html_url large_stringlengths 49 49 | title large_stringlengths 4 162 | comments large_stringlengths 66 9.88k | body large_stringlengths 5 12.2k ⌀ | text stringlengths 105 17.9k | embeddings listlengths 768 768 |
|---|---|---|---|---|---|
https://github.com/huggingface/datasets/pull/8598 | [httpx migration] Import httpx from huggingface_hub | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8598). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | Direct httpx imports will break when huggingface_hub migrates to httpx2. This updates them to use the compatibility re-export introduced in https://github.com/huggingface/huggingface_hub/pull/4803, following the migration plan in https://github.com/huggingface/huggingface_hub/issues/4802. Goal is to make datasets compa... | [httpx migration] Import httpx from huggingface_hub
Direct httpx imports will break when huggingface_hub migrates to httpx2. This updates them to use the compatibility re-export introduced in https://github.com/huggingface/huggingface_hub/pull/4803, following the migration plan in https://github.com/huggingface/hugging... | [
0.09664132446050644,
-0.3601100444793701,
-0.06200414523482323,
-0.07477857917547226,
0.1955280900001526,
0.09222032874822617,
0.13134528696537018,
0.31443706154823303,
0.20349138975143433,
-0.009685147553682327,
-0.3508707582950592,
-0.30892273783683777,
-0.03389598801732063,
0.2073323875... |
https://github.com/huggingface/datasets/pull/8598 | [httpx migration] Import httpx from huggingface_hub | Failing CI seems unrelated
```
FAILED tests/test_hub.py::test_delete_from_hub - huggingface_hub.errors.HfHubHTTPError: Server error '502 Bad Gateway' for url 'https://hub-ci.huggingface.co/api/datasets/__DUMMY_DATASETS_USER__/test-dataset-14a7d7-17890532107233/commit/main'
For more information check: https://deve... | Direct httpx imports will break when huggingface_hub migrates to httpx2. This updates them to use the compatibility re-export introduced in https://github.com/huggingface/huggingface_hub/pull/4803, following the migration plan in https://github.com/huggingface/huggingface_hub/issues/4802. Goal is to make datasets compa... | [httpx migration] Import httpx from huggingface_hub
Direct httpx imports will break when huggingface_hub migrates to httpx2. This updates them to use the compatibility re-export introduced in https://github.com/huggingface/huggingface_hub/pull/4803, following the migration plan in https://github.com/huggingface/hugging... | [
-0.06642121821641922,
-0.18476155400276184,
-0.04284598305821419,
0.018656481057405472,
0.22505217790603638,
0.0760914608836174,
0.10948563367128372,
0.31411659717559814,
-0.03423465043306351,
-0.1486196368932724,
-0.31741487979888916,
-0.1046658456325531,
0.1123252809047699,
0.44926166534... |
https://github.com/huggingface/datasets/pull/8598 | [httpx migration] Import httpx from huggingface_hub | cool ! I guess we need to bump the minimal version from 0.25.0 to 1.30.0 (both in setup.py and in the "minimum version" CI) | Direct httpx imports will break when huggingface_hub migrates to httpx2. This updates them to use the compatibility re-export introduced in https://github.com/huggingface/huggingface_hub/pull/4803, following the migration plan in https://github.com/huggingface/huggingface_hub/issues/4802. Goal is to make datasets compa... | [httpx migration] Import httpx from huggingface_hub
Direct httpx imports will break when huggingface_hub migrates to httpx2. This updates them to use the compatibility re-export introduced in https://github.com/huggingface/huggingface_hub/pull/4803, following the migration plan in https://github.com/huggingface/hugging... | [
-0.16685202717781067,
-0.32649412751197815,
-0.11599967628717422,
0.006541070993989706,
0.2499535083770752,
0.04923224449157715,
0.05442256107926369,
0.524398148059845,
0.1537390798330307,
-0.0491214320063591,
-0.2734866142272949,
-0.17101657390594482,
-0.05267921835184097,
0.2569395303726... |
https://github.com/huggingface/datasets/pull/8594 | Forward max_buffer_input_shards from IterableDatasetDict.shuffle | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8594). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | ## What breaks
`load_dataset(..., streaming=True)` without a split returns an `IterableDatasetDict`, and shuffling it cannot reach the shard-interleaving control:
```python
ds = load_dataset("HuggingFaceFW/fineweb", streaming=True)
ds = ds.shuffle(seed=42, buffer_size=10_000, max_buffer_input_shards=4)
