0xRobinR / sfai-gf-trainer

School for AI

a project where AI can be trained decentralized, using the decentralized storage provider, BNB Greenfield.

Setup

• Install the requirements

pip install - r requirements.txt

• Run the parameter server or professor nodes (PN), that will listen to the AI train requests from the BNB Chain

python main.py

• Once the PN is up and running, connect the worker nodes, this will be incharge of computing the model and maintaining a two-way communication with the professor nodes.

python .\workers\main.py

As the worker establishes the connection, the PN will coordinate with the given model and data, and using tensorflow's ParameterServerStrategy, it will distribute the model to the workers, and will start the training process.

Current State

• Currently, it’s all backend, will have to work on the user interface side as well

• Professor nodes are listening to the events

• Worker nodes are connected to the professor nodes

• The model is distributed to the workers

• Downloading the files from Greenfield and then processing it locally, and then uploading the model with the user only access to the file (encrypted model)

• while downloading, the speed of accessing the resources is very slow.

Further, I need more help from the BNB Chain team, to get the things done and in the correct way.