 3.0.1
- Platform: Linux-6.1.85+-x86_64-with-glibc2.35
- Python version: 3.10.12
- `huggingface_hub` version: 0.24.7
- PyArrow version: 16.1.0
- Pandas version: 2.2.2
- `fsspec` version: 2024.6.1
|
open
|
https://github.com/huggingface/datasets/issues/7222
| 2024-10-12T08:14:59 | 2025-07-21T03:07:32 | null |
{
"login": "nokados",
"id": 5142577,
"type": "User"
}
|
[] | false |
[] |
2,582,114,631 | 7,221 |
add CustomFeature base class to support user-defined features with encoding/decoding logic
|
intended as fix for #7220 if this kind of extensibility is something that datasets is willing to support!
```python
from datasets.features.features import CustomFeature
class ListOfStrs(CustomFeature):
requires_encoding = True
def _encode_example(self, value):
if isinstance(value, str):
return [str]
else:
return value
feats = Features(strlist=ListOfStrs())
feats.encode_example({"strlist": "a"})["strlist"] == feats["strlist"].encode_example("a")
```
|
closed
|
https://github.com/huggingface/datasets/pull/7221
| 2024-10-11T20:10:27 | 2025-01-28T09:40:29 | 2025-01-28T09:40:29 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,582,036,110 | 7,220 |
Custom features not compatible with special encoding/decoding logic
|
### Describe the bug
It is possible to register custom features using datasets.features.features.register_feature (https://github.com/huggingface/datasets/pull/6727)
However such features are not compatible with Features.encode_example/decode_example if they require special encoding / decoding logic because encode_nested_example / decode_nested_example checks whether the feature is in a fixed list of encodable types:
https://github.com/huggingface/datasets/blob/16a121d7821a7691815a966270f577e2c503473f/src/datasets/features/features.py#L1349
This prevents the extensibility of features to complex cases
### Steps to reproduce the bug
```python
class ListOfStrs:
def encode_example(self, value):
if isinstance(value, str):
return [str]
else:
return value
feats = Features(strlist=ListOfStrs())
assert feats.encode_example({"strlist": "a"})["strlist"] = feats["strlist"].encode_example("a")}
```
### Expected behavior
Registered feature types should be encoded based on some property of the feature (e.g. requires_encoding)?
### Environment info
3.0.2
|
open
|
https://github.com/huggingface/datasets/issues/7220
| 2024-10-11T19:20:11 | 2024-11-08T15:10:58 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,581,708,084 | 7,219 |
bump fsspec
| null |
closed
|
https://github.com/huggingface/datasets/pull/7219
| 2024-10-11T15:56:36 | 2024-10-14T08:21:56 | 2024-10-14T08:21:55 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,581,095,098 | 7,217 |
ds.map(f, num_proc=10) is slower than df.apply
|
### Describe the bug
pandas columns: song_id, song_name
ds = Dataset.from_pandas(df)
def has_cover(song_name):
if song_name is None or pd.isna(song_name):
return False
return 'cover' in song_name.lower()
df['has_cover'] = df.song_name.progress_apply(has_cover)
ds = ds.map(lambda x: {'has_cover': has_cover(x['song_name'])}, num_proc=10)
time cost:
1. df.apply: 100%|██████████| 12500592/12500592 [00:13<00:00, 959825.47it/s]
2. ds.map: Map (num_proc=10): 31%
3899028/12500592 [00:28<00:38, 222532.89 examples/s]
### Steps to reproduce the bug
pandas columns: song_id, song_name
ds = Dataset.from_pandas(df)
def has_cover(song_name):
if song_name is None or pd.isna(song_name):
return False
return 'cover' in song_name.lower()
df['has_cover'] = df.song_name.progress_apply(has_cover)
ds = ds.map(lambda x: {'has_cover': has_cover(x['song_name'])}, num_proc=10)
### Expected behavior
ds.map is ~num_proc faster than df.apply
### Environment info
pandas: 2.2.2
datasets: 2.19.1
|
open
|
https://github.com/huggingface/datasets/issues/7217
| 2024-10-11T11:04:05 | 2025-02-28T21:21:01 | null |
{
"login": "lanlanlanlanlanlan365",
"id": 178981231,
"type": "User"
}
|
[] | false |
[] |
2,579,942,939 | 7,215 |
Iterable dataset map with explicit features causes slowdown for Sequence features
|
### Describe the bug
When performing map, it's nice to be able to pass the new feature type, and indeed required by interleave and concatenate datasets.
However, this can cause a major slowdown for certain types of array features due to the features being re-encoded.
This is separate to the slowdown reported in #7206
### Steps to reproduce the bug
```
from datasets import Dataset, Features, Array3D, Sequence, Value
import numpy as np
import time
features=Features(**{"array0": Sequence(feature=Value("float32"), length=-1), "array1": Sequence(feature=Value("float32"), length=-1)})
dataset = Dataset.from_dict({f"array{i}": [np.zeros((x,), dtype=np.float32) for x in [5000,10000]*25] for i in range(2)}, features=features)
```
```
ds = dataset.to_iterable_dataset()
ds = ds.with_format("numpy").map(lambda x: x)
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
```
~1.5 s on main
```
ds = dataset.to_iterable_dataset()
ds = ds.with_format("numpy").map(lambda x: x, features=features)
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
```
~ 3 s on main
### Expected behavior
I'm not 100% sure whether passing new feature types to formatted outputs of map should be supported or not, but assuming it should, then there should be a cost-free way to specify the new feature type - knowing feature type is required by interleave_datasets and concatenate_datasets for example
### Environment info
3.0.2
|
open
|
https://github.com/huggingface/datasets/issues/7215
| 2024-10-10T22:08:20 | 2024-10-10T22:10:32 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,578,743,713 | 7,214 |
Formatted map + with_format(None) changes array dtype for iterable datasets
|
### Describe the bug
When applying with_format -> map -> with_format(None), array dtypes seem to change, even if features are passed
### Steps to reproduce the bug
```python
features=Features(**{"array0": Array3D((None, 10, 10), dtype="float32")})
dataset = Dataset.from_dict({f"array0": [np.zeros((100,10,10), dtype=np.float32)]*25}, features=features)
ds = dataset.to_iterable_dataset().with_format("numpy").map(lambda x: x, features=features)
ex_0 = next(iter(ds))
ds = dataset.to_iterable_dataset().with_format("numpy").map(lambda x: x, features=features).with_format(None)
ex_1 = next(iter(ds))
assert ex_1["array0"].dtype == ex_0["array0"].dtype, f"{ex_1['array0'].dtype} {ex_0['array0'].dtype}"
```
### Expected behavior
Dtypes should be preserved.
