Shypd.aiThe largest open index of AI toolsRaven Protocol is an open-source distributed deep learning training framework that enables asynchronous parallel training across heterogeneous consumer-grade devices. It allows users to train complex deep learning models using internet-connected PCs.
Trending
About
Raven Protocol, specifically Ravnest, is a distributed deep learning training framework designed to train complex deep learning models across heterogeneous consumer-grade PCs connected via the internet. It combines data and model parallelism with a novel asynchronous training approach, orchestrating distributed training through a four-stage pipeline: matchmaking & cluster formation, zero-bubble asynchronous model parallel training, parallel multi-ring all-reduce for global synchronization, and fault recovery & dynamic scaling. The framework is built to handle unreliable consumer-grade hardware, offering features like gradient compression, adaptive routing, and a flexible update rule to accommodate varying device capabilities and network conditions. It supports various models, including CNNs and small LLMs, and provides an extensible trainer API for custom architectures.
Capabilities
Pricing & Plans
Open Source
Not publicly disclosed. Check ravenprotocol.com for current pricing.
FAQs
Trending
