cuda Tools

Voicebox is an open-source AI voice studio that runs entirely on your machine. It solves the problem of expensive, cloud-based voice cloning and generation by offering a free, local alternative to services like ElevenLabs, ensuring your voice data never leaves your computer. It allows you to clone voices, generate speech across multiple TTS engines, dictate into any application, and interact with AI agents using custom voices. This tool is for developers, content creators, and privacy-conscious users who want powerful, customizable voice AI without subscriptions or data privacy concerns.

FlashKDA is a library of high-performance CUDA kernels implementing the Kimi Delta Attention (KDA) mechanism. It solves the computational bottleneck of attention layers in large language models by providing highly optimized, low-level primitives that dramatically accelerate training and inference. This library is designed for ML engineers and researchers working with Kimi-based architectures who require maximum throughput and efficiency from their hardware.