PyGo: A Deep Learning Framework Where Go Calls Python Calls C++

Short summary

PyGo is an experimental deep learning framework that embeds CPython inside a Go process, with Go as the top-level API, Python handling autograd and model zoo, and C++/CUDA doing raw compute via zero-copy shared memory tensors. The project has partial implementations of LLaMA-3, GPT-2, BERT, ViT, and Whisper with Flash Attention v2 and quantisation support. The author is seeking core contributors with CGo, autograd, and CUDA kernel experience to solve CGo call overhead at the tensor boundary.

• •PyGo embeds CPython in a Go binary for a three-layer deep learning stack with zero-copy tensors
• •Partial implementations of major models including LLaMA-3, GPT-2, BERT, ViT, and Whisper
• •Seeking contributors to solve CGo call overhead and build out kernel and autograd layers

Generated with AI, which can make mistakes.