Reliquary launches training, publishes first model checkpoints

Reliquary began training and policy updates today following a 10x miner speed improvement. The team published initial model checkpoints requiring miner updates every 10 windows. A shift to decentralized distributed reinforcement learning on policy—described as unexplored research—is underway, with results planned for publication.

• •First checkpoints published; miners update models every 10 windows
• •Miners achieved 10x speed improvement prior to training start
• •Switched to decentralized distributed RL on policy approach
• •Novel research framework planned for academic release