url stringlengths 58 61 | number int64 1 8.23k | title stringlengths 1 290 | body stringlengths 0 228k ⌀ | state stringclasses 2
values | created_at timestamp[s]date 2020-04-14 10:18:02 2026-05-30 09:38:59 | comments_url stringlengths 67 70 | pull_request dict | is_pull_request bool 2
classes | text stringlengths 2 228k | comments listlengths 0 0 |
|---|---|---|---|---|---|---|---|---|---|---|
https://api.github.com/repos/huggingface/datasets/issues/8228 | 8,228 | fix: stable fingerprint for closures capturing non-deterministic state | Fixes #7986
When `.map()` uses a closure capturing `self` with non-deterministic state
(like UUIDs or loggers), the fingerprint changes every run causing cache misses.
Fix: only hash primitive attributes of captured objects instead of the full object.
Added a regression test in `tests/test_fingerprint.py`. | open | 2026-05-30T09:38:59 | https://api.github.com/repos/huggingface/datasets/issues/8228/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8228",
"html_url": "https://github.com/huggingface/datasets/pull/8228",
"diff_url": "https://github.com/huggingface/datasets/pull/8228.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8228.patch",
"merged_at": null
} | true | fix: stable fingerprint for closures capturing non-deterministic state
Fixes #7986
When `.map()` uses a closure capturing `self` with non-deterministic state
(like UUIDs or loggers), the fingerprint changes every run causing cache misses.
Fix: only hash primitive attributes of captured objects instead of the fu... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8227 | 8,227 | Security/Logic Fix: Autonomous Code Review | ## Autonomous Bug Report & Patch
This vulnerability and fix were autonomously discovered by the Lucy Red Team swarm.
Upon reviewing the provided code snippet from `src/datasets/io/generator.py`, I didn't find any obvious critical bugs. However, there are a few areas that could be improved or clarified:
1. **Do... | open | 2026-05-29T18:50:22 | https://api.github.com/repos/huggingface/datasets/issues/8227/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8227",
"html_url": "https://github.com/huggingface/datasets/pull/8227",
"diff_url": "https://github.com/huggingface/datasets/pull/8227.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8227.patch",
"merged_at": null
} | true | Security/Logic Fix: Autonomous Code Review
## Autonomous Bug Report & Patch
This vulnerability and fix were autonomously discovered by the Lucy Red Team swarm.
Upon reviewing the provided code snippet from `src/datasets/io/generator.py`, I didn't find any obvious critical bugs. However, there are a few areas that... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8225 | 8,225 | Fix single lance file form pylance 7.0 | fix lance test in CI, due to JSON being picked up instead of lance now that latest_version_hint.json is a thing | closed | 2026-05-28T15:27:04 | https://api.github.com/repos/huggingface/datasets/issues/8225/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8225",
"html_url": "https://github.com/huggingface/datasets/pull/8225",
"diff_url": "https://github.com/huggingface/datasets/pull/8225.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8225.patch",
"merged_at": "2026-05-28T15:33... | true | Fix single lance file form pylance 7.0
fix lance test in CI, due to JSON being picked up instead of lance now that latest_version_hint.json is a thing | [] |
https://api.github.com/repos/huggingface/datasets/issues/8224 | 8,224 | fix: embed_external_files=True for mesh support | ## Fix for #8055
This is follow-up PR for fixing a specific issue related to `dataset.push_to_hub(repo_id, embed_external_files=True)`
### Test Conducted
```python
from datasets import Features, Image, Mesh, Value, load_dataset
repo_id = "VINAY-UMRETHE/My-Mesh-Dataset"
dataset = load_dataset("json", ... | open | 2026-05-27T17:14:11 | https://api.github.com/repos/huggingface/datasets/issues/8224/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8224",
"html_url": "https://github.com/huggingface/datasets/pull/8224",
"diff_url": "https://github.com/huggingface/datasets/pull/8224.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8224.patch",
"merged_at": null
} | true | fix: embed_external_files=True for mesh support
## Fix for #8055
This is follow-up PR for fixing a specific issue related to `dataset.push_to_hub(repo_id, embed_external_files=True)`
### Test Conducted
```python
from datasets import Features, Image, Mesh, Value, load_dataset
repo_id = "VINAY-UMRETHE/My-... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8223 | 8,223 | DOC: document missing save_infos and max_depth parameters | ## What this fixes
Two public API functions were missing parameters from their docstrings:
### 1. `load_dataset()` — missing `save_infos`
The `save_infos` parameter has been in the function signature but was absent from the `Args` section of the docstring. When `save_infos=True`, it overrides `verification_mod... | open | 2026-05-27T07:04:13 | https://api.github.com/repos/huggingface/datasets/issues/8223/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8223",
"html_url": "https://github.com/huggingface/datasets/pull/8223",
"diff_url": "https://github.com/huggingface/datasets/pull/8223.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8223.patch",
"merged_at": null
} | true | DOC: document missing save_infos and max_depth parameters
## What this fixes
Two public API functions were missing parameters from their docstrings:
### 1. `load_dataset()` — missing `save_infos`
The `save_infos` parameter has been in the function signature but was absent from the `Args` section of the docstri... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8222 | 8,222 | Validate split name before starting download | ## What this fixes
Closes #5523.
When you pass an invalid split name to `load_dataset()`, the error only surfaces after the entire dataset has already been downloaded. For large datasets that can mean waiting many minutes before seeing a one-liner like:
```python
from datasets import load_dataset
load_dataset("mozil... | open | 2026-05-27T01:54:14 | https://api.github.com/repos/huggingface/datasets/issues/8222/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8222",
"html_url": "https://github.com/huggingface/datasets/pull/8222",
"diff_url": "https://github.com/huggingface/datasets/pull/8222.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8222.patch",
"merged_at": null
} | true | Validate split name before starting download
## What this fixes
Closes #5523.
When you pass an invalid split name to `load_dataset()`, the error only surfaces after the entire dataset has already been downloaded. For large datasets that can mean waiting many minutes before seeing a one-liner like:
```python
from dat... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8221 | 8,221 | chore: enable Dependabot weekly GitHub Actions bumps | ## Summary
Adds `.github/dependabot.yml` so this repo's pinned GitHub Action SHAs
get bumped automatically once a week.
All action updates are **grouped into one weekly PR** (not one PR per
action) to keep the noise down, and Dependabot waits **7 days after a
release before opening the bump** (cooldown). The 7-day co... | open | 2026-05-26T10:31:28 | https://api.github.com/repos/huggingface/datasets/issues/8221/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8221",
"html_url": "https://github.com/huggingface/datasets/pull/8221",
"diff_url": "https://github.com/huggingface/datasets/pull/8221.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8221.patch",
"merged_at": null
} | true | chore: enable Dependabot weekly GitHub Actions bumps
## Summary
Adds `.github/dependabot.yml` so this repo's pinned GitHub Action SHAs
get bumped automatically once a week.
All action updates are **grouped into one weekly PR** (not one PR per
action) to keep the noise down, and Dependabot waits **7 days after a
relea... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8220 | 8,220 | Support composed splits in streaming datasets | Fixes #2699
Fixes #4804
This PR adds support for unsliced split composition when loading datasets in streaming mode, e.g. `split="train+validation"`.
Previously, `DatasetBuilder.as_streaming_dataset()` only accepted a single split name or returned all splits as an `IterableDatasetDict`, so composed split strings... | open | 2026-05-23T16:41:38 | https://api.github.com/repos/huggingface/datasets/issues/8220/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8220",
"html_url": "https://github.com/huggingface/datasets/pull/8220",
"diff_url": "https://github.com/huggingface/datasets/pull/8220.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8220.patch",
"merged_at": null
} | true | Support composed splits in streaming datasets
Fixes #2699
Fixes #4804
This PR adds support for unsliced split composition when loading datasets in streaming mode, e.g. `split="train+validation"`.
Previously, `DatasetBuilder.as_streaming_dataset()` only accepted a single split name or returned all splits as an `I... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8219 | 8,219 | Add `.conll` / `.conllu` dataset format loader (CoNLL-2003 / 2000 / U) | Closes #7757.
## Summary
Adds a `packaged_modules/conll/` builder so `.conll` and `.conllu` files load directly via `load_dataset(...)` without manual parsing scripts. Each row of the loaded dataset corresponds to one sentence, with each configured column produced as a list aligned with the token list.
Builder... | closed | 2026-05-23T12:19:24 | https://api.github.com/repos/huggingface/datasets/issues/8219/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8219",
"html_url": "https://github.com/huggingface/datasets/pull/8219",
"diff_url": "https://github.com/huggingface/datasets/pull/8219.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8219.patch",
"merged_at": "2026-05-27T14:09... | true | Add `.conll` / `.conllu` dataset format loader (CoNLL-2003 / 2000 / U)
Closes #7757.
## Summary
Adds a `packaged_modules/conll/` builder so `.conll` and `.conllu` files load directly via `load_dataset(...)` without manual parsing scripts. Each row of the loaded dataset corresponds to one sentence, with each confi... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8218 | 8,218 | Include version in DatasetInfo YAML so push_to_hub preserves it | DatasetInfo._INCLUDED_INFO_IN_YAML controls which fields land in the README.md dataset_info block written by push_to_hub, and it currently omits version. Setting version on a Dataset (e.g. via load_dataset(..., version=...)) and pushing it to the Hub silently drops the version from the rendered front matter, so reloadi... | open | 2026-05-22T10:58:49 | https://api.github.com/repos/huggingface/datasets/issues/8218/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8218",
"html_url": "https://github.com/huggingface/datasets/pull/8218",
"diff_url": "https://github.com/huggingface/datasets/pull/8218.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8218.patch",
"merged_at": null
} | true | Include version in DatasetInfo YAML so push_to_hub preserves it
DatasetInfo._INCLUDED_INFO_IN_YAML controls which fields land in the README.md dataset_info block written by push_to_hub, and it currently omits version. Setting version on a Dataset (e.g. via load_dataset(..., version=...)) and pushing it to the Hub silen... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8217 | 8,217 | Multi-GPU Training with Large Dataset — NCCL Timeouts / Slow Data Loading + Excessive Memory Usage with keep_in_memory=True | Hi, I'm training a large dataset using Lerobot (which relies on Hugging Face Datasets) and encountering severe performance issues when scaling to multi-GPU training.
### Problem Description
- Unstable/slow data loading + NCCL timeouts
- Step time fluctuates drastically: ~1s/step (fast) vs ~45s/step (extremely slow)
... | open | 2026-05-22T09:55:13 | https://api.github.com/repos/huggingface/datasets/issues/8217/comments | null | false | Multi-GPU Training with Large Dataset — NCCL Timeouts / Slow Data Loading + Excessive Memory Usage with keep_in_memory=True
Hi, I'm training a large dataset using Lerobot (which relies on Hugging Face Datasets) and encountering severe performance issues when scaling to multi-GPU training.
