pytorch/ELF
A reinforcement learning research platform for game-playing AI, featuring an AlphaZero-style implementation that achieved professional-level Go performance.

Velocity · 7d
+1.2
★ / day
Trend
→steady
star history
ELF is an extensive platform for game research built by Facebook AI Research. It implements AlphaGoZero and AlphaZero algorithms in C++ with PyTorch integration for training game-playing agents. The platform was used to develop ELF OpenGo, which achieved a 14-0 record versus top-30 Go players, demonstrating professional-level performance through self-play reinforcement learning.