url stringlengths 61 61 | repository_url stringclasses 1
value | labels_url stringlengths 75 75 | comments_url stringlengths 70 70 | events_url stringlengths 68 68 | html_url stringlengths 49 51 | id int64 1.14B 6.11B | node_id stringlengths 18 25 | number int64 3.74k 8.91k | title stringlengths 1 290 | labels listlengths 0 4 | state stringclasses 2
values | locked bool 1
class | comments listlengths 0 49 | created_at stringdate 2022-02-16 15:20:57 2026-10-29 22:11:42 | updated_at stringdate 2022-02-17 10:15:17 2026-10-30 01:00:30 | closed_at stringlengths 20 20 ⌀ | author_association stringclasses 4
values | issue_field_values listlengths 0 0 | type null | active_lock_reason null | draft bool 2
classes | pull_request dict | body stringlengths 1 59.9k ⌀ | closed_by dict | reactions dict | timeline_url stringlengths 70 70 | state_reason stringclasses 4
values | comments_count int64 0 49 | is_pull_request bool 2
classes | label_names listlengths 0 4 | days_to_close float64 0 1.46k ⌀ | created_year int64 2.02k 2.03k | title_word_count int64 1 44 | body_word_count int64 0 4.77k | has_body bool 2
classes | source_repo stringclasses 1
value | collected_at stringdate 2026-10-30 06:37:26 2026-10-30 06:37:26 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
https://api.github.com/repos/huggingface/datasets/issues/8908 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8908/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8908/comments | https://api.github.com/repos/huggingface/datasets/issues/8908/events | https://github.com/huggingface/datasets/pull/8908 | 6,109,381,173 | PR_kwDODunzps8AAAABLpILBQ | 8,908 | Close xopen handles in xnumpy_load, xpandas_read_excel, xpyarrow_parquet_read_table, xsio_loadmat | [] | open | false | [] | 2026-10-29T22:11:42Z | 2026-10-29T22:11:42Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8908.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8908",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8908.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8908"
} | Fixes #8907
Follow-up to #8786, which fixed this for `xpandas_read_csv`. The remaining streaming readers still handed `xopen(...)` to the reader without closing it:
- `xpyarrow_parquet_read_table`, `xsio_loadmat`: wrap in `with xopen(...)`.
- `xpandas_read_excel`: both the direct read and the BytesIO fallback now use... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8908/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8908/timeline | null | 0 | true | [] | null | 2,026 | 8 | 137 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8907 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8907/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8907/comments | https://api.github.com/repos/huggingface/datasets/issues/8907/events | https://github.com/huggingface/datasets/issues/8907 | 6,109,377,081 | I_kwDODunzps8AAAABbCWyOQ | 8,907 | xnumpy_load, xpandas_read_excel, xpyarrow_parquet_read_table and xsio_loadmat never close their xopen handle | [] | open | false | [] | 2026-10-29T22:11:22Z | 2026-10-29T22:11:22Z | null | NONE | [] | null | null | null | null | ### Describe the bug
#8786 fixed the leaked `xopen` handle in `xpandas_read_csv`, but the other streaming readers in `utils/file_utils.py` still pass `xopen(...)` straight to the reader and never close it:
- `xnumpy_load`
- `xpandas_read_excel` (both the first try and the BytesIO fallback)
- `xpyarrow_parquet_read_ta... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8907/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8907/timeline | null | 0 | false | [] | null | 2,026 | 10 | 206 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8905 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8905/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8905/comments | https://api.github.com/repos/huggingface/datasets/issues/8905/events | https://github.com/huggingface/datasets/pull/8905 | 6,100,014,532 | PR_kwDODunzps8AAAABLhhhYw | 8,905 | Bump doc-builder to get the docs machine-translated | [] | open | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8905). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-29T12:50:36Z | 2026-10-29T12:52:37Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8905.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8905",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8905.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8905"
} | Bumps doc-builder to the latest `main` so the docs on hf.co/docs get [machine-translated](https://github.com/huggingface/doc-builder#machine-translated-documentation) into 11 languages, as discussed on Slack. Pages built by older doc-builder versions can't be translated.
🤖 Generated with [Claude Code](https://claude.... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8905/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8905/timeline | null | 1 | true | [] | null | 2,026 | 7 | 34 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8904 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8904/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8904/comments | https://api.github.com/repos/huggingface/datasets/issues/8904/events | https://github.com/huggingface/datasets/pull/8904 | 6,093,849,240 | PR_kwDODunzps8AAAABLcheZA | 8,904 | Preserve on-the-fly transforms when flattening dataset indices | [] | open | false | [] | 2026-10-29T05:55:15Z | 2026-10-29T05:59:01Z | null | NONE | [] | null | null | true | {
"diff_url": "https://github.com/huggingface/datasets/pull/8904.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8904",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8904.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8904"
} | Dataset.flatten_indices() currently passes the dataset’s active format to an identity map. With a custom transform, the transform’s output is written to Arrow, while the result retains the same transform and applies it again on access.
For example, for stored values [2, 0] and an on-the-fly transform that adds 10, fla... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8904/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8904/timeline | null | 0 | true | [] | null | 2,026 | 7 | 472 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8903 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8903/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8903/comments | https://api.github.com/repos/huggingface/datasets/issues/8903/events | https://github.com/huggingface/datasets/pull/8903 | 6,080,889,794 | PR_kwDODunzps8AAAABLSFV4A | 8,903 | Load Harbor multi-step tasks without a top-level instruction.md | [] | open | false | [] | 2026-10-28T14:23:32Z | 2026-10-28T14:23:32Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8903.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8903",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8903.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8903"
} | Title: Load Harbor multi-step tasks without a top-level instruction.md
A multi-step Harbor task declares `[[steps]]` in its `task.toml` and keeps its instructions in `steps/<step>/instruction.md`, so it doesn't need an `instruction.md` next to its `task.toml`. Harbor handles that case in several places:
- `Task` sets... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8903/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8903/timeline | null | 0 | true | [] | null | 2,026 | 8 | 479 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8902 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8902/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8902/comments | https://api.github.com/repos/huggingface/datasets/issues/8902/events | https://github.com/huggingface/datasets/pull/8902 | 6,078,754,655 | PR_kwDODunzps8AAAABLQXf4A | 8,902 | Allow reading prepared datasets from read-only shared caches | [] | open | false | [] | 2026-10-28T12:17:51Z | 2026-10-28T12:17:51Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8902.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8902",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8902.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8902"
} | # Allow reading prepared datasets from read-only shared caches
Addresses the Datasets lock-file failure reported in huggingface/datasets#7900. Hub cache permissions are outside this patch.
## Problem
`DatasetBuilder` acquires exclusive file locks while checking cached metadata and while deciding whether a dataset ne... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8902/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8902/timeline | null | 0 | true | [] | null | 2,026 | 8 | 245 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8901 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8901/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8901/comments | https://api.github.com/repos/huggingface/datasets/issues/8901/events | https://github.com/huggingface/datasets/pull/8901 | 6,075,858,208 | PR_kwDODunzps8AAAABLODlpQ | 8,901 | Use the same class names in every split in DatasetDict.class_encode_column | [] | open | false | [] | 2026-10-28T09:06:02Z | 2026-10-28T09:06:02Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8901.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8901",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8901.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8901"
} | `DatasetDict.class_encode_column` encoded each split on its own, so a split that lacks some classes got a shorter `ClassLabel`, and the same label got a different integer id than in the other splits. Using the train features to decode the ids of another split then returned the wrong class, and concatenating the splits ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8901/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8901/timeline | null | 0 | true | [] | null | 2,026 | 10 | 310 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8900 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8900/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8900/comments | https://api.github.com/repos/huggingface/datasets/issues/8900/events | https://github.com/huggingface/datasets/pull/8900 | 6,068,218,472 | PR_kwDODunzps8AAAABLIAHIg | 8,900 | Restore parent tokenizer environment after multiprocessing map | [] | open | false | [] | 2026-10-27T21:41:44Z | 2026-10-27T21:41:44Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8900.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8900",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8900.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8900"
} | `Dataset.map(num_proc=2)` currently leaks its temporary `TOKENIZERS_PARALLELISM=false` override into the parent process. On a cold map, Python's copied `os.environ` reports the old value while a newly launched subprocess still inherits `false`. Fully cached maps and pool-constructor errors leave the Python mapping chan... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8900/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8900/timeline | null | 0 | true | [] | null | 2,026 | 7 | 248 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8899 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8899/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8899/comments | https://api.github.com/repos/huggingface/datasets/issues/8899/events | https://github.com/huggingface/datasets/pull/8899 | 6,062,954,883 | PR_kwDODunzps8AAAABLDweqg | 8,899 | fix(json): handle literal '/' in field names in mixed-type retry | [] | open | false | [] | 2026-10-27T16:03:22Z | 2026-10-27T16:19:06Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8899.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8899",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8899.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8899"
} | Fixes #8898
## Bug
When loading JSONL with the default `on_mixed_types="use_json"`, a record with a literal key such as `"a/b"` is misinterpreted: the PyArrow error column name is split into the nested path `["a", "b"]`, even though the record has one literal key `"a/b"`. Re-encoding with that nested path leaves the o... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8899/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8899/timeline | null | 0 | true | [] | null | 2,026 | 10 | 131 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8898 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8898/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8898/comments | https://api.github.com/repos/huggingface/datasets/issues/8898/events | https://github.com/huggingface/datasets/issues/8898 | 6,054,912,822 | I_kwDODunzps8AAAABaOajNg | 8,898 | JSON loading retries indefinitely when a mixed-type field name contains a slash | [] | open | false | [] | 2026-10-27T07:34:57Z | 2026-10-27T07:34:57Z | null | NONE | [] | null | null | null | null | ### Describe the bug
On main `ef61228a5756dcebfff8ee245d89c72822fb6ae6` (`datasets` 5.2.0.dev0), loading a two-row JSONL file with a literal `a/b` field mixing a number and a string does not finish with the default `on_mixed_types="use_json"`.
The same values under a plain field name load as `Json()`. A slash-contain... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8898/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8898/timeline | null | 0 | false | [] | null | 2,026 | 12 | 418 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8897 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8897/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8897/comments | https://api.github.com/repos/huggingface/datasets/issues/8897/events | https://github.com/huggingface/datasets/pull/8897 | 6,030,725,997 | PR_kwDODunzps8AAAABKpyD0A | 8,897 | Invalidate extraction caches when source archive metadata changes | [] | open | false | [] | 2026-10-25T23:43:53Z | 2026-10-25T23:46:25Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8897.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8897",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8897.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8897"
} | Replacing a local `.tar.gz` archive at the same path can return old rows from `load_dataset("json", ...)`, even after the builder detects the changed source and creates a new dataset fingerprint. The extraction cache currently identifies the archive by pathname alone.
This change includes the absolute path, file size,... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8897/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8897/timeline | null | 0 | true | [] | null | 2,026 | 8 | 396 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8896 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8896/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8896/comments | https://api.github.com/repos/huggingface/datasets/issues/8896/events | https://github.com/huggingface/datasets/pull/8896 | 6,028,268,797 | PR_kwDODunzps8AAAABKn2lxw | 8,896 | Fix batched map on a dataset with an index when the function returns a table | [] | open | false | [
"@lhoestq gentle ping: could you take a look when you have a moment? Batched `map()` on a dataset with an indices mapping crashes; the PR includes a regression test."