### Problem Description
- Uns... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8216 | 8,216 | Read cached dataset_info.json to populate config_names offline | ## What
`get_dataset_config_names` with `HF_DATASETS_OFFLINE=1` returns `['default']` even when the cache contains multiple config subdirectories for the dataset. For example, after caching `cais/mmlu`:
```
$ HF_DATASETS_OFFLINE=1 python -c "import datasets; print(datasets.get_dataset_config_names('cais/mmlu'))"
Usin... | open | 2026-05-21T23:27:32 | https://api.github.com/repos/huggingface/datasets/issues/8216/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8216",
"html_url": "https://github.com/huggingface/datasets/pull/8216",
"diff_url": "https://github.com/huggingface/datasets/pull/8216.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8216.patch",
"merged_at": null
} | true | Read cached dataset_info.json to populate config_names offline
## What
`get_dataset_config_names` with `HF_DATASETS_OFFLINE=1` returns `['default']` even when the cache contains multiple config subdirectories for the dataset. For example, after caching `cais/mmlu`:
```
$ HF_DATASETS_OFFLINE=1 python -c "import datase... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8215 | 8,215 | Warn instead of raise when user-provided data_files yields a subset | Fixes #7867.
NonMatchingSplitsSizesError currently fires whenever the loaded split size differs from the expected size, including when the user explicitly passed data_files for a known subset of the dataset. The only user-side workaround is verification_mode='no_checks', which silences ALL checks rather than just the ... | open | 2026-05-21T23:23:19 | https://api.github.com/repos/huggingface/datasets/issues/8215/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8215",
"html_url": "https://github.com/huggingface/datasets/pull/8215",
"diff_url": "https://github.com/huggingface/datasets/pull/8215.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8215.patch",
"merged_at": null
} | true | Warn instead of raise when user-provided data_files yields a subset
Fixes #7867.
NonMatchingSplitsSizesError currently fires whenever the loaded split size differs from the expected size, including when the user explicitly passed data_files for a known subset of the dataset. The only user-side workaround is verificati... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8214 | 8,214 | feat: Add return_file_name parameter to JSON builder (#5806) | ## Description
Resolves #5806.
This PR introduces an optional `return_file_name` boolean parameter to the packaged JSON builder. When enabled, it injects a `file_name` column into the generated PyArrow tables, allowing downstream users to track exactly which data shard a specific row originated from during interrup... | open | 2026-05-21T14:37:14 | https://api.github.com/repos/huggingface/datasets/issues/8214/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8214",
"html_url": "https://github.com/huggingface/datasets/pull/8214",
"diff_url": "https://github.com/huggingface/datasets/pull/8214.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8214.patch",
"merged_at": null
} | true | feat: Add return_file_name parameter to JSON builder (#5806)
## Description
Resolves #5806.
This PR introduces an optional `return_file_name` boolean parameter to the packaged JSON builder. When enabled, it injects a `file_name` column into the generated PyArrow tables, allowing downstream users to track exactly wh... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8213 | 8,213 | update single seg repos in ci | needed for https://github.com/huggingface/huggingface_hub/releases/tag/v1.16.0.rc0 | closed | 2026-05-21T11:05:56 | https://api.github.com/repos/huggingface/datasets/issues/8213/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8213",
"html_url": "https://github.com/huggingface/datasets/pull/8213",
"diff_url": "https://github.com/huggingface/datasets/pull/8213.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8213.patch",
"merged_at": "2026-05-21T11:07... | true | update single seg repos in ci
needed for https://github.com/huggingface/huggingface_hub/releases/tag/v1.16.0.rc0 | [] |
https://api.github.com/repos/huggingface/datasets/issues/8212 | 8,212 | docs: surface `repository_structure` earlier in the Get Started flow | > **Dear maintainer** — this PR has a permanent home with methodology + [opt-out](https://github.com/adv0r/tokens-for-good/blob/main/MAINTAINER_REMOVAL.md) at [tokens-for-good](https://github.com/adv0r/tokens-for-good). A one-line "no thanks" → auto-close + blacklist. Sorry for the notification this edit caused.
>
> --... | open | 2026-05-20T15:45:58 | https://api.github.com/repos/huggingface/datasets/issues/8212/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8212",
"html_url": "https://github.com/huggingface/datasets/pull/8212",
"diff_url": "https://github.com/huggingface/datasets/pull/8212.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8212.patch",
"merged_at": null
} | true | docs: surface `repository_structure` earlier in the Get Started flow
> **Dear maintainer** — this PR has a permanent home with methodology + [opt-out](https://github.com/adv0r/tokens-for-good/blob/main/MAINTAINER_REMOVAL.md) at [tokens-for-good](https://github.com/adv0r/tokens-for-good). A one-line "no thanks" → auto-c... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8211 | 8,211 | Fix duplicate keyword arguments | null | closed | 2026-05-20T05:57:38 | https://api.github.com/repos/huggingface/datasets/issues/8211/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8211",
"html_url": "https://github.com/huggingface/datasets/pull/8211",
"diff_url": "https://github.com/huggingface/datasets/pull/8211.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8211.patch",
"merged_at": null
} | true | Fix duplicate keyword arguments
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8210 | 8,210 | Fix parquet columns arg | Fix the error message "The columns and features argument must contain the same columns, but got.." when passing `columns=` to load_dataset on a streaming parquet dataset (with features declared in the README.md)
useful e.g. for https://huggingface.co/datasets/jasperai/monet to load the thumbnails without the embeddi... | closed | 2026-05-19T16:28:28 | https://api.github.com/repos/huggingface/datasets/issues/8210/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8210",
"html_url": "https://github.com/huggingface/datasets/pull/8210",
"diff_url": "https://github.com/huggingface/datasets/pull/8210.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8210.patch",
"merged_at": "2026-05-20T10:32... | true | Fix parquet columns arg
Fix the error message "The columns and features argument must contain the same columns, but got.." when passing `columns=` to load_dataset on a streaming parquet dataset (with features declared in the README.md)
useful e.g. for https://huggingface.co/datasets/jasperai/monet to load the thumbn... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8209 | 8,209 | Preserve float columns when JSON loader uses `field=` | Closes #6937.
When `load_dataset("json", data_files=..., field="data", ...)` is used, columns whose values are all integer-valued floats (`[0.0, 1.0, 2.0]`) get silently coerced to `int64`. Repro:
```python
import tempfile, json
from datasets import load_dataset
with tempfile.NamedTemporaryFile(mode="w", suffix=".js... | open | 2026-05-18T21:51:34 | https://api.github.com/repos/huggingface/datasets/issues/8209/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8209",
"html_url": "https://github.com/huggingface/datasets/pull/8209",
"diff_url": "https://github.com/huggingface/datasets/pull/8209.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8209.patch",
"merged_at": null
} | true | Preserve float columns when JSON loader uses `field=`
Closes #6937.
When `load_dataset("json", data_files=..., field="data", ...)` is used, columns whose values are all integer-valued floats (`[0.0, 1.0, 2.0]`) get silently coerced to `int64`. Repro:
```python
import tempfile, json
from datasets import load_dataset
... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8208 | 8,208 | update readme | null | closed | 2026-05-18T17:32:31 | https://api.github.com/repos/huggingface/datasets/issues/8208/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8208",
"html_url": "https://github.com/huggingface/datasets/pull/8208",
"diff_url": "https://github.com/huggingface/datasets/pull/8208.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8208.patch",
"merged_at": "2026-05-20T10:32... | true | update readme
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8207 | 8,207 | fix docs: use AutoImageProcessor for vision model examples | swaps AutoFeatureExtractor for AutoImageProcessor in the image preprocessing section of use_dataset.mdx — aligns with the transformers library direction where vision feature extractor references were already replaced in their own docs (see PRs #20111 and #20501). audio section is untouched since AutoFeatureExtractor is... | open | 2026-05-18T13:03:41 | https://api.github.com/repos/huggingface/datasets/issues/8207/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8207",
"html_url": "https://github.com/huggingface/datasets/pull/8207",
"diff_url": "https://github.com/huggingface/datasets/pull/8207.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8207.patch",
"merged_at": null
} | true | fix docs: use AutoImageProcessor for vision model examples
swaps AutoFeatureExtractor for AutoImageProcessor in the image preprocessing section of use_dataset.mdx — aligns with the transformers library direction where vision feature extractor references were already replaced in their own docs (see PRs #20111 and #20501... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8206 | 8,206 | Fix duplicate keyword arguments | ## Summary
This PR fixes a duplicate keyword argument collision in `load_dataset_builder()` when the same parameter is passed through both `builder_kwargs` and `config_kwargs`.
Previously, calling `load_dataset()` or `load_dataset_builder()` with arguments such as `base_path` could raise:
```python
TypeError:... | closed | 2026-05-17T14:53:18 | https://api.github.com/repos/huggingface/datasets/issues/8206/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8206",
"html_url": "https://github.com/huggingface/datasets/pull/8206",
"diff_url": "https://github.com/huggingface/datasets/pull/8206.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8206.patch",
"merged_at": null
} | true | Fix duplicate keyword arguments
## Summary
This PR fixes a duplicate keyword argument collision in `load_dataset_builder()` when the same parameter is passed through both `builder_kwargs` and `config_kwargs`.
Previously, calling `load_dataset()` or `load_dataset_builder()` with arguments such as `base_path` could... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8205 | 8,205 | fix: pass token_per_repo_id to embed_table_storage when pushing shards | ## Summary
- Fixes #6348
- When `push_to_hub` is called with a `token`, the per-repo token map (`token_per_repo_id`) is built correctly earlier in the function — but it was never forwarded to `embed_table_storage` during the shard-map step.
- This caused a one-line fix: wrap `embed_table_storage` with `functools.parti... | closed | 2026-05-16T15:12:27 | https://api.github.com/repos/huggingface/datasets/issues/8205/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8205",
"html_url": "https://github.com/huggingface/datasets/pull/8205",
"diff_url": "https://github.com/huggingface/datasets/pull/8205.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8205.patch",
"merged_at": null
} | true | fix: pass token_per_repo_id to embed_table_storage when pushing shards
## Summary
- Fixes #6348
- When `push_to_hub` is called with a `token`, the per-repo token map (`token_per_repo_id`) is built correctly earlier in the function — but it was never forwarded to `embed_table_storage` during the shard-map step.
- This ... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8204 | 8,204 | Please enable Private Vulnerability Reporting | Hi datasets team,
I have a security finding in datasets that I would like to disclose privately. GitHub Private Vulnerability Reporting is currently disabled on this repo.
Could you turn it on? It's at Settings > Security > Code security and analysis > Private vulnerability reporting. Once enabled I will file the ful... | open | 2026-05-16T13:28:01 | https://api.github.com/repos/huggingface/datasets/issues/8204/comments | null | false | Please enable Private Vulnerability Reporting
Hi datasets team,
I have a security finding in datasets that I would like to disclose privately. GitHub Private Vulnerability Reporting is currently disabled on this repo.
Could you turn it on? It's at Settings > Security > Code security and analysis > Private vulnerabili... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8203 | 8,203 | fix(webdataset): when loading data in WebDataset format using load_datasets during multi-matchines training. | ## Description
Fixes #8201
**The Bug:**
During distributed training with `streaming=True`, if a user provides fewer `.tar` files than the number of distributed ranks, some ranks receive an empty list of files. This caused an `IndexError: list index out of range` in `webdataset.py` because the code assumed `tar_pa... | open | 2026-05-16T06:41:37 | https://api.github.com/repos/huggingface/datasets/issues/8203/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8203",
"html_url": "https://github.com/huggingface/datasets/pull/8203",
"diff_url": "https://github.com/huggingface/datasets/pull/8203.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8203.patch",
"merged_at": null
} | true | fix(webdataset): when loading data in WebDataset format using load_datasets during multi-matchines training.