### Environment info
3.0.2
|
open
|
https://github.com/huggingface/datasets/issues/7214
| 2024-10-10T12:45:16 | 2024-10-12T16:55:57 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,578,675,565 | 7,213 |
Add with_rank to Dataset.from_generator
|
### Feature request
Add `with_rank` to `Dataset.from_generator` similar to `Dataset.map` and `Dataset.filter`.
### Motivation
As for `Dataset.map` and `Dataset.filter`, this is useful when creating cache files using multi-GPU, where the rank can be used to select GPU IDs. For now, rank can be added in the `gen_kwars` argument; however, this, in turn, includes the rank when computing the fingerprint.
### Your contribution
Added #7199 which passes rank based on the `job_id` set by `num_proc`.
|
open
|
https://github.com/huggingface/datasets/issues/7213
| 2024-10-10T12:15:29 | 2024-10-10T12:17:11 | null |
{
"login": "muthissar",
"id": 17828087,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,578,641,259 | 7,212 |
Windows do not supprot signal.alarm and singal.signal
|
### Describe the bug
signal.alarm and signal.signal are used in the load.py module, but these are not supported by Windows.
### Steps to reproduce the bug
lighteval accelerate --model_args "pretrained=gpt2,trust_remote_code=True" --tasks "community|kinit_sts" --custom_tasks "community_tasks/kinit_evals.py" --output_dir "./evals"
### Expected behavior
proceed with input(..) method
### Environment info
Windows 11
|
open
|
https://github.com/huggingface/datasets/issues/7212
| 2024-10-10T12:00:19 | 2024-10-10T12:00:19 | null |
{
"login": "TomasJavurek",
"id": 33832672,
"type": "User"
}
|
[] | false |
[] |
2,576,400,502 | 7,211 |
Describe only selected fields in README
|
### Feature request
Hi Datasets team!
Is it possible to add the ability to describe only selected fields of the dataset files in `README.md`? For example, I have this open dataset ([open-llm-leaderboard/results](https://huggingface.co/datasets/open-llm-leaderboard/results?row=0)) and I want to describe only some fields in order not to overcomplicate the Dataset Preview and filter out some fields
### Motivation
The `Results` dataset for the Open LLM Leaderboard contains json files with a complex nested structure. I would like to add `README.md` there to use the SQL console, for example. But if I describe the structure of this dataset completely, it will overcomplicate the use of Dataset Preview and the total number of columns will exceed 50
### Your contribution
I'm afraid I'm not familiar with the project structure, so I won't be able to open a PR, but I'll try to help with something else if possible
|
open
|
https://github.com/huggingface/datasets/issues/7211
| 2024-10-09T16:25:47 | 2024-10-09T16:25:47 | null |
{
"login": "alozowski",
"id": 67658835,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,575,883,939 | 7,210 |
Convert Array features to numpy arrays rather than lists by default
|
### Feature request
It is currently quite easy to cause massive slowdowns when using datasets and not familiar with the underlying data conversions by e.g. making bad choices of formatting.
Would it be more user-friendly to set defaults that avoid this as much as possible? e.g. format Array features as numpy arrays rather than python lists
### Motivation
Default array formatting leads to slow performance: e.g.
```python
import numpy as np
from datasets import Dataset, Features, Array3D
features=Features(**{"array0": Array3D((None, 10, 10), dtype="float32"), "array1": Array3D((None,10,10), dtype="float32")})
dataset = Dataset.from_dict({f"array{i}": [np.zeros((x,10,10), dtype=np.float32) for x in [2000,1000]*25] for i in range(2)}, features=features)
```
```python
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
```
~1.4 s
```python
ds = dataset.to_iterable_dataset()
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
```
~10s
```python
ds = dataset.with_format("numpy")
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
```
~0.04s
```python
ds = dataset.to_iterable_dataset().with_format("numpy")
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
```
~0.04s
### Your contribution
May be able to contribute
|
open
|
https://github.com/huggingface/datasets/issues/7210
| 2024-10-09T13:05:21 | 2024-10-09T13:05:21 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,575,526,651 | 7,209 |
Preserve features in iterable dataset.filter
|
Fixes example in #7208 - I'm not sure what other checks I should do? @lhoestq
I also haven't thought hard about the concatenate / interleaving example iterables but think this might work assuming that features are either all identical or None?
|
closed
|
https://github.com/huggingface/datasets/pull/7209
| 2024-10-09T10:42:05 | 2024-10-16T11:27:22 | 2024-10-09T16:04:07 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,575,484,256 | 7,208 |
Iterable dataset.filter should not override features
|
### Describe the bug
When calling filter on an iterable dataset, the features get set to None
### Steps to reproduce the bug
import numpy as np
import time
from datasets import Dataset, Features, Array3D
```python
features=Features(**{"array0": Array3D((None, 10, 10), dtype="float32"), "array1": Array3D((None,10,10), dtype="float32")})
dataset = Dataset.from_dict({f"array{i}": [np.zeros((x,10,10), dtype=np.float32) for x in [2000,1000]*25] for i in range(2)}, features=features)
ds = dataset.to_iterable_dataset()
orig_column_names = ds.column_names
ds = ds.filter(lambda x: True)
assert ds.column_names == orig_column_names
```
### Expected behavior
Filter should preserve features information
### Environment info
3.0.2
|
closed
|
https://github.com/huggingface/datasets/issues/7208
| 2024-10-09T10:23:45 | 2024-10-09T16:08:46 | 2024-10-09T16:08:45 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,573,582,335 | 7,207 |
apply formatting after iter_arrow to speed up format -> map, filter for iterable datasets
|
I got to this by hacking around a bit but it seems to solve #7206
I have no idea if this approach makes sense or would break something else?
Could maybe work on a full pr if this looks reasonable @lhoestq ? I imagine the same issue might affect other iterable dataset methods?
|
closed
|
https://github.com/huggingface/datasets/pull/7207
| 2024-10-08T15:44:53 | 2025-01-14T18:36:03 | 2025-01-14T16:59:30 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,573,567,467 | 7,206 |
Slow iteration for iterable dataset with numpy formatting for array data
|
### Describe the bug
When working with large arrays, setting with_format to e.g. numpy then applying map causes a significant slowdown for iterable datasets.
### Steps to reproduce the bug
```python
import numpy as np
import time
from datasets import Dataset, Features, Array3D
features=Features(**{"array0": Array3D((None, 10, 10), dtype="float32"), "array1": Array3D((None,10,10), dtype="float32")})
dataset = Dataset.from_dict({f"array{i}": [np.zeros((x,10,10), dtype=np.float32) for x in [2000,1000]*25] for i in range(2)}, features=features)
```
Then
```python
ds = dataset.to_iterable_dataset()
ds = ds.with_format("numpy").map(lambda x: x)
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
print(t1-t0)
```
takes 27 s, whereas
```python
ds = dataset.to_iterable_dataset()
ds = ds.with_format("numpy")
ds = dataset.to_iterable_dataset()
t0 = time.time()
for ex in ds:
pass
t1 = time.time()
print(t1 - t0)
```
takes ~1s
### Expected behavior
Map should not introduce a slowdown when formatting is enabled.
