NVIDIA/DALI v1.35.0 on GitHub

Key Features and Enhancements

This DALI release includes the following key features and enhancements:

Added support for PaddlePaddle and JAX iterators (#5279, #5282).
Added support for checkpointing of iterators with different last batch policies (#5298, #5278).
Added tutorial on running DALI with T5X (#5286).
Added do_not_convert decorator to address problems with parallel fn.external_source and conditional execution (#5263).

Fixed Issues

Fixed missing nvmlShutdown calls (#5311).
Fixed fn.readers.video handling of sequences bigger than 2GB (#5307).
Fixed fn.resize handling of samples larger than 2GB (#5306).
Fixed support for multi node JAX sharding (#5242).
Fixed handling of decorated callbacks and methods in fn.external_source (#5268).
Fixed insufficient synchronization when restoring random generators from a checkpoint (#5273).

Improvements

Skip slow checkpointing tests when sanitizer is enabled (#5310)
Support checkpointing in PaddlePaddle iterator (#5279)
Bump protobuf version requirements in CMake. (#5312)
Dependency update (#5308)
Add SSL support to aarch64 deps containers (needed by CMake) (#5300)
Enable OpenSSL support (allow https downloading from cmake) (#5299)
Expose checkpointing in C API (#5287)
Enable AG-transformed code to show user code in exception (#5291)
Remove stage-related public APIs (from Pipeline and Executor) (#5244)
Add T5X tutorial (#5286)
Temporarily reduce the number of epochs in SBSA torch test (#5290)
Add a tool for correcting typos (#5193)
Support checkpointing in JAX iterator (#5282)
Expose do_not_convert decorator (#5263)
Fix Pipeline docs formatting (#5283)
Generalize iterator checkpointing tests (#5278)

Bug Fixes

Add missing calls to nvmlShutdown (#5311)
Fix bugs in iterator checkpointing, enable other last batch policies (#5298)
Fixes support for video sequences bigger than 2GB (#5307)
Fix resizing of volumes larger than 2G. (#5306)
Fix memory leak in C API test (#5303)
Fix support for multi node JAX sharding (#5242)
Fix TestPytorch (#5284)
Fix capitalized property descriptions (#5285)
Fix the lack of flavor in the conda DALI package name (#5281)
Allow source to be a decorated function or method (#5268)
Fix f-string in RN50 data pipeline test (#5280)
Remove old assumption from iterator docs (#5277)
Suppress leaks from inside dlopen (#5245)
Fix PaddlePaddle plugin docs (#5276)
Fix synchronization issues in RNG checkpointing utils (#5273)

Breaking API changes

There are no breaking changes in this DALI release.

Deprecated features

No features were deprecated in this release.

Known issues:

The following operators: experimental.readers.fits, experimental.decoders.video, experimental.inputs.video, and experimental.decoders.image_random_crop do not currently support checkpointing.
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==1.35.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda120==1.35.0

or just:

pip install nvidia-dali-cuda120==1.35.0
pip install nvidia-dali-tf-plugin-cuda120==1.35.0

For CUDA 11:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda110==1.35.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda110==1.35.0

or just:

pip install nvidia-dali-cuda110==1.35.0
pip install nvidia-dali-tf-plugin-cuda110==1.35.0

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 v1.35.0 DALI v1.35.0 on GitHub