Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities

Official implementation for BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulation

Running Microsoft's BitNet inference framework via FastAPI, Uvicorn and Docker.

pure pytorch implementation of Microsoft's BitNet b1.58 2B4T

Official implementation of BitMamba-2. A scalable 1.58-bit State Space Model (Mamba-2 + BitNet) trained from scratch on 150B tokens. Includes JAX training code and high-performance C++ inference engine.

BitNet: Learning-Based-Bit-Depth-Expansion

Ultra-lightweight C++ inference engine for BitMamba-2 (1.58-bit SSM). Runs 1B models on consumer CPUs at 50+ tok/s using <700MB RAM. No heavy dependencies.

Peer-to-peer distributed AI inference using 1-bit quantized models. CPU-only, 70-82% energy savings, 103+ tokens/sec. Validated on Zen 4 & Zen 5 (+35% cross-gen improvement).

Distily: Language Model Distillation Toolkit and Library

Long term project about a custom AI architecture. Consist of cutting-edge technique in machine learning such as Flash-Attention, Group-Query-Attention, ZeRO-Infinity, BitNet, etc.

This is the repo for the MixKABRN Neural Network (Mixture of Kolmogorov-Arnold Bit Retentive Networks), and an attempt at first adapting it for training on text, and later adjust it for other modalities.

BitNet Inference Web UI: A modern web interface for running Microsoft's BitNet models efficiently on CPU. This project provides a user-friendly way to download, manage, and run inference with 1-bit quantized language models.

RSR-core: A High-Performance Engine for Low-Bit Matrix-Vector Multiplication

Windows-native BitNet and ternary LLM inference with CPU GGUF, GPU runtime, terminal and browser chat, and release zips.

🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

The JuniorOmega SDK is a sovereign, local-first spatial engineering stack optimized specifically for Apple Silicon. It is designed to bridge the gap between high-density sensor ingestion (LiDAR/TrueDepth) and automated fabrication (G-code/CNC).

🚀 Hybrid RAG: Local Neo4j + BitNet.cpp RAG System and Azure SaaS deployment. Fast vector search, instant Docker deployment via GitHub Container Registry. Complete RAG pipeline with ultra-efficient LLMs for enterprise knowledge management.

BitNet-inspired 1-bit Quantized Transformer for efficient protein function prediction and biological sequence modeling on low-power devices.

BitNet-inspired quantization-aware training and model compiler for running neural networks efficiently on ESP32 devices.

Modular AI system where independently-trained ternary specialists load on demand from disk, routed by geometric classification, sharing a unified lattice knowledge base, capable of autonomously growing new specialists — all running on a CPU laptop.