github NVIDIA/DALI v1.26.0
DALI v1.26.0

4 days ago

Key Features and Enhancements

This DALI release includes the following key features and enhancements:

  • Added O_DIRECT mode support to fn.readers.numpy (#4796, #4848).
  • Added an option to filter out iscrowd entries from COCO (#4792).
  • Moved to CUDA 12.1 update 1 (#4798).
  • Made DALI GPU tensors directly convertible to PyTorch (#4800).

Fixed Issues

  • Fixed a memory leak in the fn.experimental.remap operator (#4790).
  • Fixed the recognition of new CuPy ndarrays in fn.external_source (#4793).


  • Cumulative dependency update for May, 2023. (#4823)
  • Add O_DIRECT support in numpy_reader (#4796)
  • Add a native dataloader to RN50 PyTorch example (#4807)
  • Fix coverity issues (Apr 2023) (#4803)
  • Move to CUDA 12.1 update 1 (#4798)
  • Make DALI array_interface memory writable (#4800)
  • Add support for filtering in/our iscrowd entries from COCO (#4792)
  • Add bug and question templates to DALI github repo (#4782)
  • Rework conditional-like execution tutorial for arithmetic ops (#4795)
  • Add "depleted" operator trace (#4794)
  • Add "repeat_last" option to ExternalSource and handle it in Pipeline. (#4775)
  • Use dedicated GTC 2023 event links (#4781)

Bug Fixes

  • Fix race condition in the CPU numpy reader (#4848)
  • Update required packages for TL1_python-self-test_conda (#4843)
  • Fix FITS tests with python3.7, reduce memory usage in rand aug tests (#4844)
  • Fix FITS reader test with Python3.6 (#4835)
  • Fix TensorFlow tests (#4837)
  • Fix conda test and tests on Xavier (#4827)
  • Restrict the urllib3 version in tests to <2.0 (#4824)
  • Fix error propagation from the QA test (#4821)
  • Make TL0_python-self-test-base-cuda using the local CUDA toolkit (#4811)
  • Fix scratchpad usage in Remap. Add more documentation to scratchpad. (#4790)
  • Fix the regex that recognizes CuPy arrays. (#4793)

Breaking API changes

There are no breaking changes in this DALI release.

Deprecated features

No features were deprecated in this release.

Known issues:

  • The video loader operator requires that the key frames occur, at a minimum, every 10 to 15 frames of the video stream.
    If the key frames occur at a frequency that is less than 10-15 frames, the returned frames might be out of sync.
  • Experimental VideoReaderDecoder does not support open GOP.
    It will not report an error and might produce invalid frames. VideoReader uses a heuristic approach to detect open GOP and should work in most common cases.
  • The DALI TensorFlow plugin might not be compatible with TensorFlow versions 1.15.0 and later.
    To use DALI with the TensorFlow version that does not have a prebuilt plugin binary shipped with DALI, make sure that the compiler that is used to build TensorFlow exists on the system during the plugin installation. (Depending on the particular version, you can use GCC 4.8.4, GCC 4.8.5, or GCC 5.4.)
  • In experimental debug and eager modes, the GPU external source is not properly synchronized with DALI internal streams.
    As a workaround, you can manually synchronize the device before returning the data from the callback.
  • Due to some known issues with meltdown/spectra mitigations and DALI, DALI shows best performance when running in Docker with escalated privileges, for example:
    • privileged=yes in Extra Settings for AWS data points
    • --privileged or --security-opt seccomp=unconfined for bare Docker.

Binary builds

NOTE: DALI builds for CUDA 12 dynamically link the CUDA toolkit. To use DALI, install the latest CUDA toolkit.

CUDA 11.0 and CUDA 12.0 builds use CUDA toolkit enhanced compatibility. 
They are built with the latest CUDA 11.x/12.x toolkit respectively but they can run on the latest, 
stable CUDA 11.0/CUDA 12.0 capable drivers (450.80 or later and 525.60 or later respectively).
However, using the most recent driver may enable additional functionality. 
More details can be found in enhanced CUDA compatibility guide.

Install via pip for CUDA 12.0:
pip install --extra-index-url nvidia-dali-cuda120==1.26.0
pip install --extra-index-url nvidia-dali-tf-plugin-cuda120==1.26.0

or for CUDA 11:
pip install --extra-index-url nvidia-dali-cuda110==1.26.0
pip install --extra-index-url nvidia-dali-tf-plugin-cuda110==1.26.0

Or use direct download links (CUDA 12.0):

Or use direct download links (CUDA 11.0):

FFmpeg source code:

  • This software uses code of FFmpeg licensed under the LGPLv2.1 and its source can be downloaded here

Libsndfile source code:

Don't miss a new DALI release

NewReleases is sending notifications on new releases.