] | 2026-10-25T19:16:31Z | 2026-10-28T17:32:12Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8896.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8896",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8896.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8896"
} | `cast()` and `cast_column()` fail on any dataset that has a FAISS or Elasticsearch index attached:
```python
import numpy as np
from datasets import Dataset, Value
ds = Dataset.from_dict({"id": list(range(5)), "v": np.eye(5, dtype=np.float32).tolist()})
ds.add_faiss_index("v")
ds.cast_column("id", Value("int32"))
# A... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8896/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8896/timeline | null | 1 | true | [] | null | 2,026 | 15 | 199 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8895 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8895/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8895/comments | https://api.github.com/repos/huggingface/datasets/issues/8895/events | https://github.com/huggingface/datasets/pull/8895 | 6,027,361,872 | PR_kwDODunzps8AAAABKnJDsQ | 8,895 | feat(features): replace BioStructure with an AtomicStructure feature stored as BinaryCIF and read with atomworks | [] | open | false | [] | 2026-10-25T17:42:56Z | 2026-10-30T00:31:44Z | null | NONE | [] | null | null | true | {
"diff_url": "https://github.com/huggingface/datasets/pull/8895.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8895",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8895.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8895"
} | Replaces the experimental `BioStructure` and `BioSequence` features with one decodable feature that stores a structure as a [BinaryCIF](https://github.com/molstar/BinaryCIF) file and decodes it to the atomworks object the file describes. After this PR, nothing in `datasets` depends on Biopython.
| Type | Module | `dec... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8895/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8895/timeline | null | 0 | true | [] | null | 2,026 | 14 | 690 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8894 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8894/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8894/comments | https://api.github.com/repos/huggingface/datasets/issues/8894/events | https://github.com/huggingface/datasets/pull/8894 | 6,025,528,770 | PR_kwDODunzps8AAAABKluSSg | 8,894 | docs: propose evidence-first evolution intake and licensing gates | [] | open | false | [] | 2026-10-25T14:28:37Z | 2026-10-25T14:28:37Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8894.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8894",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8894.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8894"
} | null | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8894/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8894/timeline | null | 0 | true | [] | null | 2,026 | 8 | 0 | false | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8893 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8893/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8893/comments | https://api.github.com/repos/huggingface/datasets/issues/8893/events | https://github.com/huggingface/datasets/pull/8893 | 6,023,672,883 | PR_kwDODunzps8AAAABKkSUqw | 8,893 | Fix xgetsize fallback to count bytes instead of decoded characters | [] | open | false | [] | 2026-10-25T10:59:13Z | 2026-10-25T11:13:59Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8893.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8893",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8893.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8893"
} | When the filesystem cannot report a file size, `xgetsize` measures the file by reading it, but it opens it with `xopen(path)`, whose default mode is `"r"`. So the fallback decodes the file as text: binary files raise `UnicodeDecodeError`, and text files return a character count after newline translation instead of a by... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8893/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8893/timeline | null | 0 | true | [] | null | 2,026 | 10 | 357 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8892 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8892/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8892/comments | https://api.github.com/repos/huggingface/datasets/issues/8892/events | https://github.com/huggingface/datasets/pull/8892 | 6,018,163,293 | PR_kwDODunzps8AAAABKf_y5A | 8,892 | Forward ignore_decode_attribute in nested require_decoding calls | [] | open | false | [] | 2026-10-24T22:12:46Z | 2026-10-24T22:12:46Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8892.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8892",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8892.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8892"
} | ## What
`require_decoding(feature, ignore_decode_attribute=...)` drops `ignore_decode_attribute` when it recurses: the `dict`, `list`/`tuple`, `LargeList` and `List` branches call `require_decoding(f)` without forwarding the flag. Only a top-level feature honours it.
`ignore_decode_attribute=True` is passed by `Datas... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8892/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8892/timeline | null | 0 | true | [] | null | 2,026 | 6 | 150 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8891 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8891/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8891/comments | https://api.github.com/repos/huggingface/datasets/issues/8891/events | https://github.com/huggingface/datasets/pull/8891 | 6,014,808,547 | PR_kwDODunzps8AAAABKdZ_aw | 8,891 | Don't infer the Harbor builder from unrelated .md and .toml files | [] | open | false | [] | 2026-10-24T15:58:23Z | 2026-10-24T15:58:23Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8891.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8891",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8891.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8891"
} | Since #8729, `.md` and `.toml` are allowed data file extensions. Every Markdown or TOML file matched by the default `**` pattern now counts as a vote for `harbor` when the builder is inferred. So a directory with `data.jsonl`, `LICENSE.md` and `CHANGELOG.md` is inferred as Harbor, and so is `data.csv` with `pyproject.t... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8891/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8891/timeline | null | 0 | true | [] | null | 2,026 | 11 | 188 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8890 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8890/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8890/comments | https://api.github.com/repos/huggingface/datasets/issues/8890/events | https://github.com/huggingface/datasets/pull/8890 | 6,014,230,128 | PR_kwDODunzps8AAAABKc91DA | 8,890 | keep nullable integers intact in to_pandas | [] | open | false | [] | 2026-10-24T14:54:55Z | 2026-10-24T14:54:55Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8890.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8890",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8890.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8890"
} | an int64 column with a null was converted to float64. float64 cannot hold every int64, so `2**62 + 1` came back as a different number. `to_pandas` and the pandas formatter now keep those nulls as missing values, the same way `to_csv` / `to_json` / `to_sql` already do. a column with no nulls stays int64.
checked locall... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8890/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8890/timeline | null | 0 | true | [] | null | 2,026 | 6 | 73 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8889 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8889/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8889/comments | https://api.github.com/repos/huggingface/datasets/issues/8889/events | https://github.com/huggingface/datasets/pull/8889 | 6,008,187,639 | PR_kwDODunzps8AAAABKYT83w | 8,889 | Keep row validity and payloads in fixed-shape ArrayXD pandas round trips | [] | open | false | [
"CI note: this is the first PR from my fork, so the `CI` and `Build PR Documentation` workflows are sitting in `action_required` until someone approves the first run. Everything in the description is verified locally on the issue's pinned versions (pandas 3.1.0, pyarrow 26.0.0, numpy 2.5.4): the five new test cases... | 2026-10-24T01:50:23Z | 2026-10-26T16:06:02Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8889.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8889",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8889.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8889"
} | Fixes #8888
`PandasArrayExtensionDtype.__from_arrow__` materialized every fixed-shape ArrayXD column through `ArrayExtensionArray.to_numpy`, which pads the null rows with NaN and promotes the whole column to float64. Two consequences at the pandas boundary: the Arrow row validity was gone (a null row and a legitimate ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8889/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8889/timeline | null | 4 | true | [] | null | 2,026 | 11 | 301 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8888 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8888/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8888/comments | https://api.github.com/repos/huggingface/datasets/issues/8888/events | https://github.com/huggingface/datasets/issues/8888 | 6,006,545,971 | I_kwDODunzps8AAAABZgSeMw | 8,888 | Fixed-shape ArrayXD pandas round trips lose row nulls and integer precision | [] | open | false | [
"The repro holds up, and the `PandasArrayExtensionDtype.__from_arrow__` going through `ArrayExtensionArray.to_numpy` explains both symptoms cleanly: NaN insertion kills the row-validity distinction and the float64 promotion loses the exact integer payloads. I am taking a pass at the masked-representation path, so t... | 2026-10-23T22:45:51Z | 2026-10-24T01:50:42Z | null | NONE | [] | null | null | null | null | ### Describe the bug
`Dataset.from_pandas(ds.to_pandas())` loses missing rows in a fixed-shape ArrayXD column. A null row becomes a valid all-NaN array. A legitimate all-NaN array must remain distinct from that missing row.
### Steps to reproduce the bug
Reproduced on main `ef61228a5756dcebfff8ee245d89c72822fb6ae6` ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8888/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8888/timeline | null | 2 | false | [] | null | 2,026 | 11 | 344 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8887 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8887/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8887/comments | https://api.github.com/repos/huggingface/datasets/issues/8887/events | https://github.com/huggingface/datasets/pull/8887 | 6,001,922,606 | PR_kwDODunzps8AAAABKTRcRg | 8,887 | Preserve selected streaming rows across concatenated epochs | [] | open | false | [] | 2026-10-23T16:51:29Z | 2026-10-23T16:51:29Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8887.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8887",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8887.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8887"
} | After concatenating a streaming dataset selected with `take(4)`, `set_epoch(1)` can change its rows from `[0, 1, 2, 3]` to `[0, 3, 6, 9]`. The same problem affects `skip()` and explicit shuffling after concatenation.
Propagate shuffling through vertical concatenation's shard wrappers so the existing skip/take ordering... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8887/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8887/timeline | null | 0 | true | [] | null | 2,026 | 7 | 163 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8886 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8886/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8886/comments | https://api.github.com/repos/huggingface/datasets/issues/8886/events | https://github.com/huggingface/datasets/pull/8886 | 5,996,193,964 | PR_kwDODunzps8AAAABKOpggg | 8,886 | Fix SQLAlchemy query cache collisions for bound parameters | [] | open | false | [
"Hi @lhoestq, I saw that #8881 also addresses #8877. I reported and assigned myself the issue before either PR was opened. My PR also covers explicit parameter overrides, callable parameters, and binary values, with regression tests and validation across different SQLAlchemy versions. Could you take a look and let ... | 2026-10-23T10:33:03Z | 2026-10-25T10:48:31Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8886.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8886",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8886.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8886"
} |
## What does this PR do?
Fixes #8877
`Dataset.from_sql()` can silently return incorrect cached rows when two SQLAlchemy queries differ only in their bound parameter values. The existing cache identity includes the compiled SQL placeholders but not their values, causing queries with different `WHERE`, `LIMIT/OFFSET`,... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8886/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8886/timeline | null | 1 | true | [] | null | 2,026 | 8 | 289 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8885 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8885/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8885/comments | https://api.github.com/repos/huggingface/datasets/issues/8885/events | https://github.com/huggingface/datasets/pull/8885 | 5,996,029,451 | PR_kwDODunzps8AAAABKOhAgA | 8,885 | Close Parquet export writers with a context manager | [] | open | false | [] | 2026-10-23T10:20:18Z | 2026-10-23T10:20:18Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8885.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8885",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8885.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8885"
} | Fixes #8743.
Use `pyarrow.parquet.ParquetWriter` as a context manager so query, batch-write, and metadata failures finalize the writer before the enclosing path-owned file handle closes. Caller-provided binary streams remain open, and the successful byte-count result is unchanged.
This is an alternative to #8744 and ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8885/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8885/timeline | null | 0 | true | [] | null | 2,026 | 8 | 187 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8884 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8884/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8884/comments | https://api.github.com/repos/huggingface/datasets/issues/8884/events | https://github.com/huggingface/datasets/pull/8884 | 5,995,468,709 | PR_kwDODunzps8AAAABKOEJ2g | 8,884 | Skip the spark tests when pyspark is not installed | [] | closed | false | [] | 2026-10-23T09:38:47Z | 2026-10-24T23:28:21Z | 2026-10-24T23:28:21Z | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8884.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8884",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8884.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8884"
} | Running the test suite without the optional `pyspark` dependency aborts collection instead of skipping the spark tests:
```
ERROR tests/packaged_modules/test_spark.py
ModuleNotFoundError: No module named 'pyspark'
```
`tests/packaged_modules/test_spark.py` imports `pyspark` at module level, so collection fails before... | {
"avatar_url": "https://avatars.githubusercontent.com/u/68082325?v=4",
"events_url": "https://api.github.com/users/ege-arhan/events{/privacy}",
"followers_url": "https://api.github.com/users/ege-arhan/followers",
"following_url": "https://api.github.com/users/ege-arhan/following{/other_user}",
"gists_url": "... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8884/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8884/timeline | null | 0 | true | [] | 1.576 | 2,026 | 9 | 110 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8883 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8883/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8883/comments | https://api.github.com/repos/huggingface/datasets/issues/8883/events | https://github.com/huggingface/datasets/pull/8883 | 5,995,442,766 | PR_kwDODunzps8AAAABKOC0LQ | 8,883 | Reject duplicate dataset card configuration names | [] | open | false | [] | 2026-10-23T09:36:50Z | 2026-10-23T09:36:50Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8883.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8883",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8883.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8883"
} | Dataset cards with two entries using the same `config_name` silently keep only the last entry when `MetadataConfigs.from_dataset_card_data` builds its dictionary. For example, two `custom` entries pointing at different CSV files cause `load_dataset(..., name="custom")` to load only the second file without warning.
R... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8883/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8883/timeline | null | 0 | true | [] | null | 2,026 | 6 | 173 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8881 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8881/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8881/comments | https://api.github.com/repos/huggingface/datasets/issues/8881/events | https://github.com/huggingface/datasets/pull/8881 | 5,992,136,423 | PR_kwDODunzps8AAAABKLaGPQ | 8,881 | include bound SQL values in the dataset cache key | [] | open | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8881). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.",
"hi ! can you run `make style` to fix code formatting ? thx !",
"formatted the compile... | 2026-10-23T04:55:34Z | 2026-10-24T14:48:30Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8881.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8881",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8881.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8881"
} | fixes #8877
`str(sql.compile())` drops the bound values, so two queries that differ only by a WHERE, LIMIT, or IN parameter reused one cache entry. the key now inlines those binds.
@lhoestq | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8881/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8881/timeline | null | 3 | true | [] | null | 2,026 | 9 | 31 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8880 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8880/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8880/comments | https://api.github.com/repos/huggingface/datasets/issues/8880/events | https://github.com/huggingface/datasets/pull/8880 | 5,991,993,396 | PR_kwDODunzps8AAAABKLS0rQ | 8,880 | Fix save_to_disk sending the whole table to every worker for in-memory datasets | [] | open | false | [] | 2026-10-23T04:39:59Z | 2026-10-23T04:39:59Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8880.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8880",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8880.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8880"
} | Fixes #7290
`save_to_disk(num_proc=...)` on an in-memory dataset sends the whole table to every worker, once per shard, instead of just that shard's rows. So it is much slower and uses much more memory than it should.
```python
import pickle
from datasets import Dataset
ds = Dataset.from_dict({"text": ["some text " ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8880/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8880/timeline | null | 0 | true | [] | null | 2,026 | 12 | 663 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8879 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8879/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8879/comments | https://api.github.com/repos/huggingface/datasets/issues/8879/events | https://github.com/huggingface/datasets/pull/8879 | 5,990,221,760 | PR_kwDODunzps8AAAABKJ5kNw | 8,879 | Fix pickle RCE in pandas dataset builder (CVE-2026-57845) | [] | open | false | [] | 2026-10-23T01:26:22Z | 2026-10-23T01:26:22Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8879.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8879",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8879.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8879"
} | Addresses **CVE-2026-57845**, arbitrary code execution through pickle deserialization in the pandas dataset builder.
Replaces `Pandas._generate_tables()` with an unconditional migration error before file opening or unpickling. Builder construction and compatible prepared Arrow cache reuse remain supported. File resolu... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8879/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8879/timeline | null | 0 | true | [] | null | 2,026 | 8 | 72 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8878 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8878/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8878/comments | https://api.github.com/repos/huggingface/datasets/issues/8878/events | https://github.com/huggingface/datasets/pull/8878 | 5,988,140,756 | PR_kwDODunzps8AAAABKIPsRg | 8,878 | Fix decoding a streamed PDF dataset twice after shuffle, take or filter | [] | open | false | [] | 2026-10-22T22:01:09Z | 2026-10-29T20:21:30Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8878.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8878",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8878.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8878"
} | Since #8791 (not in a release yet), a streamed PDF folder from the Hub is decoded in threads by default. Calling `.decode()` again after another step, such as `.shuffle()`, `.take()` or `.filter()`, decodes each example a second time, and `Pdf.decode_example` fails on the already decoded `pdfplumber` PDF:
```python
fr... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8878/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8878/timeline | null | 0 | true | [] | null | 2,026 | 12 | 430 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8877 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8877/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8877/comments | https://api.github.com/repos/huggingface/datasets/issues/8877/events | https://github.com/huggingface/datasets/issues/8877 | 5,987,283,253 | I_kwDODunzps8AAAABZN6xNQ | 8,877 | Dataset.from_sql caches different SQLAlchemy queries under the same key when bound values change | [] | open | false | [
"#self-assign"
] | 2026-10-22T20:53:46Z | 2026-10-22T20:54:04Z | null | CONTRIBUTOR | [] | null | null | null | null | ### Describe the bug
When `Dataset.from_sql()` receives a SQLAlchemy `Selectable`, queries with different parameter values can use the same cache key. If the first query is cached, the next query can return its rows instead of the correct results, without warning. This affects `WHERE` conditions, `LIMIT`, and expandin... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8877/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8877/timeline | null | 1 | false | [] | null | 2,026 | 13 | 337 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8876 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8876/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8876/comments | https://api.github.com/repos/huggingface/datasets/issues/8876/events | https://github.com/huggingface/datasets/pull/8876 | 5,987,109,741 | PR_kwDODunzps8AAAABKHaeEw | 8,876 | Faster IterableDataset iteration | [] | open | false | [] | 2026-10-22T20:41:03Z | 2026-10-22T20:41:03Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8876.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8876",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8876.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8876"
} | Fixes #6133
When iterating an `IterableDataset` example by example, rows go through `RebatchedArrowExamplesIterable` with `batch_size=1`, so each row ends up as a 1-row Arrow table that gets cast and formatted on its own. That per-row overhead is why `for ex in ds` is slow, including after `to_iterable_dataset()`.
Th... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8876/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8876/timeline | null | 0 | true | [] | null | 2,026 | 3 | 217 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8875 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8875/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8875/comments | https://api.github.com/repos/huggingface/datasets/issues/8875/events | https://github.com/huggingface/datasets/pull/8875 | 5,983,246,086 | PR_kwDODunzps8AAAABKEVyyg | 8,875 | Honor table policies when exporting empty datasets to SQL | [] | open | false | [] | 2026-10-22T16:20:36Z | 2026-10-22T16:20:36Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8875.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8875",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8875.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8875"
} | Exporting an empty dataset skips the SQL writer entirely. As a result, `if_exists="replace"` returns zero while leaving old rows in the table, `if_exists="fail"` accepts an existing table, and exporting a new empty split leaves its table missing.
Write one empty batch through the existing pandas conversion so the requ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8875/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8875/timeline | null | 0 | true | [] | null | 2,026 | 9 | 231 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8874 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8874/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8874/comments | https://api.github.com/repos/huggingface/datasets/issues/8874/events | https://github.com/huggingface/datasets/pull/8874 | 5,979,277,226 | PR_kwDODunzps8AAAABKBN9VQ | 8,874 | Preserve unrestricted filters when extending data-file patterns | [] | open | false | [] | 2026-10-22T12:27:16Z | 2026-10-22T12:27:23Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8874.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8874",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8874.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8874"
} | ## Summary
Preserve unrestricted extension filters when calling `DataFilesPatternsList.filter_extensions`.
`DataFilesPatternsList.from_patterns(["*"])` stores `allowed_extensions=[None]`, where `None` means unrestricted. Calling `.filter_extensions([".json"])` currently attempts `None + [".json"]` and raises `TypeErr... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8874/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8874/timeline | null | 0 | true | [] | null | 2,026 | 7 | 239 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8873 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8873/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8873/comments | https://api.github.com/repos/huggingface/datasets/issues/8873/events | https://github.com/huggingface/datasets/pull/8873 | 5,977,320,000 | PR_kwDODunzps8AAAABJ_o3Mw | 8,873 | Fix doubled words in load.py docstring and process docs | [] | open | false | [] | 2026-10-22T10:10:32Z | 2026-10-22T10:10:32Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8873.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8873",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8873.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8873"
} | ## What
Removed a duplicated 'a' in a load.py docstring ('of a a Storage Bucket') and a duplicated 'to' in docs/source/process.mdx ('save it to to push_to_hub').
## Why
A small, objectively-verifiable improvement (typo) found during a
daily open-source maintenance pass (2026-10-22).
## How verified
Found both with a ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8873/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8873/timeline | null | 0 | true | [] | null | 2,026 | 9 | 76 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8872 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8872/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8872/comments | https://api.github.com/repos/huggingface/datasets/issues/8872/events | https://github.com/huggingface/datasets/pull/8872 | 5,975,371,876 | PR_kwDODunzps8AAAABJ-EYoA | 8,872 | Preserve grouped streaming batches when resuming checkpoints | [] | open | false | [] | 2026-10-22T07:55:50Z | 2026-10-22T07:55:50Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8872.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8872",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8872.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8872"
} | `IterableDataset.batch(by_column=...)` can silently drop a complete group after an ordinary checkpoint restore. For example, group IDs 0–8 into triples, consume the first two groups, then save and reload state: uninterrupted iteration returns `[6, 7, 8]`, while current main returns nothing.
The grouping callback buffe... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8872/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8872/timeline | null | 0 | true | [] | null | 2,026 | 7 | 324 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8871 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8871/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8871/comments | https://api.github.com/repos/huggingface/datasets/issues/8871/events | https://github.com/huggingface/datasets/pull/8871 | 5,974,424,182 | PR_kwDODunzps8AAAABJ9Ty0g | 8,871 | Honor batch_size when exporting an IterableDataset to Parquet | [] | open | false | [] | 2026-10-22T06:41:18Z | 2026-10-22T06:41:18Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8871.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8871",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8871.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8871"
} | `IterableDataset.to_parquet(..., batch_size=2)` currently ignores the requested batch size: exporting seven rows creates a single seven-row Parquet row group, while `Dataset.to_parquet` creates row groups of `[2, 2, 2, 1]`. This prevents callers from controlling the row groups used when exporting streaming training dat... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8871/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8871/timeline | null | 0 | true | [] | null | 2,026 | 8 | 176 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8870 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8870/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8870/comments | https://api.github.com/repos/huggingface/datasets/issues/8870/events | https://github.com/huggingface/datasets/pull/8870 | 5,972,833,087 | PR_kwDODunzps8AAAABJ8DYIQ | 8,870 | Avoid copying WebDataset binary payloads during schema inference | [] | open | false | [] | 2026-10-22T04:04:04Z | 2026-10-22T04:04:04Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8870.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8870",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8870.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8870"
} | WebDataset schema inference currently builds Arrow tables containing the complete binary payloads from its first examples. A single value above Arrow's binary offset limit raises `ArrowCapacityError`, even with `streaming=True`.
