From e672eac53aa3cf95520bab08a4f14f193b775092 Mon Sep 17 00:00:00 2001 From: tchaton Date: Wed, 30 Sep 2026 10:07:37 +0100 Subject: [PATCH] chore: bump version to 0.2.76 Co-Authored-By: Codex --- src/litdata/CHANGELOG.md | 16 ++++++++++++++++ src/litdata/__about__.py | 2 +- 2 files changed, 17 insertions(+), 1 deletion(-) diff --git a/src/litdata/CHANGELOG.md b/src/litdata/CHANGELOG.md index 1b4ed5bf9..8c59930c5 100644 --- a/src/litdata/CHANGELOG.md +++ b/src/litdata/CHANGELOG.md @@ -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. diff --git a/src/litdata/__about__.py b/src/litdata/__about__.py index b9f7694ba..109cc3bd4 100644 --- a/src/litdata/__about__.py +++ b/src/litdata/__about__.py @@ -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"