## Description
Fixes #8201
**The Bug:**
During distributed training with `streaming=True`, if a user provides fewer `.tar` files than the number of distributed ranks, some ranks receive an empty list of fi... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8202 | 8,202 | fix pathlib.Path support in save_to_disk and load_from_disk | `url_to_fs` from fsspec only accepts strings, but `save_to_disk` and `load_from_disk` in `arrow_dataset.py`, `dataset_dict.py`, and `load.py` pass the user-supplied path directly without converting it, causing a crash when a `pathlib.Path` is provided. Added `str()` conversion at each `url_to_fs` call site so all path-... | open | 2026-05-16T01:03:23 | https://api.github.com/repos/huggingface/datasets/issues/8202/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8202",
"html_url": "https://github.com/huggingface/datasets/pull/8202",
"diff_url": "https://github.com/huggingface/datasets/pull/8202.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8202.patch",
"merged_at": null
} | true | fix pathlib.Path support in save_to_disk and load_from_disk
`url_to_fs` from fsspec only accepts strings, but `save_to_disk` and `load_from_disk` in `arrow_dataset.py`, `dataset_dict.py`, and `load.py` pass the user-supplied path directly without converting it, causing a crash when a `pathlib.Path` is provided. Added `... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8201 | 8,201 | Encountered an error when loading data in WebDataset format using `load_datasets` during multi-matchines training. | ### Describe the bug
Single-machine training works fine, but multi-machine training throws up all sorts of weird bugs. Help me!
### Steps to reproduce the bug
train_dataset = load_dataset(
"webdataset", data_files=args.train_dataset, split="train", streaming=True,cache_dir='/dev/shm/.cache'
)
args.train... | open | 2026-05-15T19:21:24 | https://api.github.com/repos/huggingface/datasets/issues/8201/comments | null | false | Encountered an error when loading data in WebDataset format using `load_datasets` during multi-matchines training.
### Describe the bug
Single-machine training works fine, but multi-machine training throws up all sorts of weird bugs. Help me!
### Steps to reproduce the bug
train_dataset = load_dataset(
"webd... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8200 | 8,200 | replace AutoFeatureExtractor with AutoImageProcessor in docs | Closes #8199
## What does this PR do?
Replaces `AutoFeatureExtractor` with `AutoImageProcessor` in the image augmentation example in `docs/source/use_dataset.mdx`. For vision models, `AutoImageProcessor` is the recommended API and better reflects current `transformers` conventions.
This is consistent with the ... | open | 2026-05-15T17:45:15 | https://api.github.com/repos/huggingface/datasets/issues/8200/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8200",
"html_url": "https://github.com/huggingface/datasets/pull/8200",
"diff_url": "https://github.com/huggingface/datasets/pull/8200.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8200.patch",
"merged_at": null
} | true | replace AutoFeatureExtractor with AutoImageProcessor in docs
Closes #8199
## What does this PR do?
Replaces `AutoFeatureExtractor` with `AutoImageProcessor` in the image augmentation example in `docs/source/use_dataset.mdx`. For vision models, `AutoImageProcessor` is the recommended API and better reflects curren... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8199 | 8,199 | Replace AutoFeatureExtractor with AutoImageProcessor in image preprocessing docs | ## Summary
The `docs/source/use_dataset.mdx` documentation uses `AutoFeatureExtractor` from `transformers` in its image preprocessing example. For vision models, `AutoImageProcessor` is now the recommended API and better reflects current `transformers` conventions.
## Problem
In the [[Apply data augmentations](https... | open | 2026-05-15T17:34:53 | https://api.github.com/repos/huggingface/datasets/issues/8199/comments | null | false | Replace AutoFeatureExtractor with AutoImageProcessor in image preprocessing docs
## Summary
The `docs/source/use_dataset.mdx` documentation uses `AutoFeatureExtractor` from `transformers` in its image preprocessing example. For vision models, `AutoImageProcessor` is now the recommended API and better reflects current ... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8198 | 8,198 | Fix/type hints list to sequence | Fix #5354 Consider using "Sequence" instead of "List"
In datasets/arrow_writer.py, replaced all the type annotations of List method with Sequence.
Library used and changes made are as follows:
```bash
from collections.abc import Sequence as Sequence_
Before : List(Image())
After : Sequence_(Image())
| open | 2026-05-15T10:06:17 | https://api.github.com/repos/huggingface/datasets/issues/8198/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8198",
"html_url": "https://github.com/huggingface/datasets/pull/8198",
"diff_url": "https://github.com/huggingface/datasets/pull/8198.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8198.patch",
"merged_at": null
} | true | Fix/type hints list to sequence
Fix #5354 Consider using "Sequence" instead of "List"
In datasets/arrow_writer.py, replaced all the type annotations of List method with Sequence.
Library used and changes made are as follows:
```bash
from collections.abc import Sequence as Sequence_
Before : List(Image())
... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8197 | 8,197 | Feature Request / Discussion: Skip origin_metadata fetch when streaming=True? | Hi team!
I've been looking into the initialization process of the `DatasetBuilder` and noticed that the ETag/origin metadata caching path is executed unconditionally.
While fetching the `origin_metadata` and calculating the ETag is necessary for safely managing a local disk cache when using `as_dataset()`, it seems t... | open | 2026-05-15T07:43:57 | https://api.github.com/repos/huggingface/datasets/issues/8197/comments | null | false | Feature Request / Discussion: Skip origin_metadata fetch when streaming=True?
Hi team!
I've been looking into the initialization process of the `DatasetBuilder` and noticed that the ETag/origin metadata caching path is executed unconditionally.
While fetching the `origin_metadata` and calculating the ETag is necessar... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8196 | 8,196 | fix(arrow_dataset): clear stale local temp dir before re-downloading from remote FS | ## What does this PR do?
Fixes #5114 — `Dataset.load_from_disk` raises a `FileNotFoundError` when called **twice** on the same remote-filesystem path within the same Python session.
### Root cause
`Dataset.load_from_disk` uses `fs.download(src, dest, recursive=True)` to copy the remote dataset to a local temp direct... | open | 2026-05-14T11:33:11 | https://api.github.com/repos/huggingface/datasets/issues/8196/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8196",
"html_url": "https://github.com/huggingface/datasets/pull/8196",
"diff_url": "https://github.com/huggingface/datasets/pull/8196.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8196.patch",
"merged_at": null
} | true | fix(arrow_dataset): clear stale local temp dir before re-downloading from remote FS
## What does this PR do?
Fixes #5114 — `Dataset.load_from_disk` raises a `FileNotFoundError` when called **twice** on the same remote-filesystem path within the same Python session.
### Root cause
`Dataset.load_from_disk` uses `fs.do... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8195 | 8,195 | Fix spurious label column when folder builders see split-named directories | Closes #7880.
## What
`folder_based_builder._split_generators` allocates a single `labels: set[str]` outside the per-split loop and `analyze(...)` mutates it across every split. When the user passes:
```python
load_dataset("audiofolder", data_files={"train": "train/**", "test": "test/**"})
```
and the actual files ... | open | 2026-05-12T10:45:35 | https://api.github.com/repos/huggingface/datasets/issues/8195/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8195",
"html_url": "https://github.com/huggingface/datasets/pull/8195",
"diff_url": "https://github.com/huggingface/datasets/pull/8195.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8195.patch",
"merged_at": null
} | true | Fix spurious label column when folder builders see split-named directories
Closes #7880.
## What
`folder_based_builder._split_generators` allocates a single `labels: set[str]` outside the per-split loop and `analyze(...)` mutates it across every split. When the user passes:
```python
load_dataset("audiofolder", data... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8194 | 8,194 | Use multiple input shards for shuffle buffer | ```python
ds = load_dataset(..., streaming=True)
ds = ds.shuffle(seed=42)
# or configure local buffer shuffling manually, default is:
ds = ds.shuffle(seed=42, buffer_size=1000, max_buffer_input_shards=10)
```
before👎:
<img width="1061" height="529" alt="image" src="https://github.com/user-attachments/assets/d... | closed | 2026-05-11T15:04:57 | https://api.github.com/repos/huggingface/datasets/issues/8194/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8194",
"html_url": "https://github.com/huggingface/datasets/pull/8194",
"diff_url": "https://github.com/huggingface/datasets/pull/8194.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8194.patch",
"merged_at": "2026-05-18T13:59... | true | Use multiple input shards for shuffle buffer
```python
ds = load_dataset(..., streaming=True)
ds = ds.shuffle(seed=42)
# or configure local buffer shuffling manually, default is:
ds = ds.shuffle(seed=42, buffer_size=1000, max_buffer_input_shards=10)
```
before👎:
<img width="1061" height="529" alt="image" src=... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8193 | 8,193 | fix parquet reshard | .reshard() would not do anything and return the same iterable dataset for parquet, now it returns the resharded iterable dataset - sharded per row group instead of per file | closed | 2026-05-11T15:02:26 | https://api.github.com/repos/huggingface/datasets/issues/8193/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8193",
"html_url": "https://github.com/huggingface/datasets/pull/8193",
"diff_url": "https://github.com/huggingface/datasets/pull/8193.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8193.patch",
"merged_at": "2026-05-12T09:44... | true | fix parquet reshard
.reshard() would not do anything and return the same iterable dataset for parquet, now it returns the resharded iterable dataset - sharded per row group instead of per file | [] |
https://api.github.com/repos/huggingface/datasets/issues/8192 | 8,192 | Escape glob chars in `base_path` so directory paths with `[]` work (#7468) | Fixes #7468.
## Summary
`load_dataset(..., data_dir="/path/[D_DATA]/foo")` — or any path with `[`, `]`, `*`, `?` in the literal directory name — failed: `fs.glob` interpreted the bracketed segment as a glob character class, walked a much larger tree (often the whole disk on Windows) and returned no files. Users saw a... | open | 2026-05-10T12:01:55 | https://api.github.com/repos/huggingface/datasets/issues/8192/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8192",
"html_url": "https://github.com/huggingface/datasets/pull/8192",
"diff_url": "https://github.com/huggingface/datasets/pull/8192.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8192.patch",
"merged_at": null
} | true | Escape glob chars in `base_path` so directory paths with `[]` work (#7468)
Fixes #7468.
## Summary
`load_dataset(..., data_dir="/path/[D_DATA]/foo")` — or any path with `[`, `]`, `*`, `?` in the literal directory name — failed: `fs.glob` interpreted the bracketed segment as a glob character class, walked a much large... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8191 | 8,191 | Preserve `info.features` across `IterableDataset.map(remove_columns=...)` (#7568) | Fixes #7568.
## Summary
`IterableDataset.map(remove_columns=[...])` (without an explicit `features=` and without a user-provided `function`) overwrites `info.features` with `None`, so `IterableDataset.column_names` (which reads `info.features.keys()`) silently becomes `None`. The same regression bubbles up through `I... | open | 2026-05-10T11:58:54 | https://api.github.com/repos/huggingface/datasets/issues/8191/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8191",
"html_url": "https://github.com/huggingface/datasets/pull/8191",
"diff_url": "https://github.com/huggingface/datasets/pull/8191.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8191.patch",
"merged_at": null
} | true | Preserve `info.features` across `IterableDataset.map(remove_columns=...)` (#7568)
Fixes #7568.
