4.3.0

APACHE SINGA -- 2024 SIGMOD Systems Award

June 22, 2024

Apache SINGA received the 2024 SIGMOD Systems Award, which is awarded to an individual or set of individuals to recognize the development of a software or hardware system whose technical contributions have had significant impact on the theory or practice of large-scale data management systems:

The SIGMOD 2024 system award goes to Apache SINGA: "A Distributed Deep Learning System, for the development of a distributed, efficient, scalable, and easy-to-use deep learning platform for large scale data analytics."

snapshot11

Official Release from ACM SIGMOD: https://sigmod.org/sigmod-awards/sigmod-systems-award/

The SINGA team received the award at the 2024 ACM SIGMOD/PODS Conference held in Santiago, Chile from June 9 to 14, 2024.

Release Version 4.2.0

March 15, 2024

Apache SINGA V4.2.0 is released on March 15, 2024. There are six major features added in this release.

Add support for deep learning models running on top of PolarDB, including efficient model selection, dynamic model creation, flexible training configurations, and optimization of in-database analytics for scalability and efficiency.
Introduce a new horizontal federated learning example using the Bank dataset.
Enhance examples with sample training data for model selection testing.
Update the website by refining the star button on the main page and the display of star statistics.
Update the Python versions for wheel files.
Fix bugs by addressing the rat check files and updating the license files.

Please check the overview page and detailed guides.

Release Version 4.1.0

November 5, 2023

Apache SINGA V4.1.0 is released on November 5, 2023. There are six major features added in this release.

Add an example for malaria detection using cell images and structured data learning.
Add support for in-database model definition, selection, and training-free model evaluation metrics in RDBMS, along with a coordinator for balancing evaluations.
Enhance distributed training by implementing sum error loss, improving the optimizer, and iterative checking in ModelMeta class.
Enhance example code with flexible training configurations and dynamic models with varying layer sizes.
Update the website with illustrations for database integration and Apache SINGA users.
Fix bugs by updating NVIDIA_GPGKEY in Dockerfile, versions of dependencies in wheel file, and the collections module in model.py.

Please check the overview page and detailed guides.

Release Version 4.0.0

April 7, 2023

Apache SINGA v4.0.0 is released on April 7, 2023. There are seven major features added in this release.

Enhance distributed training by adding GPU configuration support, increasing max epochs, printing mini-batch information, and allowing switching between CPU and GPU.
Enhance example code by updating normalize forward function args, updating xceptionnet, adding weight decay, momentum, and learning rates arguments, adding training scripts for more datasets and models, adding resnet dist version, adding cifar 10 multi-process and sparsification implementation, and updating CIFAR datasets download and load functions.
Enhance the webpage by updating online documentation for distributed training.
Promote code quality by updating inline comments for preprocessing and data loading.
Update the PIL image module.
Update the runtime Dockerfile.
Update the conda files.

Please check the overview page and detailed guides.

Release Version 3.3.0

June 7, 2022

Apache SINGA V3.3.0 is released on June 7, 2022. There are seven major features added in this release.

New examples are added, including a CNN example for the BloodMnist dataset and an example for medical image analysis.
Distributed training is enhanced with key information printing and optimized logging for faster training.
The example code is improved with additional datasets and models, and updated scripts and paths for flexibility.
The webpage is enhanced by reconstructing the SINGA webpage, updating the Git website, and refreshing Chinese and Vietnamese documentation.
Debugging and adding assertions for input tensor data types are implemented in opt.py.
Pointer types are changed to void for generalizing data types.
Bugs are fixed, including resolving Python test errors and updating the model of pad from bytes to str.

Please check the overview page and detailed guides.

Release Version 3.2.0

August 15, 2021

Apache SINGA V3.2.0 is released on August 15, 2021. There are eight major features added in this release.

Add one CIFAR-10 distributed CNN example for benchmarking performance and one large CNN example for training with a dataset from the filesystem.
Improve the data augmentation module and add device synchronization for more accurate time measurements during distributed training.
Add support for half-precision floating-point format (fp16) in deep learning models and computational kernels.
Update new ONNX APIs and fix ONNX examples accordingly, namely, DenseNet121, ShuffleNetv1, ShuffleNetv2, SqueezeNet, VGG19.
Add a new method to resize images by given width and height.
Use Docusaurus versioning to simplify the process of generating the project homepage.
Unify the formats of docstrings and command-line arguments to promote code quality.
Fix the CI build error by downloading the TBB binaries, add a disabling graph option for accessing parameter or gradient tensors during distributed training, and solve warnings of deprecated functions in the distributed optimizer module.

Please check the overview page and detailed guides.

Release Version 3.1.0

October 30, 2020

Apache SINGA V3.1.0 is released on October 30, 2020. There are six major features added in this release.

Update the Tensor core to support tensor transformations (reshape, transpose) for tensors up to 6 dimensions and implement traverse-unary-transform in the Cuda backend, similar to the CPP backend.
Add new tensor operators into the autograd module.
Reconstruct sonnx to support creating operators from both layer and autograd, re-write SingaRep for a more powerful intermediate representation of SINGA, and add a SONNXModel for a uniform API and features.
Replace Travis CI with Github workflow and add quality and coverage management.
Add compiling and packaging scripts to create wheel packages for distribution.
Fix bugs including the IMDB LSTM model example training script, Tensor operation Mult on Broadcasting use cases, Gaussian function on Tensor with odd size, and update a testing helper function gradients() in autograd to lookup param gradient by param python object id for testing purposes

Please check the overview page and detailed guides.

Release Version 3.0.0

April 19, 2020

Apache SINGA V3.0.0 is released on April 18, 2020. This is the first official release since SINGA became the first Top-Level Apache project. There are three major features added in this release.

A new Python class Module is introduced. New models (neural networks) can be implemented by inheriting it. The benefit is that SINGA will create a computational graph at the backend for speed and memory optimization.
The soonx module is enhanced to support more ONNX operators. Multiple ONNX models have been tested for training and inference using SINGA.
Communication optimization for distributed training, including gradient sparsification and compression. NCCL is integrated as the communication library.

Please check the overview page and detailed guides.

Apache SINGA as a Top-Level Project

October 16, 2019

The Apache Software Foundation (ASF) recently announced Apache SINGA as a Top-Level Project.

The Apache Software Foundation (ASF), the all-volunteer developers, stewards, and incubators of more than 350 Open Source projects and initiatives, announced today Apache® SINGA™ as a Top-Level Project (TLP).

Apache SINGA is an Open Source distributed, scalable machine learning library. The project was originally developed in 2014 at the National University of Singapore, and was submitted to the Apache Incubator in March 2015.

Foodlg Powered by SINGA

October 3, 2019

Food(lg), the mobile app to help patients with prediabetes, is powered by SINGA and this app can now be downloaded on the APP Store or Google Play. Watch this video of Food(lg) for more details.