# TypeError: I... | Forward max_buffer_input_shards from IterableDatasetDict.shuffle
## What breaks
`load_dataset(..., streaming=True)` without a split returns an `IterableDatasetDict`, and shuffling it cannot reach the shard-interleaving control:
```python
ds = load_dataset("HuggingFaceFW/fineweb", streaming=True)
ds = ds.shuffle(seed=... | [
-0.1987999975681305,
-0.5981290936470032,
-0.02844133973121643,
0.08585389703512192,
-0.028091641142964363,
-0.16591666638851166,
0.2601541578769684,
0.13958348333835602,
-0.015885010361671448,
0.31751111149787903,
-0.1113712340593338,
0.08701872080564499,
-0.19547326862812042,
0.191598072... |
https://github.com/huggingface/datasets/pull/8591 | Better error when a column written as null gets values in a later batch | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8591). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | On #2831 the map still fails at HEAD, as `TypeError: Couldn't cast array of type double to null` now rather than `ArrowInvalid`. Your 2021 comment on that issue has the mechanism and both workarounds; the error itself says none of it.
So this does not change what the writer accepts. Before the cast, `ArrowWriter` now ... | Better error when a column written as null gets values in a later batch
On #2831 the map still fails at HEAD, as `TypeError: Couldn't cast array of type double to null` now rather than `ArrowInvalid`. Your 2021 comment on that issue has the mechanism and both workarounds; the error itself says none of it.
So this does... | [
0.034668587148189545,
-0.5024073719978333,
0.0644623190164566,
-0.047000691294670105,
0.21944843232631683,
-0.1634937822818756,
0.22597160935401917,
0.13186460733413696,
0.4030679762363434,
0.18253980576992035,
0.21798600256443024,
0.013955647125840187,
-0.12735795974731445,
0.090035483241... |
https://github.com/huggingface/datasets/pull/8589 | Advance the user-supplied generator in Dataset.shuffle even on cache hits | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8589). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | Fixes #7127.
When a `np.random.Generator` is passed to `Dataset.shuffle`, the indices cache lookup returns before `generator.permutation(...)` is called, so the caller's generator never advances on a cache hit. Because the fingerprint is derived from the generator's state, the next call recomputes the same fingerpri... | Advance the user-supplied generator in Dataset.shuffle even on cache hits
Fixes #7127.
When a `np.random.Generator` is passed to `Dataset.shuffle`, the indices cache lookup returns before `generator.permutation(...)` is called, so the caller's generator never advances on a cache hit. Because the fingerprint is deriv... | [
0.022823724895715714,
-0.06085316091775894,
0.04495702311396599,
0.01178084034472704,
0.28447529673576355,
-0.1268148571252823,
0.43660488724708557,
0.22514212131500244,
-0.020861955359578133,
0.4828891158103943,
-0.0699952021241188,
0.40311765670776367,
-0.1867009699344635,
0.004238705616... |
https://github.com/huggingface/datasets/pull/8585 | Reject a non-callable generator in GeneratorConfig | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8585). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | ### The bug
`GeneratorConfig` checks that `generator` is not `None`, but not that it is callable. A wrong type is only discovered during generation:
```python
Dataset.from_generator([{"a": 1}])
```
```
DatasetGenerationError: An error occurred while generating the dataset
caused by
TypeError: 'list' object is not ... | Reject a non-callable generator in GeneratorConfig
### The bug
`GeneratorConfig` checks that `generator` is not `None`, but not that it is callable. A wrong type is only discovered during generation:
```python
Dataset.from_generator([{"a": 1}])
```
```
DatasetGenerationError: An error occurred while generating the d... | [
-0.07609846442937851,
-0.2211441993713379,
0.014412201009690762,
0.01814841851592064,
0.34675800800323486,
-0.09240495413541794,
0.10916101187467575,
0.20534390211105347,
0.1476440280675888,
0.36890295147895813,
0.07669489830732346,
0.2192411571741104,
-0.3427641689777374,
0.15258485078811... |
https://github.com/huggingface/datasets/pull/8577 | Fix infinite repetition of empty iterable datasets | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8577). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | Infinite repetition currently restarts an empty input indefinitely. For example, a streaming dataset filtered down to zero rows hangs in `dataset.repeat(None).take(1)` instead of finishing. The same problem affects an empty distributed shard and an Arrow input containing only zero-row tables.