### Environment info
3.0.2
|
open
|
https://github.com/huggingface/datasets/issues/7206
| 2024-10-08T15:38:11 | 2024-10-17T17:14:52 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,573,490,859 | 7,205 |
fix ci benchmark
|
we're not using the benchmarks anymore + they were not working anyway due to token permissions
I keep the code in case we ever want to re-run the benchmark manually
|
closed
|
https://github.com/huggingface/datasets/pull/7205
| 2024-10-08T15:06:18 | 2024-10-08T15:25:28 | 2024-10-08T15:25:25 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,573,289,063 | 7,204 |
fix unbatched arrow map for iterable datasets
|
Fixes the bug when applying map to an arrow-formatted iterable dataset described here:
https://github.com/huggingface/datasets/issues/6833#issuecomment-2399903885
```python
from datasets import load_dataset
ds = load_dataset("rotten_tomatoes", split="train", streaming=True)
ds = ds.with_format("arrow").map(lambda x: x)
for ex in ds:
pass
```
@lhoestq
|
closed
|
https://github.com/huggingface/datasets/pull/7204
| 2024-10-08T13:54:09 | 2024-10-08T14:19:47 | 2024-10-08T14:19:47 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,573,154,222 | 7,203 |
with_format docstring
|
reported at https://github.com/huggingface/datasets/issues/3444
|
closed
|
https://github.com/huggingface/datasets/pull/7203
| 2024-10-08T13:05:19 | 2024-10-08T13:13:12 | 2024-10-08T13:13:05 |
{
"login": "lhoestq",
"id": 42851186,
"type": "User"
}
|
[] | true |
[] |
2,572,583,798 | 7,202 |
`from_parquet` return type annotation
|
### Describe the bug
As already posted in https://github.com/microsoft/pylance-release/issues/6534, the correct type hinting fails when building a dataset using the `from_parquet` constructor.
Their suggestion is to comprehensively annotate the method's return type to better align with the docstring information.
### Steps to reproduce the bug
```python
from datasets import Dataset
dataset = Dataset.from_parquet(path_or_paths="file")
dataset.map(lambda x: {"new": x["old"]}, batched=True)
```
### Expected behavior
map is a [valid](https://huggingface.co/docs/datasets/v3.0.1/en/package_reference/main_classes#datasets.Dataset.map), no error should be thrown.
### Environment info
- `datasets` version: 3.0.1
- Platform: macOS-15.0.1-arm64-arm-64bit
- Python version: 3.12.6
- `huggingface_hub` version: 0.25.1
- PyArrow version: 17.0.0
- Pandas version: 2.2.3
- `fsspec` version: 2024.6.1
|
open
|
https://github.com/huggingface/datasets/issues/7202
| 2024-10-08T09:08:10 | 2024-10-08T09:08:10 | null |
{
"login": "saiden89",
"id": 45285915,
"type": "User"
}
|
[] | false |
[] |
2,569,837,015 | 7,201 |
`load_dataset()` of images from a single directory where `train.png` image exists
|
### Describe the bug
Hey!
Firstly, thanks for maintaining such framework!
I had a small issue, where I wanted to load a custom dataset of image+text captioning. I had all of my images in a single directory, and one of the images had the name `train.png`. Then, the loaded dataset had only this image.
I guess it's related to "train" as a split name, but it's definitely an unexpected behavior :)
Unfortunately I don't have time to submit a proper PR. I'm attaching a toy example to reproduce the issue.
Thanks,
Sagi
### Steps to reproduce the bug
All of the steps I'm attaching are in a fresh env :)
```
(base) sagipolaczek@Sagis-MacBook-Pro ~ % conda activate hf_issue_env
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % python --version
Python 3.10.15
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % pip list | grep datasets
datasets 3.0.1
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % ls -la Documents/hf_datasets_issue
total 352
drwxr-xr-x 6 sagipolaczek staff 192 Oct 7 11:59 .
drwx------@ 23 sagipolaczek staff 736 Oct 7 11:46 ..
-rw-r--r--@ 1 sagipolaczek staff 72 Oct 7 11:59 metadata.csv
-rw-r--r--@ 1 sagipolaczek staff 160154 Oct 6 18:00 pika.png
-rw-r--r--@ 1 sagipolaczek staff 5495 Oct 6 12:02 pika_pika.png
-rw-r--r--@ 1 sagipolaczek staff 1753 Oct 6 11:50 train.png
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % cat Documents/hf_datasets_issue/metadata.csv
file_name,text
train.png,A train
pika.png,Pika
pika_pika.png,Pika Pika!
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % python
Python 3.10.15 (main, Oct 3 2024, 02:33:33) [Clang 14.0.6 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from datasets import load_dataset
>>> dataset = load_dataset("imagefolder", data_dir="Documents/hf_datasets_issue/")
>>> dataset
DatasetDict({
train: Dataset({
features: ['image', 'text'],
num_rows: 1
})
})
>>> dataset["train"][0]
{'image': <PIL.PngImagePlugin.PngImageFile image mode=RGBA size=354x84 at 0x10B50FD90>, 'text': 'A train'}
### DELETING `train.png` sample ###
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % vim Documents/hf_datasets_issue/metadata.csv
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % rm Documents/hf_datasets_issue/train.png
(hf_issue_env) sagipolaczek@Sagis-MacBook-Pro ~ % python
Python 3.10.15 (main, Oct 3 2024, 02:33:33) [Clang 14.0.6 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from datasets import load_dataset
>>> dataset = load_dataset("imagefolder", data_dir="Documents/hf_datasets_issue/")
Generating train split: 2 examples [00:00, 65.99 examples/s]
>>> dataset
DatasetDict({
train: Dataset({
features: ['image', 'text'],
num_rows: 2
})
})
>>> dataset["train"]
Dataset({
features: ['image', 'text'],
num_rows: 2
})
>>> dataset["train"][0],dataset["train"][1]
({'image': <PIL.PngImagePlugin.PngImageFile image mode=RGBA size=2356x1054 at 0x10DD11E70>, 'text': 'Pika'}, {'image': <PIL.PngImagePlugin.PngImageFile image mode=RGBA size=343x154 at 0x10E258C70>, 'text': 'Pika Pika!'})
```
### Expected behavior
My expected behavior would be to get a dataset with the sample `train.png` in it (along with the others data points).
