News

Open Source DL Framework 'SmallTrain' Now a Wrapper for TensorFlow and PyTorch

• By John K. Waters
• 01/06/2021

Geek Guild, the company behind the open-source SmallTrain deep learning (DL) framework, has announced the availability of the framework as a standalone deep learning library and as a wrapper for TensorFlow and PyTorch. The source code for SmallTrain 0.2.0 is open-source, free to use, and released for commercial use. "We have released the source code with the desire to deliver Artificial Intelligence (AI) that can withstand commercial and professional use," the company said in a statement.

SmallTrain is an open-source framework designed to enable users to develop transfer learning models pre-trained in various data. It uses 60+ layers of pyramid neural-network architecture built from state-of-the-art, cutting edge algorithms. These models allow for the rapid development of high-accuracy deep-neural networks without large training-data requirements, the company said.

"During the recent AI boom, companies develop customized 'Narrow AI', which are active only in limited use cases," the company said. "Here in Geek Guild, we believe AI is fundamental for machines to help humanity build symbiotic spheres as 'Harmonic AI,' aiming to integrate independent AI for developing general-purpose AI.

The framework was designed to helped developers lacking extensive data science knowledge to build learning models.. The company's motto: "We do the heavy lifting."

The company designed SmallTrain to serve as a wrapper equivalent to Keras, the deep learning API written in Python and running on top of TensorFlow. The difference is that Keras is suitable for proof of concept (PoC), while SmallTrain is suitable for both PoC and commercial use. But it's similar to Keras, in that it's easy to use for beginners, the company said.

Wrapping a library, such as TensorFlow or PyTorch, allows users to access the library via the wrapper, the company explained, "…so that even if the library is replaced or the interface of the library is changed, the changes are stopped only inside the wrapper and the effect of the change are stopped."

The Kyoto, Japan-based Guild was formed in 2018 to pursue the development of " create General Artificial-Intelligence based on harmony between humans and machines."