The ALIA platform consists of four chief technical components collectively known as the Polyglot Stack. They include: NMT system – runs in a local virtualized server that allows training of several language models concurrently within separate neural network instances.
Parameter Server – handles the NMT graph, the network layout, direction, and functionality of the nodes and feeds the variables that need to be changed as the process progresses directly to the compute nodes.
Distributed Computing Management System – responsible for managing the hardware/software environment of the cluster. It sends and receives data to determine node availability and exchanges logging information with all nodes in the cluster.
aliaOS Cluster – carries out the bulk of the NMT calculations and 100% of the document translation work. It consists of two parts: the head node, which manages the software on the cluster, and the cluster’s compute nodes, which provide the processing power.
- NMT Computations are passed through the parameter to the head node. The head node is the external interface to the cluster and receives all external network connections, processes incoming requests, and assigns work to compute nodes with GPUs.
- Language Model Translation Files are managed and distributed by the cluster management server that ensures that the correct version is sent to the aliaOS nodes via the head node and preloaded on to designated GPU cards.
- The aliaOS compute nodes will perform work in response the PolyGlot stack as well as blockchain protocols. The nodes are integrated with the Tendermint Application Blockchain Interface (ABCI) enabling them operate independent and respond directly to transaction requests from customers.
Nodes can also be configured with mining software for other cryptocurrencies. This will enhance profitability for the miners and ensure adoption of the aliaOS system. In other words, they can automatically switch back to mining Bitcoin when not serving the mining ALIA or translating documents.