### Environment info
I've attached it in the example:
Python 3.10.15
datasets 3.0.1
|
open
|
https://github.com/huggingface/datasets/issues/7201
| 2024-10-07T09:14:17 | 2024-10-07T09:14:17 | null |
{
"login": "SagiPolaczek",
"id": 56922146,
"type": "User"
}
|
[] | false |
[] |
2,567,921,694 | 7,200 |
Fix the environment variable for huggingface cache
|
Resolve #6256. As far as I tested, `HF_DATASETS_CACHE` was ignored and I could not specify the cache directory at all except for the default one by this environment variable. `HF_HOME` has worked. Perhaps the recent change on file downloading by `huggingface_hub` could affect this bug.
In my testing, I could not specify the cache directory even by `load_dataset("dataset_name" cache_dir="...")`. It might be another issue. I also welcome any advice to solve this issue.
|
closed
|
https://github.com/huggingface/datasets/pull/7200
| 2024-10-05T11:54:35 | 2024-10-30T23:10:27 | 2024-10-08T15:45:18 |
{
"login": "torotoki",
"id": 989899,
"type": "User"
}
|
[] | true |
[] |
2,566,788,225 | 7,199 |
Add with_rank to Dataset.from_generator
|
Adds `with_rank` to `Dataset.from_generator`. As for `Dataset.map` and `Dataset.filter`, this is useful when creating cache files using multi-GPU.
|
open
|
https://github.com/huggingface/datasets/pull/7199
| 2024-10-04T16:51:53 | 2024-10-04T16:51:53 | null |
{
"login": "muthissar",
"id": 17828087,
"type": "User"
}
|
[] | true |
[] |
2,566,064,849 | 7,198 |
Add repeat method to datasets
|
Following up on discussion in #6623 and #7198 I thought this would be pretty useful for my case so had a go at implementing.
My main motivation is to be able to call iterable_dataset.repeat(None).take(samples_per_epoch) to safely avoid timeout issues in a distributed training setting. This would provide a straightforward workaround for several open issues related to this situation: https://github.com/huggingface/datasets/issues/6437, https://github.com/huggingface/datasets/issues/6594, https://github.com/huggingface/datasets/issues/6623, https://github.com/huggingface/datasets/issues/6719.
@lhoestq let me know if this looks on the right track!
|
closed
|
https://github.com/huggingface/datasets/pull/7198
| 2024-10-04T10:45:16 | 2025-02-05T16:32:31 | 2025-02-05T16:32:31 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | true |
[] |
2,565,924,788 | 7,197 |
ConnectionError: Couldn't reach 'allenai/c4' on the Hub (ConnectionError)数据集下不下来，怎么回事
|
### Describe the bug
from datasets import load_dataset
print("11")
traindata = load_dataset('ptb_text_only', 'penn_treebank', split='train')
print("22")
valdata = load_dataset('ptb_text_only',
'penn_treebank',
split='validation')
### Steps to reproduce the bug
1
### Expected behavior
1
### Environment info
1
|
open
|
https://github.com/huggingface/datasets/issues/7197
| 2024-10-04T09:33:25 | 2025-02-26T02:26:16 | null |
{
"login": "Mrgengli",
"id": 114299344,
"type": "User"
}
|
[] | false |
[] |
2,564,218,566 | 7,196 |
concatenate_datasets does not preserve shuffling state
|
### Describe the bug
After concatenate datasets on an iterable dataset, the shuffling state is destroyed, similar to #7156
This means concatenation cant be used for resolving uneven numbers of samples across devices when using iterable datasets in a distributed setting as discussed in #6623
I also noticed that the number of shards is the same after concatenation, which I found surprising, but I don't understand the internals well enough to know whether this is actually surprising or not
### Steps to reproduce the bug
```python
import datasets
import torch.utils.data
def gen(shards):
yield {"shards": shards}
def main():
dataset1 = datasets.IterableDataset.from_generator(
gen, gen_kwargs={"shards": list(range(25))} # TODO: how to understand this?
)
dataset2 = datasets.IterableDataset.from_generator(
gen, gen_kwargs={"shards": list(range(25, 50))} # TODO: how to understand this?
)
dataset1 = dataset1.shuffle(buffer_size=1)
dataset2 = dataset2.shuffle(buffer_size=1)
print(dataset1.n_shards)
print(dataset2.n_shards)
dataset = datasets.concatenate_datasets(
[dataset1, dataset2]
)
print(dataset.n_shards)
# dataset = dataset1
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=8,
num_workers=0,
)
for i, batch in enumerate(dataloader):
print(batch)
print("\nNew epoch")
dataset = dataset.set_epoch(1)
for i, batch in enumerate(dataloader):
print(batch)
if __name__ == "__main__":
main()
```
### Expected behavior
Shuffling state should be preserved
### Environment info
Latest datasets
|
open
|
https://github.com/huggingface/datasets/issues/7196
| 2024-10-03T14:30:38 | 2025-03-18T10:56:47 | null |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[] | false |
[] |
2,564,070,809 | 7,195 |
Add support for 3D datasets
|
See https://huggingface.co/datasets/allenai/objaverse for example
|
open
|
https://github.com/huggingface/datasets/issues/7195
| 2024-10-03T13:27:44 | 2024-10-04T09:23:36 | null |
{
"login": "severo",
"id": 1676121,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,563,364,199 | 7,194 |
datasets.exceptions.DatasetNotFoundError for private dataset
|
### Describe the bug
The following Python code tries to download a private dataset and fails with the error `datasets.exceptions.DatasetNotFoundError: Dataset 'ClimatePolicyRadar/all-document-text-data-weekly' doesn't exist on the Hub or cannot be accessed.`. Downloading a public dataset doesn't work.