Use empty byte strings only in the temporary examples passed to schema inference. This pre... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8870/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8870/timeline | null | 0 | true | [] | null | 2,026 | 8 | 210 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8869 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8869/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8869/comments | https://api.github.com/repos/huggingface/datasets/issues/8869/events | https://github.com/huggingface/datasets/pull/8869 | 5,972,157,921 | PR_kwDODunzps8AAAABJ7hj3Q | 8,869 | Unescape doubled quotes in GenBank qualifier values | [] | closed | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8869). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-22T02:52:20Z | 2026-10-22T10:43:33Z | 2026-10-22T10:43:33Z | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8869.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8869",
"merged_at": "2026-10-22T10:43:32Z",
"patch_url": "https://github.com/huggingface/datasets/pull/8869.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/... | The GenBank loader keeps the `""` escape inside quoted qualifier values. A wrapped line that ends in an escaped `""` also closes the value early, and a stray `"` lands at the end. The INSDC feature table (section 3.3.3.1) escapes each quotation mark in free text by doubling it. Its own example is `/note="This is an exa... | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8869/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8869/timeline | null | 1 | true | [] | 0.327 | 2,026 | 7 | 388 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8868 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8868/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8868/comments | https://api.github.com/repos/huggingface/datasets/issues/8868/events | https://github.com/huggingface/datasets/pull/8868 | 5,970,386,363 | PR_kwDODunzps8AAAABJ6IhkA | 8,868 | Raise on duplicate column names in Dataset.select_columns | [] | open | false | [] | 2026-10-21T23:50:02Z | 2026-10-21T23:50:02Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8868.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8868",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8868.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8868"
} | `Dataset.select_columns` accepts a list with repeated names and returns a dataset that looks fine but breaks on first use:
```python
from datasets import Dataset
ds = Dataset.from_dict({"a": [1, 2], "b": [3, 4]})
sel = ds.select_columns(["a", "a"])
print(sel.column_names, len(sel.features)) # ['a', 'a'] 1
sel[0] ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8868/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8868/timeline | null | 0 | true | [] | null | 2,026 | 7 | 203 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8867 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8867/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8867/comments | https://api.github.com/repos/huggingface/datasets/issues/8867/events | https://github.com/huggingface/datasets/pull/8867 | 5,944,509,212 | PR_kwDODunzps8AAAABJlXwiQ | 8,867 | Preserve rows when concatenating pandas ArrayXD batches | [] | open | false | [
"Hi! Following up on this fix when you have a chance to review. It preserves row counts and shapes when concatenating pandas ArrayXD batches, with regression coverage for dense, ragged, empty, and null-containing inputs. The reproduction and local test results are in the description. Happy to adjust the implementat... | 2026-10-20T15:29:02Z | 2026-10-25T06:28:22Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8867.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8867",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8867.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8867"
} | `PandasArrayExtensionArray._concat_same_type` can turn whole input batches into individual rows. For example, concatenating dense batches with 2 and 1 rows returns 2 rows instead of 3, with an extra batch dimension inside each row. A single input chunk is also wrapped as one row. Equally sized object-backed chunks take... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8867/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8867/timeline | null | 1 | true | [] | null | 2,026 | 7 | 256 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8866 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8866/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8866/comments | https://api.github.com/repos/huggingface/datasets/issues/8866/events | https://github.com/huggingface/datasets/pull/8866 | 5,943,309,841 | PR_kwDODunzps8AAAABJkZ4jg | 8,866 | Fix forwarding custom User-Agent to fsspec HTTP requests | [] | open | false | [] | 2026-10-20T14:17:24Z | 2026-10-21T20:17:28Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8866.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8866",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8866.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8866"
} | ## Summary
Related to #6046.
Forward the configured DownloadConfig.user_agent to fsspec HTTP/HTTPS requests.
- Reuse the existing get_datasets_user_agent() helper.
- Preserve explicitly configured User-Agent headers.
- Keep existing HTTP proxy handling unchanged.
- Add regression tests for default and custom header ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8866/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8866/timeline | null | 0 | true | [] | null | 2,026 | 8 | 73 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8865 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8865/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8865/comments | https://api.github.com/repos/huggingface/datasets/issues/8865/events | https://github.com/huggingface/datasets/pull/8865 | 5,939,230,188 | PR_kwDODunzps8AAAABJhKMkg | 8,865 | Fix concatenation cache fingerprints for different axes | [] | closed | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8865). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-20T09:22:24Z | 2026-10-25T13:42:17Z | 2026-10-23T12:22:31Z | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8865.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8865",
"merged_at": "2026-10-23T12:22:31Z",
"patch_url": "https://github.com/huggingface/datasets/pull/8865.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/... |
## What does this PR do?
Fixes #8859
Concatenating the same datasets using different axes can produce the same fingerprint. As a result, `map()` can reuse cached results from the wrong concatenation without warning.
This change adds `axis` to the existing fingerprint calculation and adds two regression tests for bo... | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8865/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8865/timeline | null | 1 | true | [] | 3.125 | 2,026 | 7 | 186 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8864 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8864/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8864/comments | https://api.github.com/repos/huggingface/datasets/issues/8864/events | https://github.com/huggingface/datasets/pull/8864 | 5,939,167,230 | PR_kwDODunzps8AAAABJhHAeA | 8,864 | Keep threaded decoding reconfigurable after IterableDataset.shard | [] | open | false | [] | 2026-10-20T09:17:53Z | 2026-10-20T09:17:53Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8864.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8864",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8864.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8864"
} | Since #8791, `load_dataset(..., streaming=True)` decodes media datasets from the Hub in worker threads, and `IterableDataset.decode()` replaces that threaded step when it finds it as the last step of the dataset (`MappedExamplesIterable.decoding_pool`).
`MappedExamplesIterable.shard_data_sources`, `shuffle_data_source... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8864/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8864/timeline | null | 0 | true | [] | null | 2,026 | 6 | 180 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8863 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8863/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8863/comments | https://api.github.com/repos/huggingface/datasets/issues/8863/events | https://github.com/huggingface/datasets/issues/8863 | 5,939,160,827 | I_kwDODunzps8AAAABYgBm-w | 8,863 | IterableDataset.decode(False) has no effect after shard() on a streamed media dataset | [] | open | false | [] | 2026-10-20T09:17:23Z | 2026-10-20T09:17:23Z | null | NONE | [] | null | null | null | null | ### Describe the bug
Since #8791, `load_dataset(..., streaming=True)` decodes media datasets from the Hub in worker threads by default. `decode()` replaces that threaded decoding step when it is the last step of the dataset, but `shard()` loses track of it. After `ds.shard(...)`, `ds.decode(False)` keeps yielding deco... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8863/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8863/timeline | null | 0 | false | [] | null | 2,026 | 11 | 176 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8862 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8862/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8862/comments | https://api.github.com/repos/huggingface/datasets/issues/8862/events | https://github.com/huggingface/datasets/pull/8862 | 5,938,823,796 | PR_kwDODunzps8AAAABJg1jBQ | 8,862 | Close owned event loops and pending work in streaming async maps | [] | open | false | [] | 2026-10-20T08:52:50Z | 2026-10-20T08:52:50Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8862.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8862",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8862.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8862"
} | When an async-mapped `IterableDataset` creates an event loop for iteration, it retains that loop without closing it. Repeated epochs therefore leak file descriptors. Early iterator closure also leaves pending callbacks without completing their asynchronous cleanup.
Track loop ownership per iterator, settle that iterat... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8862/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8862/timeline | null | 0 | true | [] | null | 2,026 | 11 | 215 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8861 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8861/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8861/comments | https://api.github.com/repos/huggingface/datasets/issues/8861/events | https://github.com/huggingface/datasets/pull/8861 | 5,938,747,991 | PR_kwDODunzps8AAAABJgxs4g | 8,861 | Embed nested undecoded media in Hub Parquet shards | [] | open | false | [] | 2026-10-20T08:47:21Z | 2026-10-20T08:47:21Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8861.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8861",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8861.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8861"
} | `Dataset.push_to_hub()` can upload path-only records for images or audio nested inside a struct, `List`, or `LargeList` when `decode=False`, even with `embed_external_files=True`. The resulting Parquet shards refer to files on the uploader's machine instead of containing their bytes.
Use the existing recursive `requir... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8861/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8861/timeline | null | 0 | true | [] | null | 2,026 | 8 | 153 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8860 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8860/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8860/comments | https://api.github.com/repos/huggingface/datasets/issues/8860/events | https://github.com/huggingface/datasets/pull/8860 | 5,938,641,082 | PR_kwDODunzps8AAAABJgsbSA | 8,860 | Fix HDF5 compound array field inference | [] | open | false | [] | 2026-10-20T08:39:31Z | 2026-10-20T08:39:31Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8860.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8860",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8860.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8860"
} | HDF5 compound records can contain numeric array fields, such as `np.dtype([("id", "int32"), ("values", "float32", (3,))])`. Loading this valid file currently raises `TypeError: 'NoneType' object is not iterable` during schema inference: the subarray dtype is mistaken for another compound record.
Keep the field dimensi... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8860/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8860/timeline | null | 0 | true | [] | null | 2,026 | 6 | 156 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8859 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8859/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8859/comments | https://api.github.com/repos/huggingface/datasets/issues/8859/events | https://github.com/huggingface/datasets/issues/8859 | 5,938,116,418 | I_kwDODunzps8AAAABYfB3Qg | 8,859 | concatenate_datasets reuses preprocessing caches across different axes | [] | closed | false | [
"#self-assign"
] | 2026-10-20T07:59:58Z | 2026-10-23T12:22:33Z | 2026-10-23T12:22:33Z | CONTRIBUTOR | [] | null | null | null | null | ### Describe the bug
Concatenating the same disk-backed datasets using different axes gives the same fingerprint, even though the resulting datasets are different. As a result, calling `map()` on one can reuse the cached results of the other without warning.
In the example below, a dataset with two rows of paired tex... | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8859/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8859/timeline | completed | 1 | false | [] | 3.182 | 2,026 | 7 | 282 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8858 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8858/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8858/comments | https://api.github.com/repos/huggingface/datasets/issues/8858/events | https://github.com/huggingface/datasets/pull/8858 | 5,935,012,144 | PR_kwDODunzps8AAAABJd3NcQ | 8,858 | Fix inverted split slices reading the rest of the split | [] | open | false | [] | 2026-10-20T02:45:37Z | 2026-10-20T02:45:37Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8858.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8858",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8858.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8858"
} | A split slice whose end falls one row before its start returns the rest of the split instead of nothing:
```python
import pathlib, tempfile
from datasets import load_dataset
path = pathlib.Path(tempfile.mkdtemp()) / "data.csv"
path.write_text("x\n" + "\n".join(map(str, range(10))) + "\n")
print(list(load_dataset("csv... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8858/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8858/timeline | null | 0 | true | [] | null | 2,026 | 10 | 183 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8857 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8857/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8857/comments | https://api.github.com/repos/huggingface/datasets/issues/8857/events | https://github.com/huggingface/datasets/pull/8857 | 5,928,108,354 | PR_kwDODunzps8AAAABJYYojg | 8,857 | set dev 5.2.0 | [] | closed | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8857). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-19T17:04:51Z | 2026-10-19T17:33:23Z | 2026-10-19T17:33:21Z | MEMBER | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8857.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8857",
"merged_at": "2026-10-19T17:33:21Z",
"patch_url": "https://github.com/huggingface/datasets/pull/8857.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/... | null | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8857/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8857/timeline | null | 1 | true | [] | 0.02 | 2,026 | 3 | 0 | false | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8856 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8856/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8856/comments | https://api.github.com/repos/huggingface/datasets/issues/8856/events | https://github.com/huggingface/datasets/pull/8856 | 5,927,151,696 | PR_kwDODunzps8AAAABJXn3aA | 8,856 | Reject duplicate WebDataset fields instead of silently overwriting data | [] | closed | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8856). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-19T15:56:27Z | 2026-10-19T18:07:32Z | 2026-10-19T18:07:32Z | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8856.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8856",
"merged_at": "2026-10-19T18:07:32Z",
"patch_url": "https://github.com/huggingface/datasets/pull/8856.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/... | A TAR archive can contain consecutive members with the same name, such as two `00000.txt` entries with different contents. The WebDataset pipeline currently overwrites the first value and returns only the second without reporting that data was lost.