Stop infinite repetition ... | Fix infinite repetition of empty iterable datasets
Infinite repetition currently restarts an empty input indefinitely. For example, a streaming dataset filtered down to zero rows hangs in `dataset.repeat(None).take(1)` instead of finishing. The same problem affects an empty distributed shard and an Arrow input containi... | [
-0.14178752899169922,
-0.23508508503437042,
-0.04093222692608833,
-0.19054226577281952,
0.05191263183951378,
-0.28301629424095154,
0.2730185389518738,
0.2044234722852707,
0.16221871972084045,
0.2867491841316223,
0.13459615409374237,
0.10861001163721085,
-0.17637395858764648,
0.184957802295... |
https://github.com/huggingface/datasets/pull/8575 | Fix async map checkpoint accounting across repeated resumes | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8575). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | Resuming an iterable dataset with an asynchronous map function works once, but a checkpoint captured during the resumed iteration can skip additional examples on the next resume. A deterministic identity coroutine over IDs 0–9, checkpointed after two outputs and again after one more output, produces `[0, 1, 2, 5, 6, 7,... | Fix async map checkpoint accounting across repeated resumes
Resuming an iterable dataset with an asynchronous map function works once, but a checkpoint captured during the resumed iteration can skip additional examples on the next resume. A deterministic identity coroutine over IDs 0–9, checkpointed after two outputs a... | [
-0.31582894921302795,
0.008758404292166233,
-0.08455117791891098,
-0.35339200496673584,
-0.1498624086380005,
-0.3290441930294037,
0.44311338663101196,
0.2504814565181732,
-0.1256839632987976,
-0.11365680396556854,
0.1310632973909378,
0.45841413736343384,
-0.0701218843460083,
-0.10305752605... |
https://github.com/huggingface/datasets/pull/8569 | Shard the sources when skipping a distributed IterableDataset | The docs for this PR live [here](https://moon-ci-docs.huggingface.co/docs/datasets/pr_8569). All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. | `SkipExamplesIterable.shard_data_sources` returned `self` when `split_when_sharding` is False, so a `.skip()` applied after `split_dataset_by_node` threw the sharding away and every rank iterated the whole dataset. `TakeExamplesIterable` already handles that branch by sharding the sources and leaving `n` alone, so this... | Shard the sources when skipping a distributed IterableDataset
`SkipExamplesIterable.shard_data_sources` returned `self` when `split_when_sharding` is False, so a `.skip()` applied after `split_dataset_by_node` threw the sharding away and every rank iterated the whole dataset. `TakeExamplesIterable` already handles that... | [
-0.1972104161977768,
-0.2527852952480316,
-0.004973702132701874,
-0.004438798408955336,
0.05931602790951729,
-0.17621886730194092,
0.32067251205444336,
0.20271195471286774,
-0.06326252222061157,
0.33627304434776306,
-0.18997468054294586,
0.16454914212226868,
0.004600999876856804,
0.1173225... |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 16