``` py
from datasets import load_dataset
_ = load_dataset("ClimatePolicyRadar/all-document-text-data-weekly")
```
This seems to be just an issue with my machine config as the code above works with a colleague's machine. So far I have tried:
- logging back out and in from the Huggingface CLI using `huggingface-cli logout`
- manually removing the token cache at `/Users/kalyan/.cache/huggingface/token` (found using `huggingface-cli env`)
- manually passing a token in `load_dataset`
My output of `huggingface-cli whoami`:
```
kdutia
orgs: ClimatePolicyRadar
```
### Steps to reproduce the bug
```
python
Python 3.12.2 (main, Feb 6 2024, 20:19:44) [Clang 15.0.0 (clang-1500.1.0.2.5)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from datasets import load_dataset
>>> _ = load_dataset("ClimatePolicyRadar/all-document-text-data-weekly")
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/Users/kalyan/Library/Caches/pypoetry/virtualenvs/open-data-cnKQNmjn-py3.12/lib/python3.12/site-packages/datasets/load.py", line 2074, in load_dataset
builder_instance = load_dataset_builder(
^^^^^^^^^^^^^^^^^^^^^
File "/Users/kalyan/Library/Caches/pypoetry/virtualenvs/open-data-cnKQNmjn-py3.12/lib/python3.12/site-packages/datasets/load.py", line 1795, in load_dataset_builder
dataset_module = dataset_module_factory(
^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/kalyan/Library/Caches/pypoetry/virtualenvs/open-data-cnKQNmjn-py3.12/lib/python3.12/site-packages/datasets/load.py", line 1659, in dataset_module_factory
raise e1 from None
File "/Users/kalyan/Library/Caches/pypoetry/virtualenvs/open-data-cnKQNmjn-py3.12/lib/python3.12/site-packages/datasets/load.py", line 1597, in dataset_module_factory
raise DatasetNotFoundError(f"Dataset '{path}' doesn't exist on the Hub or cannot be accessed.") from e
datasets.exceptions.DatasetNotFoundError: Dataset 'ClimatePolicyRadar/all-document-text-data-weekly' doesn't exist on the Hub or cannot be accessed.
>>>
```
### Expected behavior
The dataset downloads successfully.
### Environment info
From `huggingface-cli env`:
```
- huggingface_hub version: 0.25.1
- Platform: macOS-14.2.1-arm64-arm-64bit
- Python version: 3.12.2
- Running in iPython ?: No
- Running in notebook ?: No
- Running in Google Colab ?: No
- Running in Google Colab Enterprise ?: No
- Token path ?: /Users/kalyan/.cache/huggingface/token
- Has saved token ?: True
- Who am I ?: kdutia
- Configured git credential helpers: osxkeychain
- FastAI: N/A
- Tensorflow: N/A
- Torch: N/A
- Jinja2: 3.1.4
- Graphviz: N/A
- keras: N/A
- Pydot: N/A
- Pillow: N/A
- hf_transfer: N/A
- gradio: N/A
- tensorboard: N/A
- numpy: 2.1.1
- pydantic: N/A
- aiohttp: 3.10.8
- ENDPOINT: https://huggingface.co
- HF_HUB_CACHE: /Users/kalyan/.cache/huggingface/hub
- HF_ASSETS_CACHE: /Users/kalyan/.cache/huggingface/assets
- HF_TOKEN_PATH: /Users/kalyan/.cache/huggingface/token
- HF_HUB_OFFLINE: False
- HF_HUB_DISABLE_TELEMETRY: False
- HF_HUB_DISABLE_PROGRESS_BARS: None
- HF_HUB_DISABLE_SYMLINKS_WARNING: False
- HF_HUB_DISABLE_EXPERIMENTAL_WARNING: False
- HF_HUB_DISABLE_IMPLICIT_TOKEN: False
- HF_HUB_ENABLE_HF_TRANSFER: False
- HF_HUB_ETAG_TIMEOUT: 10
- HF_HUB_DOWNLOAD_TIMEOUT: 10
```
from `datasets-cli env`:
```
- `datasets` version: 3.0.1
- Platform: macOS-14.2.1-arm64-arm-64bit
- Python version: 3.12.2
- `huggingface_hub` version: 0.25.1
- PyArrow version: 17.0.0
- Pandas version: 2.2.3
- `fsspec` version: 2024.6.1
```
|
closed
|
https://github.com/huggingface/datasets/issues/7194
| 2024-10-03T07:49:36 | 2024-10-03T10:09:28 | 2024-10-03T10:09:28 |
{
"login": "kdutia",
"id": 20212179,
"type": "User"
}
|
[] | false |
[] |
2,562,392,887 | 7,193 |
Support of num_workers (multiprocessing) in map for IterableDataset
|
### Feature request
Currently, IterableDataset doesn't support setting num_worker in .map(), which results in slow processing here. Could we add support for it? As .map() can be run in the batch fashion (e.g., batch_size is default to 1000 in datasets), it seems to be doable for IterableDataset as the regular Dataset.
### Motivation
Improving data processing efficiency
### Your contribution
Testing
|
open
|
https://github.com/huggingface/datasets/issues/7193
| 2024-10-02T18:34:04 | 2024-10-03T09:54:15 | null |
{
"login": "getao",
"id": 12735658,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,562,289,642 | 7,192 |
Add repeat() for iterable datasets
|
### Feature request
It would be useful to be able to straightforwardly repeat iterable datasets indefinitely, to provide complete control over starting and ending of iteration to the user.
An IterableDataset.repeat(n) function could do this automatically
### Motivation
This feature was discussed in this issue https://github.com/huggingface/datasets/issues/7147, and would resolve the need to use the hack of interleave datasets with probability 0 as a simple way to achieve this functionality.
An additional benefit might be the simplification of the use of iterable datasets in a distributed setting:
If the user can assume that datasets will repeat indefinitely, then issues around different numbers of samples appearing on different devices (e.g. https://github.com/huggingface/datasets/issues/6437, https://github.com/huggingface/datasets/issues/6594, https://github.com/huggingface/datasets/issues/6623, https://github.com/huggingface/datasets/issues/6719) can potentially be straightforwardly resolved by simply doing:
ids.repeat(None).take(n_samples_per_epoch)
### Your contribution
I'm not familiar enough with the codebase to assess how straightforward this would be to implement.
If it might be very straightforward, I could possibly have a go.
|
closed
|
https://github.com/huggingface/datasets/issues/7192
| 2024-10-02T17:48:13 | 2025-03-18T10:48:33 | 2025-03-18T10:48:32 |
{
"login": "alex-hh",
"id": 5719745,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,562,206,949 | 7,191 |
Solution to issue: #7080 Modified load_dataset function, so that it prompts the user to select a dataset when subdatasets or splits (train, test) are available
|
# Feel free to give suggestions please..
### This PR is raised because of issue: https://github.com/huggingface/datasets/issues/7080
### This PR gives solution to https://github.com/huggingface/datasets/issues/7080
1. Checking whether the dataset has splits or subdatasets.
2. Printing the available splits/subdatasets.
3. Asking the user to choose which one to load.
4. Loading only the selected dataset based on the user's input.
### Key Changes:
1. Available Splits/Subdatasets: The code checks for available splits/subdatasets using builder_instance.info.splits.keys().