## Summary
`IterableDataset.map(remove_columns=[...])` (without an explicit `features=` and without a user-provided `function`) overwrites `info.features` with `None`, so `IterableDataset.column_names` (which reads `info.fe... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8190 | 8,190 | Reject `num_shards > len(dataset)` in `Dataset.shard` (#7443) | Fixes #7443.
## Summary
`Dataset.shard(num_shards=N, index=i)` with `N > len(dataset)`:
- silently produced empty shards (`contiguous=True`)
- crashed with `IndexError: Index ... out of range for dataset of size ...` deep in `.select` (`contiguous=False`)
`push_to_hub` / `save_to_disk` then surfaced this confusing I... | open | 2026-05-10T11:54:07 | https://api.github.com/repos/huggingface/datasets/issues/8190/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8190",
"html_url": "https://github.com/huggingface/datasets/pull/8190",
"diff_url": "https://github.com/huggingface/datasets/pull/8190.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8190.patch",
"merged_at": null
} | true | Reject `num_shards > len(dataset)` in `Dataset.shard` (#7443)
Fixes #7443.
## Summary
`Dataset.shard(num_shards=N, index=i)` with `N > len(dataset)`:
- silently produced empty shards (`contiguous=True`)
- crashed with `IndexError: Index ... out of range for dataset of size ...` deep in `.select` (`contiguous=False`)
... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8189 | 8,189 | docs: clarify `num_proc` semantics in `Dataset.batch` / `Dataset.filter` (#7700) | Fixes #7700.
## Summary
`Dataset.batch.num_proc` and `Dataset.filter.num_proc` had a one-liner docstring (\"Max number of processes when generating cache. Already cached shards are loaded sequentially.\") that left readers wondering whether `None` means "no multiprocessing" or "pick a number".
Both methods simply de... | open | 2026-05-10T09:14:55 | https://api.github.com/repos/huggingface/datasets/issues/8189/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8189",
"html_url": "https://github.com/huggingface/datasets/pull/8189",
"diff_url": "https://github.com/huggingface/datasets/pull/8189.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8189.patch",
"merged_at": null
} | true | docs: clarify `num_proc` semantics in `Dataset.batch` / `Dataset.filter` (#7700)
Fixes #7700.
## Summary
`Dataset.batch.num_proc` and `Dataset.filter.num_proc` had a one-liner docstring (\"Max number of processes when generating cache. Already cached shards are loaded sequentially.\") that left readers wondering whet... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8188 | 8,188 | Don't infer labels from split-named directories in folder-based builders (#7880) | Fixes #7880.
## Summary
When an `audiofolder`/`imagefolder`/`videofolder` dataset is laid out as
```
folder/train/file0.mp3
folder/test/file1.mp3
```
with no class subdirectories, label inference accumulates the parent directory of each file across **all splits**, ends up with `labels = {\"train\", \"test\"}`, sees... | open | 2026-05-10T09:12:22 | https://api.github.com/repos/huggingface/datasets/issues/8188/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8188",
"html_url": "https://github.com/huggingface/datasets/pull/8188",
"diff_url": "https://github.com/huggingface/datasets/pull/8188.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8188.patch",
"merged_at": null
} | true | Don't infer labels from split-named directories in folder-based builders (#7880)
Fixes #7880.
## Summary
When an `audiofolder`/`imagefolder`/`videofolder` dataset is laid out as
```
folder/train/file0.mp3
folder/test/file1.mp3
```
with no class subdirectories, label inference accumulates the parent directory of eac... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8187 | 8,187 | Better error for sliced splits in streaming mode (`train[:10%]`, `train[:N]`) (#7721) | Fixes #7721.
## Summary
```python
load_dataset("user/dataset", split="train[0%:10%]", streaming=True)
```
raised the generic
```
ValueError: Bad split: train[0%:10%]. Available splits: ['train']
```
which makes it look like the slice syntax is malformed. The actual reason is that streaming mode doesn't know the da... | open | 2026-05-10T09:07:20 | https://api.github.com/repos/huggingface/datasets/issues/8187/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8187",
"html_url": "https://github.com/huggingface/datasets/pull/8187",
"diff_url": "https://github.com/huggingface/datasets/pull/8187.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8187.patch",
"merged_at": null
} | true | Better error for sliced splits in streaming mode (`train[:10%]`, `train[:N]`) (#7721)
Fixes #7721.
## Summary
```python
load_dataset("user/dataset", split="train[0%:10%]", streaming=True)
```
raised the generic
```
ValueError: Bad split: train[0%:10%]. Available splits: ['train']
```
which makes it look like the s... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8186 | 8,186 | docs: fix broken WebDataset documentation link in audio/video/image dataset pages (#7699) | Fixes #7699.
## Summary
The link to the WebDataset documentation in the audio, video and image dataset docs pages points to `https://webdataset.github.io/webdataset`, which returns **HTTP 404**.
The WebDataset library now lives at `https://github.com/webdataset/webdataset`, which is the canonical documentation entry... | open | 2026-05-10T09:02:54 | https://api.github.com/repos/huggingface/datasets/issues/8186/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8186",
"html_url": "https://github.com/huggingface/datasets/pull/8186",
"diff_url": "https://github.com/huggingface/datasets/pull/8186.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8186.patch",
"merged_at": null
} | true | docs: fix broken WebDataset documentation link in audio/video/image dataset pages (#7699)
Fixes #7699.
## Summary
The link to the WebDataset documentation in the audio, video and image dataset docs pages points to `https://webdataset.github.io/webdataset`, which returns **HTTP 404**.
The WebDataset library now lives... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8185 | 8,185 | docs: make `Dataset.map` batched example self-contained (#7703) | Fixes #7703.
## Summary
The batched-map example in `Dataset.map`, `DatasetDict.map` and `docs/README.md` referenced `tokenizer` without defining or importing it:
```python
>>> ds = ds.map(lambda example: tokenizer(example["text"]), batched=True)
```
Users copy-pasting the snippet hit `NameError: name 'tokenizer' is... | open | 2026-05-10T09:01:43 | https://api.github.com/repos/huggingface/datasets/issues/8185/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8185",
"html_url": "https://github.com/huggingface/datasets/pull/8185",
"diff_url": "https://github.com/huggingface/datasets/pull/8185.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8185.patch",
"merged_at": null
} | true | docs: make `Dataset.map` batched example self-contained (#7703)
Fixes #7703.
## Summary
The batched-map example in `Dataset.map`, `DatasetDict.map` and `docs/README.md` referenced `tokenizer` without defining or importing it:
```python
>>> ds = ds.map(lambda example: tokenizer(example["text"]), batched=True)
```
Us... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8184 | 8,184 | Preserve triple-slash in remote URLs (HDFS, file://, ...) in `_as_str` (#7934) | Fixes #7934.
## Summary
`_as_str` routed every non-`xPath` argument — including raw strings — through `xPath(str(path))`, which inherits from `pathlib.PurePosixPath`. PurePosixPath collapses runs of slashes:
```python
>>> str(PurePosixPath("hdfs:///user/path/data.parquet"))
'hdfs:/user/path/data.parquet'
```
`xPath... | open | 2026-05-10T08:59:29 | https://api.github.com/repos/huggingface/datasets/issues/8184/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8184",
"html_url": "https://github.com/huggingface/datasets/pull/8184",
"diff_url": "https://github.com/huggingface/datasets/pull/8184.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8184.patch",
"merged_at": null
} | true | Preserve triple-slash in remote URLs (HDFS, file://, ...) in `_as_str` (#7934)
Fixes #7934.
## Summary
`_as_str` routed every non-`xPath` argument — including raw strings — through `xPath(str(path))`, which inherits from `pathlib.PurePosixPath`. PurePosixPath collapses runs of slashes:
```python
>>> str(PurePosixPat... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8183 | 8,183 | Add dtype="list" to List / Sequence / LargeList features (#8002) | Fixes #8002.
## Summary
Most feature types expose a descriptive `dtype` string:
- `Value("int64").dtype == "int64"`
- `ClassLabel(...).dtype == "int64"`
- `Array2D(...).dtype == "<element-dtype>"`
- `Image().dtype == "PIL.Image.Image"`
- `Audio().dtype == "dict"`
But `List`, `LargeList` and `Sequence(non-dict)` (wh... | open | 2026-05-10T08:53:40 | https://api.github.com/repos/huggingface/datasets/issues/8183/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8183",
"html_url": "https://github.com/huggingface/datasets/pull/8183",
"diff_url": "https://github.com/huggingface/datasets/pull/8183.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8183.patch",
"merged_at": null
} | true | Add dtype="list" to List / Sequence / LargeList features (#8002)
Fixes #8002.
## Summary
Most feature types expose a descriptive `dtype` string:
- `Value("int64").dtype == "int64"`
- `ClassLabel(...).dtype == "int64"`
- `Array2D(...).dtype == "<element-dtype>"`
- `Image().dtype == "PIL.Image.Image"`
- `Audio().dtype... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8182 | 8,182 | Fix Dataset.map crash when first examples return None and later return dict (#7990) | Fixes #7990.
## Summary
`Dataset.map` crashed with `AttributeError: 'NoneType' object has no attribute 'write'` when the user's map function returned `None` for the first few examples and a `dict` (or `pa.Table` / DataFrame) for later ones.
The writer was initialized only when `i == 0` (or `i[0] == 0` in batched mod... | open | 2026-05-10T08:46:55 | https://api.github.com/repos/huggingface/datasets/issues/8182/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8182",
"html_url": "https://github.com/huggingface/datasets/pull/8182",
"diff_url": "https://github.com/huggingface/datasets/pull/8182.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8182.patch",
"merged_at": null
} | true | Fix Dataset.map crash when first examples return None and later return dict (#7990)
Fixes #7990.
## Summary
`Dataset.map` crashed with `AttributeError: 'NoneType' object has no attribute 'write'` when the user's map function returned `None` for the first few examples and a `dict` (or `pa.Table` / DataFrame) for later... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8178 | 8,178 | The full dataset viewer is not available (click to read why). Only showing a preview of the rows. Job manager crashed while running this job (missing heartbeats). Error code: JobManagerCrashedError | ### Describe the bug
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Job manager crashed while running this job (missing heartbeats).
Error code: JobManagerCrashedError
see: https://huggingface.co/datasets/Genius-Society/tt100k/viewer
### Steps to reproduce the bug... | open | 2026-05-07T08:33:48 | https://api.github.com/repos/huggingface/datasets/issues/8178/comments | null | false | The full dataset viewer is not available (click to read why). Only showing a preview of the rows. Job manager crashed while running this job (missing heartbeats). Error code: JobManagerCrashedError
### Describe the bug
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8177 | 8,177 | Nightly OSS: Bug loading `sentence-transformers/parallel-sentences` | Automated nightly Codex contribution for https://github.com/huggingface/datasets/issues/5081.
Verification:
- See the uploaded nightly artifacts for the Codex final report.
- See this workflow run for exact commands and logs.
