Download SINGA¶
To verify the downloaded tar.gz file, download the KEYS and ASC files and then execute the following commands
% gpg --import KEYS % gpg --verify downloaded_file.asc downloaded_file
You can also check the SHA512 or MD5 values to see if the download is completed.
v2.0.0 (20 April 2019):
New features and major updates,
Enhance autograd (for Convolution networks and recurrent networks)
Support ONNX
Improve the CPP operations via Intel MKL DNN lib
Implement tensor broadcasting
Move Docker images under Apache user name
Update depdent lib versions in conda-build config
v1.2.0 (6 June 2018):
New features and major updates,
Implement autograd (currently support MLP model)
Upgrade PySinga to support Python 3
Improve the Tensor class with the stride field
Upgrade cuDNN from V5 to V7
Add VGG, Inception V4, ResNet, and DenseNet for ImageNet classification
Create alias for conda packages
Complete documentation in Chinese
Add instructions for running Singa on Windows
Update the compilation, CI
Fix some bugs
v1.1.0 (12 February 2017):
New features and major updates,
Create Docker images (CPU and GPU versions)
Create Amazon AMI for SINGA (CPU version)
Integrate with Jenkins for automatically generating Wheel and Debian packages (for installation), and updating the website.
Enhance the FeedFowardNet, e.g., multiple inputs and verbose mode for debugging
Add Concat and Slice layers
Extend CrossEntropyLoss to accept instance with multiple labels
Add image_tool.py with image augmentation methods
Support model loading and saving via the Snapshot API
Compile SINGA source on Windows
Compile mandatory dependent libraries together with SINGA code
Enable Java binding (basic) for SINGA
Add version ID in checkpointing files
Add Rafiki toolkit for providing RESTFul APIs
Add examples pretrained from Caffe, including GoogleNet
v1.0.0 (8 September 2016):
New features and major updates,
Tensor abstraction for supporting more machine learning models.
Device abstraction for running on different hardware devices, including CPU, (Nvidia/AMD) GPU and FPGA (to be tested in later versions).
Replace GNU autotool with cmake for compilation.
Support Mac OS
Improve Python binding, including installation and programming
More deep learning models, including VGG and ResNet
More IO classes for reading/writing files and encoding/decoding data
New network communication components directly based on Socket.
Cudnn V5 with Dropout and RNN layers.
Replace website building tool from maven to Sphinx
Integrate Travis-CI
v0.3.0 (20 April 2016):
New features and major updates,
Training on GPU cluster enables training of deep learning models over a GPU cluster.
Python wrapper improvement makes it easy to configure the job, including neural net and SGD algorithm.
New SGD updaters are added, including Adam, AdaDelta and AdaMax.
Installation has fewer dependent libraries for single node training.
Heterogeneous training with CPU and GPU.
Support cuDNN V4.
Data prefetching.
Fix some bugs.
v0.2.0 (14 January 2016):
New features and major updates,
Training on GPU enables training of complex models on a single node with multiple GPU cards.
Hybrid neural net partitioning supports data and model parallelism at the same time.
Python wrapper makes it easy to configure the job, including neural net and SGD algorithm.
RNN model and BPTT algorithm are implemented to support applications based on RNN models, e.g., GRU.
Cloud software integration includes Mesos, Docker and HDFS.
Visualization of neural net structure and layer information, which is helpful for debugging.
Linear algebra functions and random functions against Blobs and raw data pointers.
New layers, including SoftmaxLayer, ArgSortLayer, DummyLayer, RNN layers and cuDNN layers.
Update Layer class to carry multiple data/grad Blobs.
Extract features and test performance for new data by loading previously trained model parameters.
Add Store class for IO operations.
v0.1.0 (8 October 2015):
Major features include,
Installation using GNU build utility
Scripts for job management with zookeeper
Programming model based on NeuralNet and Layer abstractions.
System architecture based on Worker, Server and Stub.
Training models from three different model categories, namely, feed-forward models, energy models and RNN models.
Synchronous and asynchronous distributed training frameworks using CPU
Checkpoint and restore
Unit test using gtest
Disclaimer
Apache SINGA is an effort undergoing incubation at The Apache Software Foundation (ASF), sponsored by the name of Apache Incubator PMC. Incubation is required of all newly accepted projects until a further review indicates that the infrastructure, communications, and decision making process have stabilized in a manner consistent with other successful ASF projects. While incubation status is not necessarily a reflection of the completeness or stability of the code, it does indicate that the project has yet to be fully endorsed by the ASF.