This change raises `ValueError` before overwriting an existing field ... | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8856/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8856/timeline | null | 1 | true | [] | 0.091 | 2,026 | 9 | 183 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8855 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8855/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8855/comments | https://api.github.com/repos/huggingface/datasets/issues/8855/events | https://github.com/huggingface/datasets/pull/8855 | 5,925,499,669 | PR_kwDODunzps8AAAABJWSn8A | 8,855 | Preserve active iterators when reconfiguring threaded decoding | [] | closed | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8855). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-19T14:12:56Z | 2026-10-19T18:02:53Z | 2026-10-19T18:02:53Z | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8855.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8855",
"merged_at": "2026-10-19T18:02:53Z",
"patch_url": "https://github.com/huggingface/datasets/pull/8855.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/... | Creating a new decoding view from a threaded `IterableDataset` currently shuts down the original dataset's decoding pool. If an existing iterator has prefetched examples queued behind active decoders, calling `ds.decode(False)` or changing its thread count cancels those examples. Continuing the original iterator then r... | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8855/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8855/timeline | null | 1 | true | [] | 0.16 | 2,026 | 7 | 226 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8854 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8854/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8854/comments | https://api.github.com/repos/huggingface/datasets/issues/8854/events | https://github.com/huggingface/datasets/pull/8854 | 5,919,837,554 | PR_kwDODunzps8AAAABJRu_8Q | 8,854 | Fix duplicated word in FileLock umask comment | [] | open | false | [] | 2026-10-19T07:38:31Z | 2026-10-19T07:38:31Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8854.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8854",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8854.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8854"
} | A one-word comment typo in `src/datasets/utils/_filelock.py`: the comment reads "In previous **previous** it was already using the current umask", duplicating a word and dropping the noun.
```diff
# The "mode" argument is required if we want to use the current umask in filelock >= 3.10
- # In previous ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8854/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8854/timeline | null | 0 | true | [] | null | 2,026 | 7 | 92 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8853 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8853/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8853/comments | https://api.github.com/repos/huggingface/datasets/issues/8853/events | https://github.com/huggingface/datasets/pull/8853 | 5,918,203,410 | PR_kwDODunzps8AAAABJQbZvA | 8,853 | Drop a leading UTF-8 BOM in the conll loader | [] | open | false | [] | 2026-10-19T05:10:48Z | 2026-10-19T09:13:29Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8853.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8853",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8853.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8853"
} | ### What was wrong
`load_dataset("conll", ...)` keeps a leading UTF-8 BOM. Because the BOM sits at the very start of the first line, it also defeats the two line-level checks (`skip_docstart` and `comment_prefix`), which compare the beginning of a line, so the first marker/comment line is turned into a data row:
```p... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8853/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8853/timeline | null | 0 | true | [] | null | 2,026 | 9 | 460 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8852 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8852/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8852/comments | https://api.github.com/repos/huggingface/datasets/issues/8852/events | https://github.com/huggingface/datasets/pull/8852 | 5,918,068,071 | PR_kwDODunzps8AAAABJQUfIQ | 8,852 | Drop a leading UTF-8 BOM in the text loader | [] | open | false | [] | 2026-10-19T04:56:25Z | 2026-10-19T09:13:23Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8852.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8852",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8852.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8852"
} | ### What was wrong
`load_dataset("text", ...)` keeps a leading UTF-8 BOM as part of the data:
```python
import codecs
import os
import tempfile
from datasets import load_dataset
path = os.path.join(tempfile.mkdtemp(), "corpus.txt")
with open(path, "wb") as f:
f.write(codecs.BOM_UTF8 + b"hello\nworld\n")
print(... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8852/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8852/timeline | null | 0 | true | [] | null | 2,026 | 9 | 458 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8851 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8851/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8851/comments | https://api.github.com/repos/huggingface/datasets/issues/8851/events | https://github.com/huggingface/datasets/pull/8851 | 5,910,819,059 | PR_kwDODunzps8AAAABJKqRaA | 8,851 | Fix WebDataset feature inference with empty TAR shards | [] | open | false | [] | 2026-10-18T14:56:37Z | 2026-10-22T09:15:10Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8851.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8851",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8851.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8851"
} | null | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8851/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8851/timeline | null | 0 | true | [] | null | 2,026 | 8 | 0 | false | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8850 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8850/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8850/comments | https://api.github.com/repos/huggingface/datasets/issues/8850/events | https://github.com/huggingface/datasets/pull/8850 | 5,910,024,252 | PR_kwDODunzps8AAAABJKDDOg | 8,850 | Apply cast and cast_column below threaded decoding in IterableDataset | [] | open | false | [] | 2026-10-18T13:29:49Z | 2026-10-18T13:29:49Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8850.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8850",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8850.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8850"
} | Since #8791, streaming a media dataset from the Hub decodes it in worker threads by default, through the `map` step added by `IterableDataset.decode(num_threads=...)`. `cast` and `cast_column` only replaced `info.features` on top of that step, so they had no effect on decoding:
* `cast_column("audio", Audio(sampling_r... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8850/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8850/timeline | null | 0 | true | [] | null | 2,026 | 9 | 254 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8849 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8849/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8849/comments | https://api.github.com/repos/huggingface/datasets/issues/8849/events | https://github.com/huggingface/datasets/issues/8849 | 5,910,024,035 | I_kwDODunzps8AAAABYEPPYw | 8,849 | cast_column / cast are ignored on streaming datasets decoded in threads (audio sampling_rate, decode=False) | [] | open | false | [] | 2026-10-18T13:29:47Z | 2026-10-18T13:29:47Z | null | NONE | [] | null | null | null | null | ### Describe the bug
Since #8791, `load_dataset(..., streaming=True)` decodes media in worker threads by default when the dataset is on the Hub. The threaded decoding is a `map` step created by `IterableDataset.decode(num_threads=...)`, so the examples coming out of the dataset are already decoded.
`IterableDataset.c... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8849/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8849/timeline | null | 0 | false | [] | null | 2,026 | 14 | 373 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8848 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8848/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8848/comments | https://api.github.com/repos/huggingface/datasets/issues/8848/events | https://github.com/huggingface/datasets/pull/8848 | 5,907,459,842 | PR_kwDODunzps8AAAABJIFkzg | 8,848 | Read string_to_dict fields by group name instead of position | [] | open | false | [] | 2026-10-18T07:59:07Z | 2026-10-18T07:59:07Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8848.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8848",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8848.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8848"
} | `string_to_dict` turns every `{field}` in the pattern into a named group `(?P<_field>.+)` and then reads the match with `result.groups()`, which is positional. Any other capture group in the pattern shifts the values, so `string_to_dict("train/alice", "(train|test)/{name}")` returns `{'name': 'train'}` instead of `{'na... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8848/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8848/timeline | null | 0 | true | [] | null | 2,026 | 9 | 237 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8847 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8847/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8847/comments | https://api.github.com/repos/huggingface/datasets/issues/8847/events | https://github.com/huggingface/datasets/pull/8847 | 5,905,750,493 | PR_kwDODunzps8AAAABJGyEnQ | 8,847 | Hide GPUs before spawning TensorFlow data workers | [] | open | false | [] | 2026-10-18T03:26:59Z | 2026-10-18T03:30:11Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8847.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8847",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8847.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8847"
} | `to_tf_dataset(num_workers=...)` can import a user's collator and initialize TensorFlow before `worker_loop` hides its GPUs. Hide GPUs in the spawn environment so initialization already uses the CPU, then restore the parent environment. Fixes #6948.
Six new cases cover imported collators with eager TensorFlow initia... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8847/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8847/timeline | null | 0 | true | [] | null | 2,026 | 7 | 71 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8846 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8846/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8846/comments | https://api.github.com/repos/huggingface/datasets/issues/8846/events | https://github.com/huggingface/datasets/pull/8846 | 5,905,747,984 | PR_kwDODunzps8AAAABJGx8nQ | 8,846 | Restore packaged type information and public re-exports | [] | open | false | [] | 2026-10-18T03:26:32Z | 2026-10-18T03:29:34Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8846.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8846",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8846.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8846"
} | Restores packaged inline type information. The `py.typed` marker is included in the wheel and source archive and explicit aliases mark the existing public re exports. Annotating the `Features` constructor's variadic arguments as `Any` also lets ordinary dictionary inputs type check. Fixes #7390.
Checked an installed... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8846/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8846/timeline | null | 0 | true | [] | null | 2,026 | 7 | 81 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8845 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8845/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8845/comments | https://api.github.com/repos/huggingface/datasets/issues/8845/events | https://github.com/huggingface/datasets/pull/8845 | 5,905,745,577 | PR_kwDODunzps8AAAABJGx1Jw | 8,845 | Copy existing tensors without a PyTorch warning | [] | open | false | [] | 2026-10-18T03:26:07Z | 2026-10-18T03:28:57Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8845.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8845",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8845.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8845"
} | Streaming `with_format("torch").map(...)` warns for every row when a transform returns tensors. Copy existing tensors directly while keeping dtype/device overrides, independent storage and leaf gradient settings. Fixes #7088.
Added 18 regression cases, including batched maps. Formatting checks: 59 passed, 12 skipped... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8845/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8845/timeline | null | 0 | true | [] | null | 2,026 | 7 | 58 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8844 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8844/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8844/comments | https://api.github.com/repos/huggingface/datasets/issues/8844/events | https://github.com/huggingface/datasets/pull/8844 | 5,905,574,467 | PR_kwDODunzps8AAAABJGpeMw | 8,844 | Warn once when resuming a shuffled dataset | [] | open | false | [
"I think the issue is that it's logged once per dataloader worker, maybe we should have `_is_main_process` from `IterableDataset` in examples iterables too"
] | 2026-10-18T02:57:43Z | 2026-10-19T17:32:35Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8844.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8844",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8844.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8844"
} | Repeated checkpoint restores can print the same shuffle-buffer warning thousands of times. This shows it once per process, including when several datasets are restored. Fixes #7444.
Added four cases covering repeated restores and initial checkpoints, with checks on the returned rows. The iterable suite has 655 passes ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8844/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8844/timeline | null | 1 | true | [] | null | 2,026 | 7 | 72 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8843 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8843/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8843/comments | https://api.github.com/repos/huggingface/datasets/issues/8843/events | https://github.com/huggingface/datasets/pull/8843 | 5,905,519,061 | PR_kwDODunzps8AAAABJGmxQg | 8,843 | Fix mask resizing in the semantic segmentation guide | [] | open | false | [] | 2026-10-18T02:48:19Z | 2026-10-18T02:49:15Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8843.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8843",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8843.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8843"
} | The Torchvision example now resizes masks with nearest neighbor interpolation so it doesn't create new class labels. I also removed the old output image, which shows the incorrect mask boundaries. Fixes #6865.