2. User Prompt: If splits are found, it prints them out and prompts the user to select one.
3. Loading Based on User Input: The dataset is loaded based on the user's choice.
This way, the dataset loading function will interactively prompt the user to select which subdataset or split they want to load instead of automatically loading all of them.
|
closed
|
https://github.com/huggingface/datasets/pull/7191
| 2024-10-02T17:02:45 | 2024-11-10T08:48:21 | 2024-11-10T08:48:21 |
{
"login": "negativenagesh",
"id": 148525245,
"type": "User"
}
|
[] | true |
[] |
2,562,162,725 | 7,190 |
Datasets conflicts with fsspec 2024.9
|
### Describe the bug
Installing both in latest versions are not possible
`pip install "datasets==3.0.1" "fsspec==2024.9.0"`
But using older version of datasets is ok
`pip install "datasets==1.24.4" "fsspec==2024.9.0"`
### Steps to reproduce the bug
`pip install "datasets==3.0.1" "fsspec==2024.9.0"`
### Expected behavior
install both versions.
### Environment info
debian 11.
python 3.10.15
|
open
|
https://github.com/huggingface/datasets/issues/7190
| 2024-10-02T16:43:46 | 2024-10-10T07:33:18 | null |
{
"login": "cw-igormorgado",
"id": 162599174,
"type": "User"
}
|
[] | false |
[] |
2,562,152,845 | 7,189 |
Audio preview in dataset viewer for audio array data without a path/filename
|
### Feature request
Huggingface has quite a comprehensive set of guides for [audio datasets](https://huggingface.co/docs/datasets/en/audio_dataset). It seems, however, all these guides assume the audio array data to be decoded/inserted into a HF dataset always originates from individual files. The [Audio-dataclass](https://github.com/huggingface/datasets/blob/3.0.1/src/datasets/features/audio.py#L20) appears designed with this assumption in mind. Looking at its source code it returns a dictionary with the keys `path`, `array` and `sampling_rate`.
However, sometimes users may have different pipelines where they themselves decode the audio array. This feature request has to do with wishing some clarification in guides on whether it is possible, and in such case how users can insert already decoded audio array data into datasets (pandas DataFrame, HF dataset or whatever) that are later saved as parquet, and still get a functioning audio preview in the dataset viewer.
Do I perhaps need to write a tempfile of my audio array slice to wav and capture the bytes object with `io.BytesIO` and pass that to `Audio()`?
### Motivation
I'm working with large audio datasets, and my pipeline reads (decodes) audio from larger files, and slices the relevant portions of audio from that larger file based on metadata I have available.
The pipeline is designed this way to avoid having to store multiple copies of data, and to avoid having to store tens of millions of small files.
I tried [test-uploading parquet files](https://huggingface.co/datasets/Lauler/riksdagen_test) where I store the audio array data of decoded slices of audio in an `audio` column with a dictionary with the keys `path`, `array` and `sampling_rate`. But I don't know the secret sauce of what the Huggingface Hub expects and requires to be able to display audio previews correctly.
### Your contribution
I could contribute a tool agnostic guide of creating HF audio datasets directly as parquet to the HF documentation if there is an interest. Provided you help me figure out the secret sauce of what the dataset viewer expects to display the preview correctly.
|
open
|
https://github.com/huggingface/datasets/issues/7189
| 2024-10-02T16:38:38 | 2024-10-02T17:01:40 | null |
{
"login": "Lauler",
"id": 7157234,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,560,712,689 | 7,188 |
Pin multiprocess<0.70.1 to align with dill<0.3.9
|
Pin multiprocess<0.70.1 to align with dill<0.3.9.
Note that multiprocess-0.70.1 requires dill-0.3.9: https://github.com/uqfoundation/multiprocess/releases/tag/0.70.17
Fix #7186.
|
closed
|
https://github.com/huggingface/datasets/pull/7188
| 2024-10-02T05:40:18 | 2024-10-02T06:08:25 | 2024-10-02T06:08:23 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true |
[] |
2,560,501,308 | 7,187 |
shard_data_sources() got an unexpected keyword argument 'worker_id'
|
### Describe the bug
```
[rank0]: File "/home/qinghao/miniconda3/envs/doremi/lib/python3.10/site-packages/datasets/iterable_dataset.py", line 238, in __iter__
[rank0]: for key_example in islice(self.generate_examples_fn(**gen_kwags), shard_example_idx_start, None):
[rank0]: File "/home/qinghao/miniconda3/envs/doremi/lib/python3.10/site-packages/datasets/packaged_modules/generator/generator.py", line 32, in _generate_examples
[rank0]: for idx, ex in enumerate(self.config.generator(**gen_kwargs)):
[rank0]: File "/home/qinghao/workdir/doremi/doremi/dataloader.py", line 337, in take_data_generator
[rank0]: for ex in ds:
[rank0]: File "/home/qinghao/miniconda3/envs/doremi/lib/python3.10/site-packages/datasets/iterable_dataset.py", line 1791, in __iter__
[rank0]: yield from self._iter_pytorch()
[rank0]: File "/home/qinghao/miniconda3/envs/doremi/lib/python3.10/site-packages/datasets/iterable_dataset.py", line 1704, in _iter_pytorch
[rank0]: ex_iterable = ex_iterable.shard_data_sources(worker_id=worker_info.id, num_workers=worker_info.num_workers)
[rank0]: TypeError: UpdatableRandomlyCyclingMultiSourcesExamplesIterable.shard_data_sources() got an unexpected keyword argument 'worker_id'
```
### Steps to reproduce the bug
IterableDataset cannot use
### Expected behavior
can work on datasets==2.10, but will raise error for later versions.
### Environment info
datasets==3.0.1
|
open
|
https://github.com/huggingface/datasets/issues/7187
| 2024-10-02T01:26:35 | 2024-10-02T01:26:35 | null |
{
"login": "Qinghao-Hu",
"id": 27758466,
"type": "User"
}
|
[] | false |
[] |
2,560,323,917 | 7,186 |
pinning `dill<0.3.9` without pinning `multiprocess`
|
### Describe the bug
The [latest `multiprocess` release](https://github.com/uqfoundation/multiprocess/releases/tag/0.70.17) requires `dill>=0.3.9` which causes issues when installing `datasets` without backtracking during package version resolution. Is it possible to add a pin for multiprocess so something like `multiprocess<=0.70.16` so that the `dill` version is compatible?
### Steps to reproduce the bug
NA
### Expected behavior
NA
### Environment info
NA
|
closed
|
https://github.com/huggingface/datasets/issues/7186
| 2024-10-01T22:29:32 | 2024-10-02T06:08:24 | 2024-10-02T06:08:24 |
{
"login": "shubhbapna",
"id": 38372682,
"type": "User"
}
|
[] | false |
[] |
2,558,508,748 | 7,185 |
CI benchmarks are broken
|
Since Aug 30, 2024, CI benchmarks are broken: https://github.com/huggingface/datasets/actions/runs/11108421214/job/30861323975
```
{"level":"error","message":"Resource not accessible by integration","name":"HttpError","request":{"body":"{\"body\":\"<details>\\n<summary>Show benchmarks</summary>\\n\\nPyArrow==8.0.0\\n\\n<details>\\n<summary>Show updated benchmarks!</summary>\\n\\n### Benchmark: benchmark_array_xd.json\\n\\n| metric | read_batch_formatted_as_numpy after write_array2d |
...