This PR is generated by xodn348/oss-nightly-control and is opened ready for maintainer revie... | open | 2026-05-07T08:04:50 | https://api.github.com/repos/huggingface/datasets/issues/8177/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8177",
"html_url": "https://github.com/huggingface/datasets/pull/8177",
"diff_url": "https://github.com/huggingface/datasets/pull/8177.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8177.patch",
"merged_at": null
} | true | Nightly OSS: Bug loading `sentence-transformers/parallel-sentences`
Automated nightly Codex contribution for https://github.com/huggingface/datasets/issues/5081.
Verification:
- See the uploaded nightly artifacts for the Codex final report.
- See this workflow run for exact commands and logs.
This PR is generated by ... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8176 | 8,176 | Fix Parquet streaming hangs at the end of script | the fix is applied for pyarrow<=24 for now since it involves a gc.collect() call that won't be needed as a fix is being added in arrow
fix https://github.com/huggingface/datasets/issues/8169
fix https://github.com/huggingface/datasets/issues/7467 | closed | 2026-05-06T14:38:47 | https://api.github.com/repos/huggingface/datasets/issues/8176/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8176",
"html_url": "https://github.com/huggingface/datasets/pull/8176",
"diff_url": "https://github.com/huggingface/datasets/pull/8176.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8176.patch",
"merged_at": "2026-05-06T16:06... | true | Fix Parquet streaming hangs at the end of script
the fix is applied for pyarrow<=24 for now since it involves a gc.collect() call that won't be needed as a fix is being added in arrow
fix https://github.com/huggingface/datasets/issues/8169
fix https://github.com/huggingface/datasets/issues/7467 | [] |
https://api.github.com/repos/huggingface/datasets/issues/8175 | 8,175 | Support fsspec 2026.4.0 | add a fix for https://github.com/fsspec/filesystem_spec/pull/2012 | closed | 2026-05-06T12:23:52 | https://api.github.com/repos/huggingface/datasets/issues/8175/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8175",
"html_url": "https://github.com/huggingface/datasets/pull/8175",
"diff_url": "https://github.com/huggingface/datasets/pull/8175.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8175.patch",
"merged_at": "2026-05-06T12:39... | true | Support fsspec 2026.4.0
add a fix for https://github.com/fsspec/filesystem_spec/pull/2012 | [] |
https://api.github.com/repos/huggingface/datasets/issues/8174 | 8,174 | feat: add `on_error` to Image/Audio/Video features and IterableDatase… | …t.decode()
Adds a new `on_error` parameter that controls how decoding failures are handled. This addresses the long-standing problem where a single corrupted sample in a streaming dataset (bad EXIF, truncated bytes, unidentified format, ...) raises an exception inside the generator frame and silently kills the enti... | open | 2026-05-06T07:29:30 | https://api.github.com/repos/huggingface/datasets/issues/8174/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8174",
"html_url": "https://github.com/huggingface/datasets/pull/8174",
"diff_url": "https://github.com/huggingface/datasets/pull/8174.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8174.patch",
"merged_at": null
} | true | feat: add `on_error` to Image/Audio/Video features and IterableDatase…
…t.decode()
Adds a new `on_error` parameter that controls how decoding failures are handled. This addresses the long-standing problem where a single corrupted sample in a streaming dataset (bad EXIF, truncated bytes, unidentified format, ...) rai... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8173 | 8,173 | fix(image): guard exif_transpose against corrupted EXIF rational tags | ## Summary
When an `IterableDataset` contains images with corrupted EXIF metadata (e.g. a TIFF rational tag whose denominator is zero), `PIL.ImageOps.exif_transpose` raises `ZeroDivisionError` inside `Image.decode_example`. Because this exception escapes the internal generator frame, the streaming iterator dies immedi... | open | 2026-05-05T05:16:04 | https://api.github.com/repos/huggingface/datasets/issues/8173/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8173",
"html_url": "https://github.com/huggingface/datasets/pull/8173",
"diff_url": "https://github.com/huggingface/datasets/pull/8173.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8173.patch",
"merged_at": null
} | true | fix(image): guard exif_transpose against corrupted EXIF rational tags
## Summary
When an `IterableDataset` contains images with corrupted EXIF metadata (e.g. a TIFF rational tag whose denominator is zero), `PIL.ImageOps.exif_transpose` raises `ZeroDivisionError` inside `Image.decode_example`. Because this exception es... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8172 | 8,172 | Add batch(by_column=...) | Will be useful for robotics dataset to batch samples by episode cc @pkooij
example of usage:
```python
from datasets import Dataset
ds = Dataset.from_dict({"episode": [0] * 10 + [1] * 10, "frame": list(range(10)) * 2})
# ds = ds.to_iterable_dataset()
ds = ds.batch(by_column="episode")
for x in ds:
pr... | closed | 2026-05-04T18:49:33 | https://api.github.com/repos/huggingface/datasets/issues/8172/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8172",
"html_url": "https://github.com/huggingface/datasets/pull/8172",
"diff_url": "https://github.com/huggingface/datasets/pull/8172.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8172.patch",
"merged_at": "2026-05-09T13:16... | true | Add batch(by_column=...)
Will be useful for robotics dataset to batch samples by episode cc @pkooij
example of usage:
```python
from datasets import Dataset
ds = Dataset.from_dict({"episode": [0] * 10 + [1] * 10, "frame": list(range(10)) * 2})
# ds = ds.to_iterable_dataset()
ds = ds.batch(by_column="episod... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8171 | 8,171 | Dataset UI shows 0:00/0:00 for audio after push_to_hub | I’m seeing an issue where audio previews in the dataset viewer show `0:00/0:00` after uploading with push_to_hub.
## What I did
I tried to avoid `torchcodec` because I want to keep the packages small using the following code:
```python
import io
import os
import shutil
import tempfile
from pathlib import Path
imp... | closed | 2026-05-04T13:17:25 | https://api.github.com/repos/huggingface/datasets/issues/8171/comments | null | false | Dataset UI shows 0:00/0:00 for audio after push_to_hub
I’m seeing an issue where audio previews in the dataset viewer show `0:00/0:00` after uploading with push_to_hub.
## What I did
I tried to avoid `torchcodec` because I want to keep the packages small using the following code:
```python
import io
import os
impo... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8170 | 8,170 | fix(map): fix progress bar exceeding total when load_from_cache_file=False | ## Summary
When `load_from_cache_file=False`, the progress bar in `Dataset.map()` displayed a count that exceeded the dataset size (e.g. "800 examples" for a 400-row dataset).
The bug: `pbar_initial` was computed using `len(existing_cache_files)`, which counts cache files on disk regardless of whether they would... | closed | 2026-05-03T14:49:46 | https://api.github.com/repos/huggingface/datasets/issues/8170/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8170",
"html_url": "https://github.com/huggingface/datasets/pull/8170",
"diff_url": "https://github.com/huggingface/datasets/pull/8170.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8170.patch",
"merged_at": "2026-05-29T10:59... | true | fix(map): fix progress bar exceeding total when load_from_cache_file=False
## Summary
When `load_from_cache_file=False`, the progress bar in `Dataset.map()` displayed a count that exceeded the dataset size (e.g. "800 examples" for a 400-row dataset).
The bug: `pbar_initial` was computed using `len(existing_cache... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8169 | 8,169 | Streaming dataset hangs consistently | ### Describe the bug
When loading a dataset with `streaming=True`, some background process prevents the script from ever returning (or at least for a very long time).
This did not happen with `huggingface-hub < 1`.
### Steps to reproduce the bug
from datasets import load_dataset
ds = load_dataset("IRIIS-RESEARCH/N... | closed | 2026-05-01T20:27:58 | https://api.github.com/repos/huggingface/datasets/issues/8169/comments | null | false | Streaming dataset hangs consistently
### Describe the bug
When loading a dataset with `streaming=True`, some background process prevents the script from ever returning (or at least for a very long time).
This did not happen with `huggingface-hub < 1`.
### Steps to reproduce the bug
from datasets import load_dataset... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8168 | 8,168 | docs(map): clarify that fn_kwargs yields a stable cache fingerprint vs closures | ## Summary
When `Dataset.map()` or `Dataset.filter()` receives a closure that closes over an object containing non-deterministic state — for example a UUID, a random seed, or any mutable attribute of a class instance — `dill` serialises the **entire** captured object into the fingerprint. Two calls that are logically... | open | 2026-05-01T09:33:30 | https://api.github.com/repos/huggingface/datasets/issues/8168/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8168",
"html_url": "https://github.com/huggingface/datasets/pull/8168",
"diff_url": "https://github.com/huggingface/datasets/pull/8168.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8168.patch",
"merged_at": null
} | true | docs(map): clarify that fn_kwargs yields a stable cache fingerprint vs closures
## Summary
When `Dataset.map()` or `Dataset.filter()` receives a closure that closes over an object containing non-deterministic state — for example a UUID, a random seed, or any mutable attribute of a class instance — `dill` serialises th... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8167 | 8,167 | [BUG] Abnormal progress bar in dataset.map when load_from_cache_file=False | ### Describe the bug
datasets 4.8.5
```python
from datasets import load_dataset, load_from_disk
data = load_from_disk('pretrain_cached_dataset/train')
def dummy_map(example):
return example
dataset = data.map(dummy_map)
dataset = data.map(dummy_map, load_from_cache_file=True) ... | closed | 2026-04-30T08:47:53 | https://api.github.com/repos/huggingface/datasets/issues/8167/comments | null | false | [BUG] Abnormal progress bar in dataset.map when load_from_cache_file=False
### Describe the bug
datasets 4.8.5
```python
from datasets import load_dataset, load_from_disk
data = load_from_disk('pretrain_cached_dataset/train')
def dummy_map(example):
return example
dataset = data... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8166 | 8,166 | Fix storage_options lookup for streaming Lance datasets | In `Lance._split_generators`, the `storage_options` lookup used `files[0].split("://", 0)[0] + "://"`, which has two bugs that compound: `split("://", 0)` does not split at all (maxsplit=0), and the `storage_options` dict is keyed by the bare scheme (`"hf"`), not `"hf://"`. The lookup therefore always returned `None`, ... | closed | 2026-04-30T02:34:57 | https://api.github.com/repos/huggingface/datasets/issues/8166/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8166",
"html_url": "https://github.com/huggingface/datasets/pull/8166",
"diff_url": "https://github.com/huggingface/datasets/pull/8166.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8166.patch",
"merged_at": "2026-04-30T15:38... | true | Fix storage_options lookup for streaming Lance datasets
In `Lance._split_generators`, the `storage_options` lookup used `files[0].split("://", 0)[0] + "://"`, which has two bugs that compound: `split("://", 0)` does not split at all (maxsplit=0), and the `storage_options` dict is keyed by the bare scheme (`"hf"`), not ... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8165 | 8,165 | IterableDataset: corrupted EXIF image silently terminates streaming iterator instead of skipping the sample | ## Summary
When iterating over a streaming `IterableDataset` that contains images with corrupted EXIF metadata (e.g., a TIFF rational tag with denominator=0), PIL raises `ZeroDivisionError` inside `exif_transpose`. This exception propagates through the HuggingFace `datasets` pipeline and **terminates the streaming ite... | open | 2026-04-30T02:20:54 | https://api.github.com/repos/huggingface/datasets/issues/8165/comments | null | false | IterableDataset: corrupted EXIF image silently terminates streaming iterator instead of skipping the sample
## Summary
When iterating over a streaming `IterableDataset` that contains images with corrupted EXIF metadata (e.g., a TIFF rational tag with denominator=0), PIL raises `ZeroDivisionError` inside `exif_transpos... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8164 | 8,164 | Lance: HF auth token dropped for private repos in streaming mode (broken `storage_options` lookup) | ### Describe the bug
When `Lance._split_generators` receives `hf://` URIs - i.e. anywhere a `StreamingDownloadManager` is in play (`load_dataset(..., streaming=True)`, `get_dataset_split_names`, and the dataset-viewer worker) - the HF auth token is silently dropped before reaching `lance.dataset(...)`. Lance then atte... | closed | 2026-04-30T02:03:33 | https://api.github.com/repos/huggingface/datasets/issues/8164/comments | null | false | Lance: HF auth token dropped for private repos in streaming mode (broken `storage_options` lookup)
### Describe the bug
When `Lance._split_generators` receives `hf://` URIs - i.e. anywhere a `StreamingDownloadManager` is in play (`load_dataset(..., streaming=True)`, `get_dataset_split_names`, and the dataset-viewer wo... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8163 | 8,163 | add agent trace prompt, sent_at, count fields | - add agent trace metadata columns: `prompt`, `sent_at`, `num_user_messages`, and `num_tool_calls`.
