TL;DR: Project Maya allows private, local inference of the 321B MoE GLM-5.3-Flash model on consumer GPUs, offering an OpenAI/Anthropic-compatible API.
Summary: Project Maya provides an engine, server, and dashboard for running the GLM-5.3-Flash (321B MoE) model privately on personal GPUs. It supports both NVIDIA and AMD hardware, from single cards to 16, across Linux and Windows, and includes quantization tools.
Why it matters: This enables indie developers to experiment with a large Mixture-of-Experts model locally, reducing API costs and enhancing data privacy. Builders should explore its OpenAI/Anthropic-compatible API for integrating powerful local LLM capabilities into their applications.
Source: github_topics