"headers":{"accept":"application/vnd.github.v3+json","authorization":"token [REDACTED]","content-type":"application/json; charset=utf-8","user-agent":"octokit-rest.js/18.0.0 octokit-core.js/3.6.0 Node.js/16.20.2 (linux; x64)"},"method":"POST","request":{"agent":{"_events":{},"_eventsCount":2,"cache":
...
"response":{"data":{"documentation_url":"https://docs.github.com/rest/issues/comments#create-an-issue-comment","message":"Resource not accessible by integration","status":"403"},
...
"stack":"HttpError: Resource not accessible by integration\n at /usr/lib/node_modules/@dvcorg/cml/node_modules/@octokit/request/dist-node/index.js:86:21\n at processTicksAndRejections (node:internal/process/task_queues:96:5)\n at async Job.doExecute (/usr/lib/node_modules/@dvcorg/cml/node_modules/bottleneck/light.js:405:18)","status":403}
```
|
closed
|
https://github.com/huggingface/datasets/issues/7185
| 2024-10-01T08:16:08 | 2024-10-09T16:07:48 | 2024-10-09T16:07:48 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "maintenance",
"color": "d4c5f9"
}
] | false |
[] |
2,556,855,150 | 7,184 |
Pin dill<0.3.9 to fix CI
|
Pin dill<0.3.9 to fix CI for deps-latest.
Note that dill-0.3.9 was released yesterday Sep 29, 2024:
- https://pypi.org/project/dill/0.3.9/
- https://github.com/uqfoundation/dill/releases/tag/0.3.9
Fix #7183.
|
closed
|
https://github.com/huggingface/datasets/pull/7184
| 2024-09-30T14:26:25 | 2024-09-30T14:38:59 | 2024-09-30T14:38:57 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true |
[] |
2,556,789,055 | 7,183 |
CI is broken for deps-latest
|
See: https://github.com/huggingface/datasets/actions/runs/11106149906/job/30853879890
```
=========================== short test summary info ============================
FAILED tests/test_arrow_dataset.py::BaseDatasetTest::test_filter_caching_on_disk - AssertionError: Lists differ: [{'fi[44 chars] {'filename': '/tmp/tmp6xcyyjs4/cache-9533fe2601cd3e48.arrow'}] != [{'fi[44 chars] {'filename': '/tmp/tmp6xcyyjs4/cache-e6e0a8b830976289.arrow'}]
First differing element 1:
{'filename': '/tmp/tmp6xcyyjs4/cache-9533fe2601cd3e48.arrow'}
{'filename': '/tmp/tmp6xcyyjs4/cache-e6e0a8b830976289.arrow'}
[{'filename': '/tmp/tmp6xcyyjs4/dataset0.arrow'},
- {'filename': '/tmp/tmp6xcyyjs4/cache-9533fe2601cd3e48.arrow'}]
? ^^^^^ --------
+ {'filename': '/tmp/tmp6xcyyjs4/cache-e6e0a8b830976289.arrow'}]
? ++++++++++ ^^ +
FAILED tests/test_arrow_dataset.py::BaseDatasetTest::test_map_caching_on_disk - AssertionError: Lists differ: [{'filename': '/tmp/tmp5gxrti_n/cache-e58d327daec8626f.arrow'}] != [{'filename': '/tmp/tmp5gxrti_n/cache-d87234c5763e54a3.arrow'}]
First differing element 0:
{'filename': '/tmp/tmp5gxrti_n/cache-e58d327daec8626f.arrow'}
{'filename': '/tmp/tmp5gxrti_n/cache-d87234c5763e54a3.arrow'}
- [{'filename': '/tmp/tmp5gxrti_n/cache-e58d327daec8626f.arrow'}]
? ^^ -----------
+ [{'filename': '/tmp/tmp5gxrti_n/cache-d87234c5763e54a3.arrow'}]
? +++++++++++ ^^
FAILED tests/test_fingerprint.py::TokenizersHashTest::test_hash_regex - NameError: name 'log' is not defined
FAILED tests/test_fingerprint.py::RecurseHashTest::test_hash_ignores_line_definition_of_function - AssertionError: '52e56ee04ad92499' != '0a4f75cec280f634'
- 52e56ee04ad92499
+ 0a4f75cec280f634
FAILED tests/test_fingerprint.py::RecurseHashTest::test_hash_ipython_function - AssertionError: 'a6bd2041ca63d6c0' != '517bf36b7eecdef5'
- a6bd2041ca63d6c0
+ 517bf36b7eecdef5
FAILED tests/test_fingerprint.py::HashingTest::test_hash_tiktoken_encoding - NameError: name 'log' is not defined
FAILED tests/test_fingerprint.py::HashingTest::test_hash_torch_compiled_module - NameError: name 'log' is not defined
FAILED tests/test_fingerprint.py::HashingTest::test_hash_torch_generator - NameError: name 'log' is not defined
FAILED tests/test_fingerprint.py::HashingTest::test_hash_torch_tensor - NameError: name 'log' is not defined
FAILED tests/test_fingerprint.py::HashingTest::test_set_doesnt_depend_on_order - NameError: name 'log' is not defined
FAILED tests/test_fingerprint.py::HashingTest::test_set_stable - NameError: name 'log' is not defined
ERROR tests/test_iterable_dataset.py::test_iterable_dataset_from_file - NameError: name 'log' is not defined
= 11 failed, 2850 passed, 3 skipped, 23 warnings, 1 error in 191.06s (0:03:11) =
```
|
closed
|
https://github.com/huggingface/datasets/issues/7183
| 2024-09-30T14:02:07 | 2024-09-30T14:38:58 | 2024-09-30T14:38:58 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | false |
[] |
2,556,333,671 | 7,182 |
Support features in metadata configs
|
Support features in metadata configs, like:
```
configs:
- config_name: default
features:
- name: id
dtype: int64
- name: name
dtype: string
- name: score
dtype: float64
```
This will allow to avoid inference of data types.
Currently, we allow passing this information in the `dataset_info` (instead of `configs`) field, but this is not intuitive and it is not properly documented.
