Skip to content

chore: bump version to 0.2.76 #935

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tchaton merged 1 commit into from
Sep 30, 2026
Merged

chore: bump version to 0.2.76 #935

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions src/litdata/CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,12 +8,28 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).

## [unreleased] - YYYY-MM-DD

## [0.2.76] - 2026-09-30

### Added

- GPU-local CPU affinity and optional NUMA memory binding helpers for ranks and DataLoader workers. ([#932](https://github.com/Lightning-AI/litData/pull/932))
- Opt-in NFS direct reads for temporal windows with `window_direct_io=True`, plus reusable Python and experimental native-loader helpers. Direct reads bypass the client page cache; the native launcher requires a C compiler. ([#934](https://github.com/Lightning-AI/litData/pull/934))

- `Downloader.adownload_bytes` supports validated async range reads, with native S3/R2 requests and direct local slices. Added `StreamingDataset.read_window` / `aread_window` and `TemporalArrayLoader` for built-in field/window selection, grouped array records, and index-only frame counts through the normal optimize workflow.

### Changed

- Declare PyTorch 2.4.0 as the minimum supported version. ([#909](https://github.com/Lightning-AI/litData/pull/909))
- Remove the PyArrow upper version bound from optional dependencies. ([#929](https://github.com/Lightning-AI/litData/pull/929))

### Fixed

- Preserve mounted storage paths and use exact, unbuffered local range reads for temporal windows. ([#931](https://github.com/Lightning-AI/litData/pull/931))
- Honor `index_path` for local dataset directories. ([#917](https://github.com/Lightning-AI/litData/pull/917))
- Preserve output prefixes when deriving checkpoint filenames. ([#926](https://github.com/Lightning-AI/litData/pull/926))
- Reset wrapper sample and cycle counters with `reset_state_dict` and avoid mutating restored checkpoint counters. ([#928](https://github.com/Lightning-AI/litData/pull/928))
- Refresh temporary credentials against their reported expiry and send S3 reads to the bucket's reported endpoint and region. ([#906](https://github.com/Lightning-AI/litData/pull/906), [#904](https://github.com/Lightning-AI/litData/pull/904))

- Temporal window reads on local and parallel POSIX filesystems reuse bounded chunk mappings and prefetch selected ranges, preserving owned outputs and safe concurrent mapping lifetimes.
- S3/R2 range reads close SDK response bodies on success and failure and reject truncated responses.
- Resumed optimize workers with no remaining inputs preserve their checkpoint's schema and compression metadata, so index merging also succeeds when a worker completed before interruption.
Expand Down
2 changes: 1 addition & 1 deletion src/litdata/__about__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@

import time

__version__ = "0.2.75"
__version__ = "0.2.76"
__author__ = "Lightning AI et al."
__author_email__ = "pytorch@lightning.ai"
__license__ = "Apache-2.0"
Expand Down
Loading