Checked the updated example with upscaling and downscaling, for single examples and batches. Class labels a... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8843/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8843/timeline | null | 0 | true | [] | null | 2,026 | 8 | 53 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8842 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8842/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8842/comments | https://api.github.com/repos/huggingface/datasets/issues/8842/events | https://github.com/huggingface/datasets/pull/8842 | 5,905,464,748 | PR_kwDODunzps8AAAABJGkDSA | 8,842 | Include the config name when no data files are found | [] | open | false | [] | 2026-10-18T02:40:09Z | 2026-10-18T02:42:13Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8842.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8842",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8842.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8842"
} | When the first config in a README has no matching data files, the error now includes its name and the dataset path. This makes it easier to find the broken entry. Added tests for README and standalone YAML configs, Hub and bucket loaders and explicit file overrides.
Fixes #6819.
Tests: 302 local tests passed. lin... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8842/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8842/timeline | null | 0 | true | [] | null | 2,026 | 10 | 64 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8841 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8841/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8841/comments | https://api.github.com/repos/huggingface/datasets/issues/8841/events | https://github.com/huggingface/datasets/pull/8841 | 5,905,383,660 | PR_kwDODunzps8AAAABJGf6vQ | 8,841 | Handle UTF-8 BOMs in every JSON input file | [] | open | false | [] | 2026-10-18T02:28:30Z | 2026-10-18T02:29:37Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8841.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8841",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8841.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8841"
} | Follow up to #8243. Strip the UTF-8 BOM at the start of each JSON input file. Multifile loads with inferred or explicit `List(Json())` columns otherwise fail when a later file has a BOM.
Adds eight regression cases covering inferred/explicit features and regular/streaming loads, including multiple batches and BOM ch... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8841/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8841/timeline | null | 0 | true | [] | null | 2,026 | 8 | 75 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8840 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8840/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8840/comments | https://api.github.com/repos/huggingface/datasets/issues/8840/events | https://github.com/huggingface/datasets/pull/8840 | 5,905,312,433 | PR_kwDODunzps8AAAABJGcXiw | 8,840 | Fix indices buffer loading in Dataset.from_buffer | [] | closed | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8840). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-18T02:16:59Z | 2026-10-19T17:53:36Z | 2026-10-19T17:53:36Z | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8840.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8840",
"merged_at": "2026-10-19T17:53:36Z",
"patch_url": "https://github.com/huggingface/datasets/pull/8840.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/... | `Dataset.from_buffer` reads the data buffer twice when `indices_buffer` is provided. this raises a type error for string columns and can silently return the wrong rows for uint64 data. Read the supplied indices buffer instead, with 8 regression cases covering empty, reordered and repeated indices, plus the no index cas... | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8840/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8840/timeline | null | 1 | true | [] | 1.65 | 2,026 | 6 | 89 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8839 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8839/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8839/comments | https://api.github.com/repos/huggingface/datasets/issues/8839/events | https://github.com/huggingface/datasets/pull/8839 | 5,904,302,121 | PR_kwDODunzps8AAAABJFrhjA | 8,839 | Create the table before the workers append in to_sql with num_proc | [] | open | false | [
"Checked this against the race in #8838: running batch 0 in the parent before the pool starts means the table exists (with `if_exists` applied once) before any worker appends, and the empty-dataset and single-batch paths still behave. The test discriminates the right way too: the sleep on offset 0 only matters with... | 2026-10-17T23:42:13Z | 2026-10-19T10:15:53Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8839.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8839",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8839.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8839"
} | `to_sql` with `num_proc > 1` hands every batch to the pool at once. The batch at offset 0 runs `DataFrame.to_sql` with `if_exists` as given (`"fail"` by default) and is meant to create the table; the later batches run with `if_exists="append"`, which also creates the table when it is missing. Whenever a later batch rea... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8839/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8839/timeline | null | 3 | true | [] | null | 2,026 | 11 | 303 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8838 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8838/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8838/comments | https://api.github.com/repos/huggingface/datasets/issues/8838/events | https://github.com/huggingface/datasets/issues/8838 | 5,904,301,750 | I_kwDODunzps8AAAABX-x-tg | 8,838 | to_sql with num_proc > 1 fails with 'table already exists' when a later batch reaches the database first | [] | open | false | [] | 2026-10-17T23:42:10Z | 2026-10-17T23:42:10Z | null | NONE | [] | null | null | null | null | ### Describe the bug
`Dataset.to_sql(name, con, num_proc=N)` with more than one batch fails with `table already exists` (sqlite: `(sqlite3.OperationalError) table t already exists`, or pandas' `ValueError: Table 't' already exists.`), and whether it fails depends on which worker reaches the database first.
`SqlDatase... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8838/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8838/timeline | null | 0 | false | [] | null | 2,026 | 18 | 265 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8837 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8837/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8837/comments | https://api.github.com/repos/huggingface/datasets/issues/8837/events | https://github.com/huggingface/datasets/pull/8837 | 5,904,286,825 | PR_kwDODunzps8AAAABJFqyBw | 8,837 | Refuse an iterable of bools as a row index instead of reading it as indices 0 and 1 | [] | open | false | [
"The first version only looked at `dtype.kind`, so an object-dtype array of bools and a torch-style bool tensor still slipped through as rows 0 and 1. The check now covers those, Arrow boolean arrays, and `Dataset.select`, which had the same silent path for a numpy or Arrow mask. Each case has a test that fails on ... | 2026-10-17T23:39:59Z | 2026-10-18T08:06:25Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8837.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8837",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8837.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8837"
} | `Dataset.__getitem__` reads a list, tuple, numpy array or pandas Series of booleans as row indices: `_query_table` turns the key into integers with `np.fromiter(key, np.int64)`, so a mask becomes the rows 0 and 1, and the result is silently wrong.
```python
>>> ds = Dataset.from_dict({"a": [10, 20, 30, 40, 50]})
>>> d... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8837/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8837/timeline | null | 1 | true | [] | null | 2,026 | 18 | 524 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8836 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8836/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8836/comments | https://api.github.com/repos/huggingface/datasets/issues/8836/events | https://github.com/huggingface/datasets/issues/8836 | 5,904,286,347 | I_kwDODunzps8AAAABX-xCiw | 8,836 | Indexing a Dataset with a list or array of bools silently returns rows 0 and 1 | [] | open | false | [] | 2026-10-17T23:39:55Z | 2026-10-17T23:39:55Z | null | NONE | [] | null | null | null | null | ### Describe the bug
`Dataset.__getitem__` accepts a list, tuple, numpy array or pandas Series of booleans and reads it as integer row indices: every `True` becomes row 1 and every `False` row 0. No error is raised, and the result has the length of the mask, so it looks like a valid selection. The docstring says rows ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8836/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8836/timeline | null | 0 | false | [] | null | 2,026 | 16 | 258 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8835 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8835/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8835/comments | https://api.github.com/repos/huggingface/datasets/issues/8835/events | https://github.com/huggingface/datasets/pull/8835 | 5,904,197,347 | PR_kwDODunzps8AAAABJFmVbQ | 8,835 | Write the encoding passed to to_csv | [] | open | false | [
"Two small follow-ups: an empty dataset wrote a two-byte UTF-16 byte order mark where main wrote nothing, so the encoder's final flush now runs only after a batch was written, and the test checks the raw bytes for exactly one mark instead of relying on the round trip alone. #8834 touches the same lines, so whicheve... | 2026-10-17T23:27:22Z | 2026-10-19T11:23:29Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8835.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8835",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8835.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8835"
} | `Dataset.to_csv(path, encoding="utf-16")` wrote UTF-8. `CsvDatasetWriter._batch_csv` hands the batch to pandas with `path_or_buf=None`, so pandas returns a `str` and never applies `encoding`; the writer then encoded that `str` with a hard-coded `"utf-8"`. The docstring points at `pandas.DataFrame.to_csv` for the keywor... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8835/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8835/timeline | null | 1 | true | [] | null | 2,026 | 6 | 357 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8834 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8834/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8834/comments | https://api.github.com/repos/huggingface/datasets/issues/8834/events | https://github.com/huggingface/datasets/pull/8834 | 5,904,197,099 | PR_kwDODunzps8AAAABJFmUog | 8,834 | Number the index across batches in to_csv and to_sql | [] | open | false | [
"The RangeIndex renumbering is the right fix and covers both writers. One case worth a pin: a dataset that went through `select()` carries `_indices`, and `offset` is then a position inside the sliced view, so the written index numbers the slice 0..len rather than the original rows. That matches the natural \"row n... | 2026-10-17T23:27:20Z | 2026-10-19T08:02:54Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8834.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8834",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8834.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8834"
} | `to_csv(index=True)` and `to_sql(index=True)` wrote an index column that restarts at 0 in every batch: `CsvDatasetWriter._batch_csv` and `SqlDatasetWriter._batch_sql` convert each batch with `batch.to_pandas()`, which gives the batch a fresh `RangeIndex` from 0, and pandas writes that. With the default `batch_size` the... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8834/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8834/timeline | null | 2 | true | [] | null | 2,026 | 9 | 322 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8833 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8833/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8833/comments | https://api.github.com/repos/huggingface/datasets/issues/8833/events | https://github.com/huggingface/datasets/pull/8833 | 5,904,196,838 | PR_kwDODunzps8AAAABJFmT0A | 8,833 | Cast dictionary-encoded storage such as pandas Categorical columns in ClassLabel | [] | open | false | [
"A second look at my own change turned up a regression: the first version sent every non-integer dictionary down the label-name path, so a float dictionary that `array_cast` converted on main started raising. The name path is now gated on a string value type and a test pins the float case. I also noted #8688 in the... | 2026-10-17T23:27:18Z | 2026-10-25T13:07:30Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8833.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8833",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8833.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8833"
} | `ClassLabel.cast_storage` handles `pa.StringArray` and `pa.IntegerArray` and hands anything else to `array_cast(storage, int64)`. A pandas `Categorical` column arrives as a `pa.DictionaryArray` (`dictionary<values=large_string, indices=int8>`), so Arrow tries to parse the category names as integers:
```
ArrowInvalid: ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8833/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8833/timeline | null | 2 | true | [] | null | 2,026 | 10 | 496 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8832 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8832/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8832/comments | https://api.github.com/repos/huggingface/datasets/issues/8832/events | https://github.com/huggingface/datasets/issues/8832 | 5,904,196,259 | I_kwDODunzps8AAAABX-riow | 8,832 | to_csv ignores the encoding argument and always writes UTF-8 | [] | open | false | [] | 2026-10-17T23:27:13Z | 2026-10-17T23:27:13Z | null | NONE | [] | null | null | null | null | ### Describe the bug
`Dataset.to_csv(path, encoding="utf-16")` writes UTF-8. The docstring says the extra keyword arguments go to `pandas.DataFrame.to_csv`, which documents `encoding` (and `errors`), and both are accepted without complaint, but the file is always UTF-8.