TODO:
- [ ] Document usage
|
closed
|
https://github.com/huggingface/datasets/pull/7182
| 2024-09-30T11:14:53 | 2024-10-09T16:03:57 | 2024-10-09T16:03:54 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true |
[] |
2,554,917,019 | 7,181 |
Fix datasets export to JSON
| null |
closed
|
https://github.com/huggingface/datasets/pull/7181
| 2024-09-29T12:45:20 | 2024-11-01T11:55:36 | 2024-11-01T11:55:36 |
{
"login": "varadhbhatnagar",
"id": 20443618,
"type": "User"
}
|
[] | true |
[] |
2,554,244,750 | 7,180 |
Memory leak when wrapping datasets into PyTorch Dataset without explicit deletion
|
### Describe the bug
I've encountered a memory leak when wrapping the HuggingFace dataset into a PyTorch Dataset. The RAM usage constantly increases during iteration if items are not explicitly deleted after use.
### Steps to reproduce the bug
Steps to reproduce:
Create a PyTorch Dataset wrapper for 'nebula/cc12m':
````
from torch.utils.data import Dataset
from tqdm import tqdm
from datasets import load_dataset
from torchvision import transforms
Image.MAX_IMAGE_PIXELS = None
class CC12M(Dataset):
def __init__(self, path_or_name='nebula/cc12m', split='train', transform=None, single_caption=True):
self.raw_dataset = load_dataset(path_or_name)[split]
if transform is None:
self.transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.48145466, 0.4578275, 0.40821073],
std=[0.26862954, 0.26130258, 0.27577711]
)
])
else:
self.transform = transforms.Compose(transform)
self.single_caption = single_caption
self.length = len(self.raw_dataset)
def __len__(self):
return self.length
def __getitem__(self, index):
item = self.raw_dataset[index]
caption = item['txt']
with io.BytesIO(item['webp']) as buffer:
image = Image.open(buffer).convert('RGB')
if self.transform:
image = self.transform(image)
# del item # Uncomment this line to prevent the memory leak
return image, caption
````
Iterate through the dataset without the del item line in __getitem__.
Observe RAM usage increasing constantly.
Add del item at the end of __getitem__:
```
def __getitem__(self, index):
item = self.raw_dataset[index]
caption = item['txt']
with io.BytesIO(item['webp']) as buffer:
image = Image.open(buffer).convert('RGB')
if self.transform:
image = self.transform(image)
del item # This line prevents the memory leak
return image, caption
```
Iterate through the dataset again and observe that RAM usage remains stable.
### Expected behavior
Expected behavior:
RAM usage should remain stable during iteration without needing to explicitly delete items.
Actual behavior:
RAM usage constantly increases unless items are explicitly deleted after use
### Environment info
- `datasets` version: 2.21.0
- Platform: Linux-4.18.0-513.5.1.el8_9.x86_64-x86_64-with-glibc2.28
- Python version: 3.12.4
- `huggingface_hub` version: 0.24.6
- PyArrow version: 17.0.0
- Pandas version: 2.2.2
- `fsspec` version: 2024.6.1
|
closed
|
https://github.com/huggingface/datasets/issues/7180
| 2024-09-28T14:00:47 | 2024-09-30T12:07:56 | 2024-09-30T12:07:56 |
{
"login": "iamwangyabin",
"id": 38123329,
"type": "User"
}
|
[] | false |
[] |
2,552,387,980 | 7,179 |
Support Python 3.11
|
Support Python 3.11.
Fix #7178.
|
closed
|
https://github.com/huggingface/datasets/pull/7179
| 2024-09-27T08:55:44 | 2024-10-08T16:21:06 | 2024-10-08T16:21:03 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true |
[] |
2,552,378,330 | 7,178 |
Support Python 3.11
|
Support Python 3.11: https://peps.python.org/pep-0664/
|
closed
|
https://github.com/huggingface/datasets/issues/7178
| 2024-09-27T08:50:47 | 2024-10-08T16:21:04 | 2024-10-08T16:21:04 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[
{
"name": "enhancement",
"color": "a2eeef"
}
] | false |
[] |
2,552,371,082 | 7,177 |
Fix release instructions
|
Fix release instructions.
During last release, I had to make this additional update.
|
closed
|
https://github.com/huggingface/datasets/pull/7177
| 2024-09-27T08:47:01 | 2024-09-27T08:57:35 | 2024-09-27T08:57:32 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true |
[] |
2,551,025,564 | 7,176 |
fix grammar in fingerprint.py
|
I see this error all the time and it was starting to get to me.
|
open
|
https://github.com/huggingface/datasets/pull/7176
| 2024-09-26T16:13:42 | 2024-09-26T16:13:42 | null |
{
"login": "jxmorris12",
"id": 13238952,
"type": "User"
}
|
[] | true |
[] |
2,550,957,337 | 7,175 |
[FSTimeoutError] load_dataset
|
### Describe the bug
When using `load_dataset`to load [HuggingFaceM4/VQAv2](https://huggingface.co/datasets/HuggingFaceM4/VQAv2), I am getting `FSTimeoutError`.
### Error
```
TimeoutError:
The above exception was the direct cause of the following exception:
FSTimeoutError Traceback (most recent call last)
[/usr/local/lib/python3.10/dist-packages/fsspec/asyn.py](https://klh9mr78js-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab_20240924-060116_RC00_678132060#) in sync(loop, func, timeout, *args, **kwargs)
99 if isinstance(return_result, asyncio.TimeoutError):
100 # suppress asyncio.TimeoutError, raise FSTimeoutError
--> 101 raise FSTimeoutError from return_result
102 elif isinstance(return_result, BaseException):
103 raise return_result
FSTimeoutError:
```
It usually fails around 5-6 GB.
<img width="847" alt="Screenshot 2024-09-26 at 9 10 19 PM" src="https://github.com/user-attachments/assets/ff91995a-fb55-4de6-8214-94025d6c8470">
### Steps to reproduce the bug
To reproduce it, run this in colab notebook:
```
!pip install -q -U datasets
from datasets import load_dataset
ds = load_dataset('HuggingFaceM4/VQAv2', split="train[:10%]")
```
### Expected behavior
It should download properly.
### Environment info
Using Colab Notebook.
|
closed
|
https://github.com/huggingface/datasets/issues/7175
| 2024-09-26T15:42:29 | 2025-02-01T09:09:35 | 2024-09-30T17:28:35 |
{
"login": "cosmo3769",
"id": 53268607,
"type": "User"
}
|
[] | false |
[] |
2,549,892,315 | 7,174 |
Set dev version
| null |
closed
|
https://github.com/huggingface/datasets/pull/7174
| 2024-09-26T08:30:11 | 2024-09-26T08:32:39 | 2024-09-26T08:30:21 |
{
"login": "albertvillanova",
"id": 8515462,
"type": "User"
}
|
[] | true |
[] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.