- extract `prompt` / `sent_at` from the first real user message for Claude Code, Pi/OpenClaw, and Codex traces.
| closed | 2026-04-29T23:13:58 | https://api.github.com/repos/huggingface/datasets/issues/8163/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8163",
"html_url": "https://github.com/huggingface/datasets/pull/8163",
"diff_url": "https://github.com/huggingface/datasets/pull/8163.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8163.patch",
"merged_at": "2026-05-04T12:03... | true | add agent trace prompt, sent_at, count fields
- add agent trace metadata columns: `prompt`, `sent_at`, `num_user_messages`, and `num_tool_calls`.
- extract `prompt` / `sent_at` from the first real user message for Claude Code, Pi/OpenClaw, and Codex traces.
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8162 | 8,162 | Include all test files in sdist via MANIFEST.in | Fixes #8159.
## What's wrong
The sdist on PyPI for `datasets==4.8.5` only ships **37 of the 95** files under `tests/`. Without an explicit `MANIFEST.in`, setuptools' default sdist behavior includes only top-level `tests/test_*.py` files and drops everything else.
Files missing from the current sdist:
- `tests/conft... | open | 2026-04-29T19:29:32 | https://api.github.com/repos/huggingface/datasets/issues/8162/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8162",
"html_url": "https://github.com/huggingface/datasets/pull/8162",
"diff_url": "https://github.com/huggingface/datasets/pull/8162.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8162.patch",
"merged_at": null
} | true | Include all test files in sdist via MANIFEST.in
Fixes #8159.
## What's wrong
The sdist on PyPI for `datasets==4.8.5` only ships **37 of the 95** files under `tests/`. Without an explicit `MANIFEST.in`, setuptools' default sdist behavior includes only top-level `tests/test_*.py` files and drops everything else.
Files... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8161 | 8,161 | Pass library_name/version to HfApi in dataset push and delete paths | Adds `library_name="datasets"` and `library_version=__version__` to 12 bare `HfApi(endpoint=..., token=...)` instantiations across `hub.py`, `arrow_dataset.py`, `iterable_dataset.py`, and `dataset_dict.py`. Commits made via `push_to_hub` / `delete_from_hub` currently report a User-Agent of `unknown/None; hf_hub/X.Y.Z; ... | closed | 2026-04-29T14:08:33 | https://api.github.com/repos/huggingface/datasets/issues/8161/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8161",
"html_url": "https://github.com/huggingface/datasets/pull/8161",
"diff_url": "https://github.com/huggingface/datasets/pull/8161.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8161.patch",
"merged_at": "2026-04-30T10:29... | true | Pass library_name/version to HfApi in dataset push and delete paths
Adds `library_name="datasets"` and `library_version=__version__` to 12 bare `HfApi(endpoint=..., token=...)` instantiations across `hub.py`, `arrow_dataset.py`, `iterable_dataset.py`, and `dataset_dict.py`. Commits made via `push_to_hub` / `delete_from... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8160 | 8,160 | feat: add TsFile (Apache IoTDB) packaged builder with per-device wide format | Closes #7922
## Summary
Add a packaged builder for [TsFile](https://tsfile.apache.org/) — the columnar time-series file format used as the native storage layer of [Apache IoTDB](https://iotdb.apache.org/). This enables `load_dataset("tsfile", data_files="...")` with automatic `.tsfile` extension detection.
### Data ... | open | 2026-04-29T09:07:25 | https://api.github.com/repos/huggingface/datasets/issues/8160/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8160",
"html_url": "https://github.com/huggingface/datasets/pull/8160",
"diff_url": "https://github.com/huggingface/datasets/pull/8160.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8160.patch",
"merged_at": null
} | true | feat: add TsFile (Apache IoTDB) packaged builder with per-device wide format
Closes #7922
## Summary
Add a packaged builder for [TsFile](https://tsfile.apache.org/) — the columnar time-series file format used as the native storage layer of [Apache IoTDB](https://iotdb.apache.org/). This enables `load_dataset("tsfile"... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8159 | 8,159 | tests/conftest.py, tests/_test_patching.py, tests/fixtures/fsspec.py and other files are missing from the PYPI distribution | ### Describe the bug
Version: 4.8.5
### Steps to reproduce the bug
n/a
### Expected behavior
n/a
### Environment info
n/a | open | 2026-04-28T20:01:05 | https://api.github.com/repos/huggingface/datasets/issues/8159/comments | null | false | tests/conftest.py, tests/_test_patching.py, tests/fixtures/fsspec.py and other files are missing from the PYPI distribution
### Describe the bug
Version: 4.8.5
### Steps to reproduce the bug
n/a
### Expected behavior
n/a
### Environment info
n/a | [] |
https://api.github.com/repos/huggingface/datasets/issues/8158 | 8,158 | Dev version | null | closed | 2026-04-27T15:47:37 | https://api.github.com/repos/huggingface/datasets/issues/8158/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8158",
"html_url": "https://github.com/huggingface/datasets/pull/8158",
"diff_url": "https://github.com/huggingface/datasets/pull/8158.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8158.patch",
"merged_at": "2026-04-27T15:47... | true | Dev version
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8157 | 8,157 | Release: 4.8.5 | null | closed | 2026-04-27T15:43:15 | https://api.github.com/repos/huggingface/datasets/issues/8157/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8157",
"html_url": "https://github.com/huggingface/datasets/pull/8157",
"diff_url": "https://github.com/huggingface/datasets/pull/8157.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8157.patch",
"merged_at": "2026-04-27T15:44... | true | Release: 4.8.5
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8156 | 8,156 | Fix revision-aware cache lookup in load_dataset | Fixes #7928
When `load_dataset()` falls back to the local cache, for example when offline or when the Hub is unreachable, it ignores the `revision` argument and simply returns the most recently modified cached version. This can lead to unexpected behavior when a specific revision is requested.
This happens becaus... | open | 2026-04-26T05:21:17 | https://api.github.com/repos/huggingface/datasets/issues/8156/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8156",
"html_url": "https://github.com/huggingface/datasets/pull/8156",
"diff_url": "https://github.com/huggingface/datasets/pull/8156.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8156.patch",
"merged_at": null
} | true | Fix revision-aware cache lookup in load_dataset
Fixes #7928
When `load_dataset()` falls back to the local cache, for example when offline or when the Hub is unreachable, it ignores the `revision` argument and simply returns the most recently modified cached version. This can lead to unexpected behavior when a specif... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8155 | 8,155 | Fix base_path in integration tests | null | closed | 2026-04-24T17:18:11 | https://api.github.com/repos/huggingface/datasets/issues/8155/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8155",
"html_url": "https://github.com/huggingface/datasets/pull/8155",
"diff_url": "https://github.com/huggingface/datasets/pull/8155.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8155.patch",
"merged_at": "2026-04-24T17:20... | true | Fix base_path in integration tests
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8154 | 8,154 | Fix null in embed storage | null | closed | 2026-04-24T16:28:18 | https://api.github.com/repos/huggingface/datasets/issues/8154/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8154",
"html_url": "https://github.com/huggingface/datasets/pull/8154",
"diff_url": "https://github.com/huggingface/datasets/pull/8154.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8154.patch",
"merged_at": "2026-04-24T17:13... | true | Fix null in embed storage
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8153 | 8,153 | Fix tilde expansion for local data_files paths | Summary
- expand `~` before classifying local `data_files` patterns
- add a regression test covering `resolve_pattern(os.path.join("~", ...), ...)`
Impact
- `load_dataset(..., data_files="~/path/to/file.parquet")` currently treats `~` as a literal relative directory under the working tree instead of the user's home di... | open | 2026-04-24T15:23:52 | https://api.github.com/repos/huggingface/datasets/issues/8153/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8153",
"html_url": "https://github.com/huggingface/datasets/pull/8153",
"diff_url": "https://github.com/huggingface/datasets/pull/8153.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8153.patch",
"merged_at": null
} | true | Fix tilde expansion for local data_files paths
Summary
- expand `~` before classifying local `data_files` patterns
- add a regression test covering `resolve_pattern(os.path.join("~", ...), ...)`
Impact
- `load_dataset(..., data_files="~/path/to/file.parquet")` currently treats `~` as a literal relative directory under... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8152 | 8,152 | fix original_files missing in ci json test | null | closed | 2026-04-23T17:29:29 | https://api.github.com/repos/huggingface/datasets/issues/8152/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8152",
"html_url": "https://github.com/huggingface/datasets/pull/8152",
"diff_url": "https://github.com/huggingface/datasets/pull/8152.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8152.patch",
"merged_at": "2026-04-23T17:29... | true | fix original_files missing in ci json test
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8151 | 8,151 | fix mask in embed_storage for remote files | following https://github.com/huggingface/datasets/pull/8150 | closed | 2026-04-23T17:27:22 | https://api.github.com/repos/huggingface/datasets/issues/8151/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8151",
"html_url": "https://github.com/huggingface/datasets/pull/8151",
"diff_url": "https://github.com/huggingface/datasets/pull/8151.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8151.patch",
"merged_at": "2026-04-23T17:27... | true | fix mask in embed_storage for remote files
following https://github.com/huggingface/datasets/pull/8150 | [] |
https://api.github.com/repos/huggingface/datasets/issues/8150 | 8,150 | don't embed remote files in download_and_prepare to parquet | fix the viewer at https://huggingface.co/datasets/raivn/vn-test/discussions/1
kinda related: https://github.com/huggingface/datasets/issues/5717 | closed | 2026-04-23T16:45:46 | https://api.github.com/repos/huggingface/datasets/issues/8150/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8150",
"html_url": "https://github.com/huggingface/datasets/pull/8150",
"diff_url": "https://github.com/huggingface/datasets/pull/8150.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8150.patch",
"merged_at": "2026-04-23T16:48... | true | don't embed remote files in download_and_prepare to parquet
fix the viewer at https://huggingface.co/datasets/raivn/vn-test/discussions/1
kinda related: https://github.com/huggingface/datasets/issues/5717 | [] |
https://api.github.com/repos/huggingface/datasets/issues/8149 | 8,149 | [Optimization] Prevent per-thread instantiation of Cloud Storage FileSystem during Data loading initialization | ### Feature request
Modify the dataset loading initialization, so that fsspec filesystem instances (like GCSFileSystem or S3FileSystem) are instantiated once in the main thread and explicitly passed down to the background threads.