The cause is in `datasets/io/csv.py`: `CsvDatas... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8832/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8832/timeline | null | 0 | false | [] | null | 2,026 | 9 | 189 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8831 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8831/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8831/comments | https://api.github.com/repos/huggingface/datasets/issues/8831/events | https://github.com/huggingface/datasets/issues/8831 | 5,904,196,115 | I_kwDODunzps8AAAABX-riEw | 8,831 | to_csv and to_sql with index=True restart the index at 0 in every batch | [] | open | false | [] | 2026-10-17T23:27:11Z | 2026-10-17T23:27:11Z | null | NONE | [] | null | null | null | null | ### Describe the bug
`Dataset.to_csv(..., index=True)` and `Dataset.to_sql(..., index=True)` write an index column whose values restart at 0 at every batch boundary. The docstrings of both methods say "If you would like to write the index, pass `index=True` and also set a name for the index column by passing `index_la... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8831/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8831/timeline | null | 0 | false | [] | null | 2,026 | 13 | 268 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8830 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8830/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8830/comments | https://api.github.com/repos/huggingface/datasets/issues/8830/events | https://github.com/huggingface/datasets/issues/8830 | 5,904,196,010 | I_kwDODunzps8AAAABX-rhqg | 8,830 | ClassLabel.cast_storage fails on a pandas Categorical column (dictionary-encoded Arrow storage) | [] | open | false | [
"Related: #8681 reports the plain-string side of the same pandas 3 change (`str` columns arrive as `large_string`, which `ClassLabel.cast_storage` does not map either), with the fix in #8688. On current main both shapes fail with the same `ArrowInvalid`; #8833 fixes the `Categorical` (dictionary) shape and #8688 th... | 2026-10-17T23:27:11Z | 2026-10-18T07:56:15Z | null | NONE | [] | null | null | null | null | ### Describe the bug
`Dataset.from_pandas(df, features=...)` fails when the label column of the DataFrame is a pandas `Categorical` and the feature is a `ClassLabel`. pandas hands the column to Arrow as a `DictionaryArray` (`dictionary<values=large_string, indices=int8>`), and `ClassLabel.cast_storage` only knows `pa.... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8830/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8830/timeline | null | 1 | false | [] | null | 2,026 | 10 | 239 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8829 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8829/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8829/comments | https://api.github.com/repos/huggingface/datasets/issues/8829/events | https://github.com/huggingface/datasets/pull/8829 | 5,900,342,930 | PR_kwDODunzps8AAAABJCneLw | 8,829 | docs: clarify checksum recording defaults | [] | open | false | [] | 2026-10-17T16:07:16Z | 2026-10-17T16:07:16Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8829.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8829",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8829.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8829"
} | Fixes #8590
Clarifies that checksum recording has been disabled by default since Datasets 4.8.5, and that `verification_mode="all_checks"` does not enable it. The integrity guide now explains when checksum comparison runs, shows how to pin a Hub dataset commit for reproducibility, and distinguishes that from verifying... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8829/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8829/timeline | null | 0 | true | [] | null | 2,026 | 5 | 75 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8828 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8828/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8828/comments | https://api.github.com/repos/huggingface/datasets/issues/8828/events | https://github.com/huggingface/datasets/pull/8828 | 5,890,424,494 | PR_kwDODunzps8AAAABI67Uqw | 8,828 | Fix save_to_disk failing when a batch holds more than 2GB of image or audio files | [] | open | false | [] | 2026-10-16T20:02:59Z | 2026-10-16T20:02:59Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8828.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8828",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8828.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8828"
} | Fixes #5717
Supersedes [#8141](https://github.com/huggingface/datasets/pull/8141), which @kasperg3 [agreed to hand over](https://github.com/huggingface/datasets/pull/8141#issuecomment-6208799638). The first commit is kasperg3's, minus a `to_dict` line that [a later commit](https://github.com/huggingface/datasets/pull/... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8828/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8828/timeline | null | 0 | true | [] | null | 2,026 | 15 | 1,438 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8827 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8827/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8827/comments | https://api.github.com/repos/huggingface/datasets/issues/8827/events | https://github.com/huggingface/datasets/pull/8827 | 5,886,472,877 | PR_kwDODunzps8AAAABI3wpGA | 8,827 | Release closed streaming batch iterators | [] | open | false | [
"`use_threads=False` can cause important slow downs when it comes to streaming parquet datasets, I would look for a better fix. I'm pretty sure you will see a difference if you run a benchmark",
"thanks for that 🙇 converting back to draft until i know more here\r\n\r\nI compared upstream with this branch using a... | 2026-10-16T15:08:05Z | 2026-10-21T22:37:55Z | null | NONE | [] | null | null | true | {
"diff_url": "https://github.com/huggingface/datasets/pull/8827.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8827",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8827.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8827"
} | Related to #8793.
`IterableDataset.iter(...).close()` can leave nested generators alive while the caller keeps the closed iterator. On Python 3.12 this leaves shuffled streaming sources and their files open until that iterator is deleted. Return an iterator that closes and releases its owned batch generator, so early ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8827/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8827/timeline | null | 4 | true | [] | null | 2,026 | 5 | 391 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8826 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8826/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8826/comments | https://api.github.com/repos/huggingface/datasets/issues/8826/events | https://github.com/huggingface/datasets/pull/8826 | 5,885,601,100 | PR_kwDODunzps8AAAABI3EBnA | 8,826 | Speed up Dataset iteration ~7x for the default Python format | [] | open | false | [] | 2026-10-16T14:09:45Z | 2026-10-16T14:09:45Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8826.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8826",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8826.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8826"
} | `for row in ds` slices a one-row table and runs `format_table` on it for every row. With the default Python format that means `to_pydict()` on a 1-row table, `_unnest`, and `decode_example`, so most of the time goes to per-row overhead rather than conversion. Datasets with an indices mapping (after `shuffle` or `select... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8826/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8826/timeline | null | 0 | true | [] | null | 2,026 | 10 | 324 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8825 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8825/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8825/comments | https://api.github.com/repos/huggingface/datasets/issues/8825/events | https://github.com/huggingface/datasets/pull/8825 | 5,884,732,891 | PR_kwDODunzps8AAAABI2Xovw | 8,825 | bump fsspec and remove aiohttp dep | [] | closed | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8825). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update."
] | 2026-10-16T13:10:13Z | 2026-10-16T13:20:07Z | 2026-10-16T13:20:03Z | MEMBER | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8825.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8825",
"merged_at": "2026-10-16T13:20:03Z",
"patch_url": "https://github.com/huggingface/datasets/pull/8825.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/... | null | {
"avatar_url": "https://avatars.githubusercontent.com/u/42851186?v=4",
"events_url": "https://api.github.com/users/lhoestq/events{/privacy}",
"followers_url": "https://api.github.com/users/lhoestq/followers",
"following_url": "https://api.github.com/users/lhoestq/following{/other_user}",
"gists_url": "https:... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8825/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8825/timeline | null | 1 | true | [] | 0.007 | 2,026 | 6 | 0 | false | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8824 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8824/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8824/comments | https://api.github.com/repos/huggingface/datasets/issues/8824/events | https://github.com/huggingface/datasets/pull/8824 | 5,882,312,053 | PR_kwDODunzps8AAAABI0a4pg | 8,824 | Fix duplicated article in load.py docstring | [] | open | false | [] | 2026-10-16T10:10:42Z | 2026-10-16T10:10:42Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8824.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8824",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8824.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8824"
} | ## What
Fixed the duplicated article 'a a Storage Bucket' in a docstring in load.py.
## Why
A small, objectively-verifiable improvement (typo) found during a
daily open-source maintenance pass (2026-10-16).
## How verified
Found with a grep for repeated words. Checked the diff: 1 line changed in 1 file. Committed on ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8824/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8824/timeline | null | 0 | true | [] | null | 2,026 | 6 | 53 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8823 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8823/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8823/comments | https://api.github.com/repos/huggingface/datasets/issues/8823/events | https://github.com/huggingface/datasets/pull/8823 | 5,878,560,687 | PR_kwDODunzps8AAAABIxbtfQ | 8,823 | Accept path-like objects in `save_to_disk` / `load_from_disk` on older fsspec | [] | open | false | [] | 2026-10-16T04:54:40Z | 2026-10-16T04:54:40Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8823.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8823",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8823.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8823"
} | `Dataset.save_to_disk`, `Dataset.load_from_disk`, `DatasetDict.save_to_disk`, `DatasetDict.load_from_disk` and `datasets.load_from_disk` all document their path argument as `path-like`, but they pass it straight to `fsspec.core.url_to_fs`, which only started stringifying `os.PathLike` inputs in fsspec 2024.6. On any fs... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8823/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8823/timeline | null | 0 | true | [] | null | 2,026 | 10 | 186 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8822 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8822/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8822/comments | https://api.github.com/repos/huggingface/datasets/issues/8822/events | https://github.com/huggingface/datasets/pull/8822 | 5,873,892,807 | PR_kwDODunzps8AAAABItwR3g | 8,822 | Fix streaming raising FileNotFoundError for existing files when the Hub is briefly unavailable | [] | open | false | [
"The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8822). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.",
"where does the FileNotFoundError come from ? I see one that is raised when OSError is c... | 2026-10-15T21:03:30Z | 2026-10-19T18:11:41Z | null | CONTRIBUTOR | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8822.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8822",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8822.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8822"
} | Fixes #7440
When the Hub is briefly unavailable (429, 5xx or a timeout), streaming can raise `FileNotFoundError` for a file that exists, and a long training run stops.
```python
from unittest.mock import patch
from huggingface_hub import HfFileSystem
from huggingface_hub.errors import HfHubHTTPError
from hu... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8822/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8822/timeline | null | 3 | true | [] | null | 2,026 | 13 | 619 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8821 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8821/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8821/comments | https://api.github.com/repos/huggingface/datasets/issues/8821/events | https://github.com/huggingface/datasets/pull/8821 | 5,869,027,491 | PR_kwDODunzps8AAAABIp1xqQ | 8,821 | Stop streaming take after the requested input boundary | [] | open | false | [] | 2026-10-15T15:28:45Z | 2026-10-15T15:28:45Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8821.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8821",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8821.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8821"
} | Fully exhausting `dataset.take(n)` currently asks the upstream iterable for one extra example (or Arrow table) before stopping. If that next read raises, a sample containing all requested rows still fails. The extra read can also run unnecessary mapping or I/O.
Stop both iteration paths immediately after yielding the ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8821/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8821/timeline | null | 0 | true | [] | null | 2,026 | 8 | 162 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8820 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8820/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8820/comments | https://api.github.com/repos/huggingface/datasets/issues/8820/events | https://github.com/huggingface/datasets/pull/8820 | 5,868,959,085 | PR_kwDODunzps8AAAABIpyP0w | 8,820 | Preserve column order and metadata when unpickling concatenated datasets | [] | open | false | [] | 2026-10-15T15:24:27Z | 2026-10-15T15:24:27Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8820.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8820",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8820.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8820"
} | Pickling a dataset concatenated along columns can undo `select_columns()` ordering and replace the full feature metadata with metadata from its first source. For example, two memory-mapped datasets concatenated as `a, b` and selected as `b, a` come back from a pickle roundtrip as `a, b`.
Restore the saved schema when ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8820/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8820/timeline | null | 0 | true | [] | null | 2,026 | 9 | 122 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8819 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8819/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8819/comments | https://api.github.com/repos/huggingface/datasets/issues/8819/events | https://github.com/huggingface/datasets/pull/8819 | 5,868,925,089 | PR_kwDODunzps8AAAABIpwgmA | 8,819 | Synchronize feature decoding after in-place dictionary unions | [] | open | false | [] | 2026-10-15T15:22:13Z | 2026-10-15T15:22:13Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8819.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8819",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8819.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8819"
} | `Features |= other` inherits `dict.__ior__`, so it changes the schema without updating the cached decoding flags. Adding a column then raises `KeyError` during decoding. Replacing a `Value("string")` or `Json(decode=False)` column with `Json()` silently leaves JSON payloads encoded as strings.
