gorgonia

For Go shops that want to keep their ML inference and training code in the same language ecosystem as their backend services avoiding Python dependency management, GIL limitations, and cross-language FFI overhead Gorgonia provides a native alternative to wrapping Python libraries.

The library uses a computation graph paradigm similar to TensorFlow 1.x, where users define operations as nodes in a directed acyclic graph and Gorgonia handles forward computation, gradient computation via automatic differentiation, and optimization updates.

It supports standard neural network building blocks dense layers, convolutional layers, recurrent cells, attention mechanisms and integrates with CUDA via the cu package for GPU acceleration on NVIDIA hardware.

Gorgonia is open-source under the Apache 2.0 license and is the most mature ML framework in the Go ecosystem, though Go remains significantly less popular than Python for ML research and development.

It is most relevant for Go engineers integrating ML inference into high-performance services where the operational simplicity of a single-language stack outweighs the broader ecosystem available in Python.

The project is actively maintained with a contributor community focused on expanding operator coverage and improving documentation.