By pre-instantiating the fs object in the main thread (where the directory cache from g... | open | 2026-04-23T13:09:19 | https://api.github.com/repos/huggingface/datasets/issues/8149/comments | null | false | [Optimization] Prevent per-thread instantiation of Cloud Storage FileSystem during Data loading initialization
### Feature request
Modify the dataset loading initialization, so that fsspec filesystem instances (like GCSFileSystem or S3FileSystem) are instantiated once in the main thread and explicitly passed down to t... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8148 | 8,148 | Add Apache Iceberg format support | # Add Apache Iceberg format support
## Motivation
Apache Iceberg is the most widely adopted open table format for data lakes, supported by Databricks,
Snowflake, AWS Glue, Dremio, and others. A large amount of ML training data lives in Iceberg tables.
Currently, users must manually export Iceberg data t... | closed | 2026-04-23T11:29:11 | https://api.github.com/repos/huggingface/datasets/issues/8148/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8148",
"html_url": "https://github.com/huggingface/datasets/pull/8148",
"diff_url": "https://github.com/huggingface/datasets/pull/8148.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8148.patch",
"merged_at": "2026-05-27T14:41... | true | Add Apache Iceberg format support
# Add Apache Iceberg format support
## Motivation
Apache Iceberg is the most widely adopted open table format for data lakes, supported by Databricks,
Snowflake, AWS Glue, Dremio, and others. A large amount of ML training data lives in Iceberg tables.
Currently, users m... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8147 | 8,147 | Fix iterable map resume state |
Fixes: https://github.com/huggingface/datasets/issues/8145
## Root Cause
`FormattedExamplesIterable.__iter__` ignored `force_convert_to_python` and used Arrow batch iteration, advancing the
underlying state by a full Arrow chunk before yielding Python examples.
## Fix
Only use Arrow batch iteration whe... | closed | 2026-04-23T00:03:14 | https://api.github.com/repos/huggingface/datasets/issues/8147/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8147",
"html_url": "https://github.com/huggingface/datasets/pull/8147",
"diff_url": "https://github.com/huggingface/datasets/pull/8147.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8147.patch",
"merged_at": "2026-04-27T14:03... | true | Fix iterable map resume state
Fixes: https://github.com/huggingface/datasets/issues/8145
## Root Cause
`FormattedExamplesIterable.__iter__` ignored `force_convert_to_python` and used Arrow batch iteration, advancing the
underlying state by a full Arrow chunk before yielding Python examples.
## Fix
Only... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8146 | 8,146 | fix: tolerate chmod PermissionError on flat-permission filesystems | ## Summary
Fixes #8125.
`Dataset.map()` (and the indices writer in `Dataset.flatten_indices` / `Dataset.select` path) calls `os.chmod(cache_file, 0o666 & ~umask)` on the freshly moved cache file. On flat-permission filesystems — GCS FUSE, S3 FUSE mounts, and similar object-storage-backed mounts — `chmod` raises `Perm... | open | 2026-04-22T13:39:46 | https://api.github.com/repos/huggingface/datasets/issues/8146/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8146",
"html_url": "https://github.com/huggingface/datasets/pull/8146",
"diff_url": "https://github.com/huggingface/datasets/pull/8146.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8146.patch",
"merged_at": null
} | true | fix: tolerate chmod PermissionError on flat-permission filesystems
## Summary
Fixes #8125.
`Dataset.map()` (and the indices writer in `Dataset.flatten_indices` / `Dataset.select` path) calls `os.chmod(cache_file, 0o666 & ~umask)` on the freshly moved cache file. On flat-permission filesystems — GCS FUSE, S3 FUSE moun... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8145 | 8,145 | `.map()` on a streaming IterableDataset silently skips examples after load_state_dict | ### Describe the bug
After `ds.map(fn)` on a streaming IterableDataset, saving state_dict() and loading it into a fresh dataset does not resume at the correct position. It jumps forward by up to one Arrow read-chunk (~1000 rows).
### Steps to reproduce the bug
```python
from datasets import Dataset
ds1 = Dataset.fr... | closed | 2026-04-22T03:08:39 | https://api.github.com/repos/huggingface/datasets/issues/8145/comments | null | false | `.map()` on a streaming IterableDataset silently skips examples after load_state_dict
### Describe the bug
After `ds.map(fn)` on a streaming IterableDataset, saving state_dict() and loading it into a fresh dataset does not resume at the correct position. It jumps forward by up to one Arrow read-chunk (~1000 rows).
##... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8144 | 8,144 | Add Apache TsFile packaged module and Time Series docs category | ## What does this PR do?
Adds first-class support for the [Apache TsFile](https://tsfile.apache.org/)
columnar time-series format (used by [Apache IoTDB](https://iotdb.apache.org/)
and other time-series systems) to 🤗 Datasets, and surfaces time-series
datasets as a dedicated modality in the documentation sidebar... | closed | 2026-04-21T07:15:42 | https://api.github.com/repos/huggingface/datasets/issues/8144/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8144",
"html_url": "https://github.com/huggingface/datasets/pull/8144",
"diff_url": "https://github.com/huggingface/datasets/pull/8144.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8144.patch",
"merged_at": null
} | true | Add Apache TsFile packaged module and Time Series docs category
## What does this PR do?
Adds first-class support for the [Apache TsFile](https://tsfile.apache.org/)
columnar time-series format (used by [Apache IoTDB](https://iotdb.apache.org/)
and other time-series systems) to 🤗 Datasets, and surfaces time-serie... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8143 | 8,143 | tag openclaw sessions via cwd in pi agent traces | Tag OpenClaw datasets from Pi agent sessions. Will add a PR to add icon for it in moon too
cc: @merveenoyan | closed | 2026-04-20T14:45:50 | https://api.github.com/repos/huggingface/datasets/issues/8143/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8143",
"html_url": "https://github.com/huggingface/datasets/pull/8143",
"diff_url": "https://github.com/huggingface/datasets/pull/8143.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8143.patch",
"merged_at": "2026-04-21T09:02... | true | tag openclaw sessions via cwd in pi agent traces
Tag OpenClaw datasets from Pi agent sessions. Will add a PR to add icon for it in moon too
cc: @merveenoyan | [] |
https://api.github.com/repos/huggingface/datasets/issues/8142 | 8,142 | fix: replace list with Sequence in public API type hints | Fixes #5354
Replaced `list` with `Sequence` in public API function signatures across:
- `arrow_dataset.py` — from_csv, from_json, from_parquet, from_text
- `iterable_dataset.py` — from_csv, from_json, from_parquet, from_text
- `combine.py` — interleave_datasets, concatenate_datasets
- `dataset_dict.py` — remov... | closed | 2026-04-19T23:22:30 | https://api.github.com/repos/huggingface/datasets/issues/8142/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8142",
"html_url": "https://github.com/huggingface/datasets/pull/8142",
"diff_url": "https://github.com/huggingface/datasets/pull/8142.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8142.patch",
"merged_at": null
} | true | fix: replace list with Sequence in public API type hints
Fixes #5354
Replaced `list` with `Sequence` in public API function signatures across:
- `arrow_dataset.py` — from_csv, from_json, from_parquet, from_text
- `iterable_dataset.py` — from_csv, from_json, from_parquet, from_text
- `combine.py` — interleave_d... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8141 | 8,141 | Fix Image feature to dynamically select pyarrow storage type for large binary/string inputs | Add dynamic pyarrow storage type selection to the Image feature so that pa.large_binary() and pa.large_string() inputs are preserved natively instead of being cast down to pa.binary() / pa.string()
Introduce _get_target_pa_type() to detect large types from incoming storage and _to_struct_storage() to build struct arra... | open | 2026-04-17T07:28:01 | https://api.github.com/repos/huggingface/datasets/issues/8141/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8141",
"html_url": "https://github.com/huggingface/datasets/pull/8141",
"diff_url": "https://github.com/huggingface/datasets/pull/8141.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8141.patch",
"merged_at": null
} | true | Fix Image feature to dynamically select pyarrow storage type for large binary/string inputs
Add dynamic pyarrow storage type selection to the Image feature so that pa.large_binary() and pa.large_string() inputs are preserved natively instead of being cast down to pa.binary() / pa.string()
Introduce _get_target_pa_type... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8140 | 8,140 | fix warning in ci | otherwise it raises an issue from transformers smh
```
../../.pyenv/versions/3.12.2/envs/hf-datasets/lib/python3.12/site-packages/transformers/__init__.py:843: in _getattr
return getattr(importlib.import_module(target, __name__), new_name)
../../.pyenv/versions/3.12.2/lib/python3.12/importlib/__init__.py:90: ... | closed | 2026-04-16T13:35:06 | https://api.github.com/repos/huggingface/datasets/issues/8140/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8140",
"html_url": "https://github.com/huggingface/datasets/pull/8140",
"diff_url": "https://github.com/huggingface/datasets/pull/8140.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8140.patch",
"merged_at": "2026-04-16T13:35... | true | fix warning in ci
otherwise it raises an issue from transformers smh
```
../../.pyenv/versions/3.12.2/envs/hf-datasets/lib/python3.12/site-packages/transformers/__init__.py:843: in _getattr
return getattr(importlib.import_module(target, __name__), new_name)
../../.pyenv/versions/3.12.2/lib/python3.12/importli... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8139 | 8,139 | update ci uer | null | closed | 2026-04-16T13:16:21 | https://api.github.com/repos/huggingface/datasets/issues/8139/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8139",
"html_url": "https://github.com/huggingface/datasets/pull/8139",
"diff_url": "https://github.com/huggingface/datasets/pull/8139.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8139.patch",
"merged_at": "2026-04-16T13:17... | true | update ci uer
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8138 | 8,138 | chore: bump doc-builder SHA for main doc build workflow | Bump the pinned doc-builder SHA so that main documentation builds also sync to the HF bucket (dual-write).
| closed | 2026-04-16T12:03:28 | https://api.github.com/repos/huggingface/datasets/issues/8138/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8138",
"html_url": "https://github.com/huggingface/datasets/pull/8138",
"diff_url": "https://github.com/huggingface/datasets/pull/8138.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8138.patch",
"merged_at": null
} | true | chore: bump doc-builder SHA for main doc build workflow
Bump the pinned doc-builder SHA so that main documentation builds also sync to the HF bucket (dual-write).
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8137 | 8,137 | Fix: decode JSON type before to_list or to_dict is called | **Motivation**:
There is a change in the approach to decoding JSON types, added in https://github.com/huggingface/datasets/commit/d560b58e1f1ec96591a95a1780ee118f5eea6c74. Since 4.7.0, to_list() returns raw JSON strings for columns stored as Json(), while direct access via getitem returns dicts.