Apply the existing synch... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8819/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8819/timeline | null | 0 | true | [] | null | 2,026 | 7 | 135 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8818 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8818/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8818/comments | https://api.github.com/repos/huggingface/datasets/issues/8818/events | https://github.com/huggingface/datasets/pull/8818 | 5,868,924,229 | PR_kwDODunzps8AAAABIpwdyQ | 8,818 | Preserve unread values in lazy reverse dictionary unions | [] | open | false | [] | 2026-10-15T15:22:10Z | 2026-10-15T15:22:10Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8818.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8818",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8818.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8818"
} | Using `defaults | example` inside `Dataset.map()` should keep existing example values. For an unread overlapping column, the reverse union currently removes its deferred-formatting marker while retaining its `None` placeholder. Row mapping silently writes null values; batched mapping raises `TypeError`.
Keep the right... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8818/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8818/timeline | null | 0 | true | [] | null | 2,026 | 8 | 139 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8817 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8817/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8817/comments | https://api.github.com/repos/huggingface/datasets/issues/8817/events | https://github.com/huggingface/datasets/pull/8817 | 5,868,923,203 | PR_kwDODunzps8AAAABIpwaXw | 8,817 | Skip empty streams when rebatching the entire dataset | [] | open | false | [] | 2026-10-15T15:22:06Z | 2026-10-15T15:22:06Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8817.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8817",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8817.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8817"
} | Mapping an empty Arrow or pandas stream with `batched=True` and `batch_size=None` (also `0` or a negative value) crashes with `ArrowInvalid: Must pass at least one table` after `take(0)`. If a filter instead emits empty tables, the same path invokes the callback with an empty batch. Positive-size batches already skip b... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8817/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8817/timeline | null | 0 | true | [] | null | 2,026 | 8 | 158 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8816 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8816/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8816/comments | https://api.github.com/repos/huggingface/datasets/issues/8816/events | https://github.com/huggingface/datasets/pull/8816 | 5,868,922,300 | PR_kwDODunzps8AAAABIpwXZw | 8,816 | Preserve selected features through streaming filters | [] | open | false | [] | 2026-10-15T15:22:02Z | 2026-10-15T15:22:02Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8816.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8816",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8816.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8816"
} | After a typed Arrow or pandas map, `select_columns()` removes columns from rows but retains the parent iterable's full feature schema. A subsequent `filter()` reintroduces those removed columns during formatting, then raises `CastError` because they no longer match the dataset's selected schema.
Project `SelectColumns... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8816/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8816/timeline | null | 0 | true | [] | null | 2,026 | 6 | 130 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8815 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8815/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8815/comments | https://api.github.com/repos/huggingface/datasets/issues/8815/events | https://github.com/huggingface/datasets/pull/8815 | 5,868,919,463 | PR_kwDODunzps8AAAABIpwOLA | 8,815 | Accept NumPy integer indices on nested columns | [] | open | false | [] | 2026-10-15T15:21:51Z | 2026-10-15T15:21:51Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8815.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8815",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8815.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8815"
} | Indexing a nested column with a NumPy integer, such as `dataset["metadata"]["score"][np.int64(0)]`, treats the returned row as an iterable and raises `TypeError`. Ordinary dataset columns already accept NumPy integer indices.
Recognize NumPy integers in the nested-column scalar path as well. Regression tests cover sig... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8815/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8815/timeline | null | 0 | true | [] | null | 2,026 | 7 | 124 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8814 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8814/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8814/comments | https://api.github.com/repos/huggingface/datasets/issues/8814/events | https://github.com/huggingface/datasets/pull/8814 | 5,868,821,873 | PR_kwDODunzps8AAAABIprLrw | 8,814 | Honor new_fingerprint when casting non-decoder columns | [] | open | false | [] | 2026-10-15T15:15:44Z | 2026-10-15T15:15:44Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8814.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8814",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8814.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8814"
} | `Dataset.cast_column(..., new_fingerprint="cast-result")` ignores the supplied fingerprint for features without a decoder, including `Value`, `ClassLabel`, and `List`. The returned dataset and its saved `state.json` instead contain the fingerprint produced by the nested `cast()` call. Decoder-backed features already ho... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8814/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8814/timeline | null | 0 | true | [] | null | 2,026 | 6 | 171 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8813 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8813/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8813/comments | https://api.github.com/repos/huggingface/datasets/issues/8813/events | https://github.com/huggingface/datasets/pull/8813 | 5,868,709,247 | PR_kwDODunzps8AAAABIplX1A | 8,813 | Keep streaming Arrow map indices tied to input rows | [] | open | false | [] | 2026-10-15T15:08:59Z | 2026-10-15T15:08:59Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8813.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8813",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8813.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8813"
} | Arrow-formatted batched maps advance `with_indices=True` by the number of output rows when read through the Python row iterator. If a batch is duplicated, filtered, or dropped, later callbacks receive the wrong source indices. For example, concatenating a mapped Arrow stream with a generator-backed stream makes source ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8813/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8813/timeline | null | 0 | true | [] | null | 2,026 | 9 | 144 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8812 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8812/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8812/comments | https://api.github.com/repos/huggingface/datasets/issues/8812/events | https://github.com/huggingface/datasets/pull/8812 | 5,868,686,868 | PR_kwDODunzps8AAAABIpkOrA | 8,812 | Preserve dataset info and split when converting to streaming | [] | open | false | [] | 2026-10-15T15:07:33Z | 2026-10-15T15:07:33Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8812.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8812",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8812.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8812"
} | `Dataset.to_iterable_dataset()` keeps the rows and features but currently drops the split name and all other `DatasetInfo` fields. For example, a dataset with `split="train"`, a description and a license becomes a stream with `split=None` and empty description/license values. This loses attribution and provenance when ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8812/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8812/timeline | null | 0 | true | [] | null | 2,026 | 9 | 171 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8811 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8811/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8811/comments | https://api.github.com/repos/huggingface/datasets/issues/8811/events | https://github.com/huggingface/datasets/pull/8811 | 5,868,620,168 | PR_kwDODunzps8AAAABIpg0Hg | 8,811 | Preserve headers and document structure when exporting empty datasets | [] | open | false | [] | 2026-10-15T15:03:23Z | 2026-10-15T15:03:23Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8811.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8811",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8811.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8811"
} | An empty `Dataset` currently writes zero bytes for both CSV and JSON exports. That drops the CSV header and makes `to_json(..., lines=False)` produce an invalid JSON document. This is reachable when filtering removes every record or a partition has no rows.
Serialize one empty batch so CSV retains the requested header... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8811/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8811/timeline | null | 0 | true | [] | null | 2,026 | 9 | 217 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8810 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8810/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8810/comments | https://api.github.com/repos/huggingface/datasets/issues/8810/events | https://github.com/huggingface/datasets/pull/8810 | 5,862,055,527 | PR_kwDODunzps8AAAABIkOjjQ | 8,810 | Restore legacy unsorted dict hashing for the 2.15 cache lookup | [] | open | false | [] | 2026-10-15T07:32:28Z | 2026-10-15T09:10:37Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8810.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8810",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8810.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8810"
} | Fixes #8410
### Problem
`DatasetBuilder._check_legacy_cache2` patches `Pickler._legacy_no_dict_keys_sorting = True` to recompute the `config_id` that datasets 2.15 wrote, since 2.15 hashed dicts without sorting their keys. Since 4.4.0 the flag does nothing, so the lookup computes the current (sorted) hash and misses ... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8810/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8810/timeline | null | 0 | true | [] | null | 2,026 | 10 | 226 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8809 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8809/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8809/comments | https://api.github.com/repos/huggingface/datasets/issues/8809/events | https://github.com/huggingface/datasets/pull/8809 | 5,860,994,951 | PR_kwDODunzps8AAAABIjY6oA | 8,809 | Fix CSV column_names overrides for configured datasets | [] | open | false | [] | 2026-10-15T05:59:34Z | 2026-10-15T05:59:34Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8809.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8809",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8809.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8809"
} | Fixes #7077.
When loading a CSV dataset directory, `column_names` can be ignored and the first data row used as the header. It is also ignored when overriding column names from README configuration. Loading with the explicit `"csv"` builder already works.
Normalize incoming `column_names` to `names` before the CSV bu... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8809/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8809/timeline | null | 0 | true | [] | null | 2,026 | 7 | 164 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8808 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8808/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8808/comments | https://api.github.com/repos/huggingface/datasets/issues/8808/events | https://github.com/huggingface/datasets/pull/8808 | 5,858,807,953 | PR_kwDODunzps8AAAABIhqitQ | 8,808 | Always release open Parquet fragments on early stream stop (pyarrow >= 25) | [] | closed | false | [
"Closing this. The issue reporter already noted that an explicit `gc.collect()` before exit doesn't prevent the crash, the upstream Arrow bug (#45214) was fixed, and I couldn't reproduce or verify the pyarrow 25 crash with this change, so the version gate removal isn't justified. Unassigning myself from #8793 so so... | 2026-10-15T02:06:07Z | 2026-10-15T02:09:51Z | 2026-10-15T02:09:51Z | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8808.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8808",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8808.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8808"
} | Fixes #8793
The workaround for apache/arrow#45214 in the Parquet builder (`del parquet_fragment` + `gc.collect()` when a stream stops before the fragment is fully read) was gated on `datasets.config.PYARROW_VERSION <= 24.0.0`. The issue reports that the fatal `PyGILState_Release` at interpreter exit still happens with... | {
"avatar_url": "https://avatars.githubusercontent.com/u/241105014?v=4",
"events_url": "https://api.github.com/users/jipeng6036-del/events{/privacy}",
"followers_url": "https://api.github.com/users/jipeng6036-del/followers",
"following_url": "https://api.github.com/users/jipeng6036-del/following{/other_user}",
... | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8808/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8808/timeline | null | 1 | true | [] | 0.003 | 2,026 | 12 | 137 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
https://api.github.com/repos/huggingface/datasets/issues/8807 | https://api.github.com/repos/huggingface/datasets | https://api.github.com/repos/huggingface/datasets/issues/8807/labels{/name} | https://api.github.com/repos/huggingface/datasets/issues/8807/comments | https://api.github.com/repos/huggingface/datasets/issues/8807/events | https://github.com/huggingface/datasets/pull/8807 | 5,858,786,339 | PR_kwDODunzps8AAAABIhpcHQ | 8,807 | Support remote streaming reads for TsFile | [] | open | false | [
"LGTB, @lhoestq Would you mind taking a look at this PR when you get a chance?"
] | 2026-10-15T02:03:57Z | 2026-10-16T03:43:47Z | null | NONE | [] | null | null | false | {
"diff_url": "https://github.com/huggingface/datasets/pull/8807.diff",
"html_url": "https://github.com/huggingface/datasets/pull/8807",
"merged_at": null,
"patch_url": "https://github.com/huggingface/datasets/pull/8807.patch",
"url": "https://api.github.com/repos/huggingface/datasets/pulls/8807"
} | Streaming a remote TsFile currently validates its header through `xopen`, then passes the URL string to `TsFileReader`, whose path-based constructor opens local files. Keep the fsspec seekable binary source alive and pass it to the file-object reader introduced in TsFile 2.5.0. Local files continue using the native pat... | null | {
"+1": 0,
"-1": 0,
"confused": 0,
"eyes": 0,
"heart": 0,
"hooray": 0,
"laugh": 0,
"rocket": 0,
"total_count": 0,
"url": "https://api.github.com/repos/huggingface/datasets/issues/8807/reactions"
} | https://api.github.com/repos/huggingface/datasets/issues/8807/timeline | null | 1 | true | [] | null | 2,026 | 6 | 424 | true | huggingface/datasets | 2026-10-30T06:37:26+00:00 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.