Nodes
The "Node" is the fundamental building block of the OSMI decentralized network, responsible for enabling AI operations by providing computational power from various sources, such as GPUs from individual users. This unique piece of software allows anyone with sufficient hardware to contribute to the OSMI network by running a node, earning $OSMI tokens in return. Nodes execute AI inference tasks and distribute computational power across the network, reducing costs and increasing accessibility for users and developers. These decentralized nodes create the peer-to-peer network that can operate independently of any central authority. This framework enables distributed computation, storage, and communication among its participants while promoting transparency, collaboration, and community-driven innovation.
Types of Nodes
The OSMI framework consists of several nodes that perform specific functions within the decentralized system. These nodes include:
Inference Nodes
Currently there are two powerhouse node types doing the heavy lifting and performing inference tasks for OSMI. They are:
LLM Node (xgllm)
Open to anyone, an xgllm node offers language model completion services for supported languages. It requires a beefy GPU with 12GB VRAM or more and is subject to regular performance audits by xgcoordinator for continued participation.
Image Node (xgimage)
Similar to an LLM node, xgimage nodes load a large language model trained and dedicated to generating images. It also requires an impressive GPU with 12GB VRAM or more and is subject to regular performance audits by xgcoordinator for continued participation. This node is in Alpha state and is not publicly available at this time.
Memory Node (xgstore)
Open to all participants, xgstore provides indexed storage of ingested knowledge and similarity search capabilities. It requires a low-to-mid-level GPU with 6GB VRAM or more and is subject to regular performance audits for continued participation. Xgstore utilizes Storj for permanent storage.
Ingest Node (xgingest)
Also open to anyone, an xgingest node crawls the web for documents and extracts scraped knowledge, which it sends along to xgstore for indexing. It does not require a GPU but can utilize one if available.
Coordinator Node (xgcoordinator)
Run by trusted entities, xgcoordinator provides connection details for other nodes, orchestrates audits of these nodes, and validates their participation to determine distribution. Operators must agree to legal agreements before participating in this role.
Example Node Implementation (LLM)
Although OSMI supports several types of nodes, currently the most beneficial node to run is the inference node. Proprietors of these nodes will be obligated to fulfill certain GPU specifications. As a result of fulfilling these requirements, they will be entitled to more extensive distribution rewards.
(With the "xg" prefix, we are acknowledging the influence of knowledge graphs, which serve as a distinctive characteristic and source of inspiration for this project)
Last updated