In versions prior t... | closed | 2026-04-16T11:54:09 | https://api.github.com/repos/huggingface/datasets/issues/8137/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8137",
"html_url": "https://github.com/huggingface/datasets/pull/8137",
"diff_url": "https://github.com/huggingface/datasets/pull/8137.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8137.patch",
"merged_at": "2026-04-27T15:32... | true | Fix: decode JSON type before to_list or to_dict is called
**Motivation**:
There is a change in the approach to decoding JSON types, added in https://github.com/huggingface/datasets/commit/d560b58e1f1ec96591a95a1780ee118f5eea6c74. Since 4.7.0, to_list() returns raw JSON strings for columns stored as Json(), while direc... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8136 | 8,136 | Remove print statement in JSON processing | Remove debug print statement for JSON field paths. | closed | 2026-04-15T23:42:21 | https://api.github.com/repos/huggingface/datasets/issues/8136/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8136",
"html_url": "https://github.com/huggingface/datasets/pull/8136",
"diff_url": "https://github.com/huggingface/datasets/pull/8136.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8136.patch",
"merged_at": "2026-04-15T23:43... | true | Remove print statement in JSON processing
Remove debug print statement for JSON field paths. | [] |
https://api.github.com/repos/huggingface/datasets/issues/8135 | 8,135 | Add Zarr / OME-Zarr Dataset Support | # Add Zarr / OME-Zarr Dataset Support
## TL;DR
This PR adds support for Zarr-backed datasets to `datasets`, with a focus on lazy, streaming-friendly access to large multidimensional arrays. It introduces a new `Zarr` feature, lazy proxy objects for arrays, groups, and OME-Zarr multiscale images, a `zarrfolder` pa... | open | 2026-04-15T10:57:18 | https://api.github.com/repos/huggingface/datasets/issues/8135/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8135",
"html_url": "https://github.com/huggingface/datasets/pull/8135",
"diff_url": "https://github.com/huggingface/datasets/pull/8135.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8135.patch",
"merged_at": null
} | true | Add Zarr / OME-Zarr Dataset Support
# Add Zarr / OME-Zarr Dataset Support
## TL;DR
This PR adds support for Zarr-backed datasets to `datasets`, with a focus on lazy, streaming-friendly access to large multidimensional arrays. It introduces a new `Zarr` feature, lazy proxy objects for arrays, groups, and OME-Zarr ... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8134 | 8,134 | chore: bump doc-builder SHA for PR upload workflow | Switch the PR doc upload flow from the legacy dataset push to the new HF bucket.
| closed | 2026-04-15T08:19:11 | https://api.github.com/repos/huggingface/datasets/issues/8134/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8134",
"html_url": "https://github.com/huggingface/datasets/pull/8134",
"diff_url": "https://github.com/huggingface/datasets/pull/8134.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8134.patch",
"merged_at": "2026-04-15T10:15... | true | chore: bump doc-builder SHA for PR upload workflow
Switch the PR doc upload flow from the legacy dataset push to the new HF bucket.
| [] |
https://api.github.com/repos/huggingface/datasets/issues/8133 | 8,133 | use full repo-relative file path instead of filename for agent traces | We want to include full path of the file so it's easier to fetch the full trace from repo
- Store the full repo-relative path in `file_name` instead of just the basename
- Extracts the path from the `hf://` origin URL set by the download manager
Before: `file_name = "614b8d50-097f-4e36-aee0-1fdd053335e7.js... | closed | 2026-04-14T21:56:43 | https://api.github.com/repos/huggingface/datasets/issues/8133/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8133",
"html_url": "https://github.com/huggingface/datasets/pull/8133",
"diff_url": "https://github.com/huggingface/datasets/pull/8133.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8133.patch",
"merged_at": "2026-04-15T11:18... | true | use full repo-relative file path instead of filename for agent traces
We want to include full path of the file so it's easier to fetch the full trace from repo
- Store the full repo-relative path in `file_name` instead of just the basename
- Extracts the path from the `hf://` origin URL set by the download mana... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8132 | 8,132 | fix(Dataset.map): initialize writer on first non-None output instead of first index | ## What does this PR do?
Fixes #7990
## The Bug
When a map function returns `None` for the first examples and a dict for
later ones, `Dataset.map()` crashes with:
AttributeError: 'NoneType' object has no attribute 'write'
This happens because the ArrowWriter was initialized only when processing
ind... | open | 2026-04-14T08:48:37 | https://api.github.com/repos/huggingface/datasets/issues/8132/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8132",
"html_url": "https://github.com/huggingface/datasets/pull/8132",
"diff_url": "https://github.com/huggingface/datasets/pull/8132.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8132.patch",
"merged_at": null
} | true | fix(Dataset.map): initialize writer on first non-None output instead of first index
## What does this PR do?
Fixes #7990
## The Bug
When a map function returns `None` for the first examples and a dict for
later ones, `Dataset.map()` crashes with:
AttributeError: 'NoneType' object has no attribute 'writ... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8131 | 8,131 | datasets.to_list: Behaviour change in version 4.7.0 with the addition of JSON type | ### Describe the bug
There is a change in the approach to decoding JSON types, added in https://github.com/huggingface/datasets/commit/d560b58e1f1ec96591a95a1780ee118f5eea6c74. Since 4.7.0, to_list() returns raw JSON strings for columns stored as Json(), while direct access via __getitem__ returns dicts.
In versions... | closed | 2026-04-13T06:06:43 | https://api.github.com/repos/huggingface/datasets/issues/8131/comments | null | false | datasets.to_list: Behaviour change in version 4.7.0 with the addition of JSON type
### Describe the bug
There is a change in the approach to decoding JSON types, added in https://github.com/huggingface/datasets/commit/d560b58e1f1ec96591a95a1780ee118f5eea6c74. Since 4.7.0, to_list() returns raw JSON strings for columns... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8130 | 8,130 | fix: mutable default argument in `_visit_with_path` | Replace `visit_path: _VisitPath = []` with `Optional[_VisitPath] = None` to avoid the mutable default argument anti-pattern ([B006](https://docs.astral.sh/ruff/rules/mutable-argument-default/)).
While the current callers always use `visit_path + [k]` (creating new lists), the shared default list is a latent bug that w... | open | 2026-04-11T05:36:02 | https://api.github.com/repos/huggingface/datasets/issues/8130/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8130",
"html_url": "https://github.com/huggingface/datasets/pull/8130",
"diff_url": "https://github.com/huggingface/datasets/pull/8130.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8130.patch",
"merged_at": null
} | true | fix: mutable default argument in `_visit_with_path`
Replace `visit_path: _VisitPath = []` with `Optional[_VisitPath] = None` to avoid the mutable default argument anti-pattern ([B006](https://docs.astral.sh/ruff/rules/mutable-argument-default/)).
While the current callers always use `visit_path + [k]` (creating new li... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8129 | 8,129 | `load_dataset()` hangs when `hf_xet` is enabled | ### Describe the bug
Hi! I am getting a peculiar download issue on the latest versions of `datasets` and `hfh` where datasets hang indefinitely:
```py
from datasets import load_dataset
load_dataset("openai/gsm8k", "main", split="test")
```
Note that:
- streaming=True works.
- Direct Hub access works.
- Th... | closed | 2026-04-10T21:00:24 | https://api.github.com/repos/huggingface/datasets/issues/8129/comments | null | false | `load_dataset()` hangs when `hf_xet` is enabled
### Describe the bug
Hi! I am getting a peculiar download issue on the latest versions of `datasets` and `hfh` where datasets hang indefinitely:
```py
from datasets import load_dataset
load_dataset("openai/gsm8k", "main", split="test")
```
Note that:
- streaming... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8128 | 8,128 | Don't include files list DatasetInfo (and remove old stuff) | this will fix the viewer when it saves the dataset info in mongo, for datasets with a lot of files this can make the mongo doc too big for mongo
I also removed old deprecated stuff linked to files list like the old concept of "post-processing" and cleaned the checksum recording logic | closed | 2026-04-10T15:30:24 | https://api.github.com/repos/huggingface/datasets/issues/8128/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8128",
"html_url": "https://github.com/huggingface/datasets/pull/8128",
"diff_url": "https://github.com/huggingface/datasets/pull/8128.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8128.patch",
"merged_at": "2026-04-16T09:23... | true | Don't include files list DatasetInfo (and remove old stuff)
this will fix the viewer when it saves the dataset info in mongo, for datasets with a lot of files this can make the mongo doc too big for mongo
I also removed old deprecated stuff linked to files list like the old concept of "post-processing" and cleaned t... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8127 | 8,127 | Support explicit harness field in agent traces | Hi!
This small PR allows an optional `harness` field to be parsed.
When present, the harness field takes priority over type-value heuristics.
This allows agents to self-identify even when they contain property names that collide with an existing harness ("session", "system", etc.) | closed | 2026-04-08T15:39:48 | https://api.github.com/repos/huggingface/datasets/issues/8127/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8127",
"html_url": "https://github.com/huggingface/datasets/pull/8127",
"diff_url": "https://github.com/huggingface/datasets/pull/8127.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8127.patch",
"merged_at": null
} | true | Support explicit harness field in agent traces
Hi!
This small PR allows an optional `harness` field to be parsed.
When present, the harness field takes priority over type-value heuristics.
This allows agents to self-identify even when they contain property names that collide with an existing harness ("session"... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8126 | 8,126 | Fix batching for table-formatted datasets | ## Summary
Fix #8075
Fix `.batch()` on table-formatted `Dataset` and `IterableDataset`.
Before this change, calling `.batch()` on datasets formatted as `pyarrow`, `pandas`, or `polars` could fail because the batching path assumed dict-like inputs. This updates batching to use an Arrow-based path for table-style ... | closed | 2026-04-08T13:05:36 | https://api.github.com/repos/huggingface/datasets/issues/8126/comments | {
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8126",
"html_url": "https://github.com/huggingface/datasets/pull/8126",
"diff_url": "https://github.com/huggingface/datasets/pull/8126.diff",
"patch_url": "https://github.com/huggingface/datasets/pull/8126.patch",
"merged_at": "2026-04-10T13:14... | true | Fix batching for table-formatted datasets
## Summary
Fix #8075
Fix `.batch()` on table-formatted `Dataset` and `IterableDataset`.
Before this change, calling `.batch()` on datasets formatted as `pyarrow`, `pandas`, or `polars` could fail because the batching path assumed dict-like inputs. This updates batching t... | [] |
https://api.github.com/repos/huggingface/datasets/issues/8125 | 8,125 | Feature request: being able to disable chmod for flat permission filesystems | ### Feature request
For flat permission file systems, `os.chmod` simply fails when trying to cache datasets - in some places such as: https://github.com/huggingface/datasets/blob/main/src/datasets/arrow_dataset.py#L4043-L4048 for arrow datasets.
This seems to be optional to obtained desired behavior as simply comment... | open | 2026-04-08T08:17:14 | https://api.github.com/repos/huggingface/datasets/issues/8125/comments | null | false | Feature request: being able to disable chmod for flat permission filesystems
### Feature request
For flat permission file systems, `os.chmod` simply fails when trying to cache datasets - in some places such as: https://github.com/huggingface/datasets/blob/main/src/datasets/arrow_dataset.py#L4043-L4048 for arrow datase... | [] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.