NVIDIA/DALI v1.32.0 on GitHub

Key Features and Enhancements

This DALI release includes the following key features and enhancements:

Added Python signatures/type hints to the DALI Python API (#5096, #5039, #5112, #5118, #5124, #5143).
Added experimental support for checkpointing DALI pipelines at arbitrary iterations (fn.readers.file, CPU fn.random generators, and stateless operators) (#5085, #5088, #5103, #5114, #5113, #5142, #5128, #5144).
Added support for CUDA 12.3 (#5106).

Fixed Issues

Fixed a potential crash on process teardown when using the fn.python_function in the DALI pipeline (#5138).
Removed unused arguments from fn.fast_resize_crop_mirror. The operator was deprecated in favor of fn.resize_crop_mirror (#5123).
Fixed a potential crash on process teardown when using CPU fn.resize in the DALI pipeline (#5133).
Fixed constructing tensors from stream-aware __cuda_array_interface__ v3 (#5125).
Fixed the crop_pos_z handling for a fixed crop window in the fn.crop operator (#5119).
Fixed releasing Python tensors without GIL in fn.external_source. The problem led to crashes when using fn.external_source in no_copy or parallel mode with conditional execution enabled in the pipeline (#5101).

Improvements

Improve checkpointing docs (#5142)
Add type hints to python-defined ops, run and tfrecord APIs (#5118)
Allow mid-epoch checkpointing in FileReader (#5113)
Mark more stateless operators (#5114)
Load ASAN during the build with sanitizers (#5121)
Add default arg values to JAX decorator (#5115)
Improve docs of variance and stddev (#5130)
Update python op tutorial (#5120)
Generalize __module__ handling and hide private modules docs (#5112)
Add mid-epoch checkpointing to Loader (#5103)
Add ViT data processing pipeline to hw_bench_script (#5110)
Add JAX Getting Started Tutorial (#5095)
Generate type hints for fn and ops APIs (#5096)
Move to CUDA 12.3 (#5106)
Add the BUILD ID to the Xavier wheel name (#5099)
Add Fast-Forward to Loader base (#5088)
Update installation guide to move https://pypi.nvidia.com/ or just pypi (#4815)
Remove Python 3.7 support and replace defaults with Python 3.8 (#5089)
Add better error for VFR check (#5092)
Move the snapshot queue from loader to reader (#5085)
Refactor ops and fn APIs (#5039)

Bug Fixes

Fix Python Functions ops signature (#5143)
Fix stateless tests (#5144)
Fix exit sequence. (#5138)
Fix data_iterator docs (#5140)
Reimplement (Fast)ResizeCropMirror in terms of Resize. Remove dead code. (#5123)
Ensure that the default host resource is destroyed after ResamplingFilters CPU instance. (#5133)
Fix CVE-2023-45853 in zlib (#5116)
Fix stub generation during DALI build (#5124)
Improve checkpointing error message (#5128)
Fix constructing tensors from cuda_array_interface v3. (#5125)
Fix lack of crop_pos_z handling for fixed crop window (#5119)
Fix CVE-2022-33065 in libsnd (#5105)
Add DALI package dependencies to the custom build test (#5122)
Make sure that the relase of python memory from the DALI tensor happens inside GIL (#5101)
Relax UpdatePropertiesFromSamples check constraints (#5098)

Breaking API changes

DALI 1.31 was the final release that supported Python 3.7.

Deprecated features

The operator fn.fast_resize_crop_mirror was deprecated in favour of fn.resize_crop_mirror.

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 https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda120==1.32.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda120==1.32.0

or for CUDA 11:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda110==1.32.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda110==1.32.0

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:

https://developer.download.nvidia.com/compute/redist/nvidia-dali/libsndfile-1.2.2.tar.gz

NVIDIA/DALI v1.32.0 DALI v1.32.0 on GitHub