Benchmarks¶
Scripts and evaluation artifacts live under benchmarks/, results under
benchmarks/results/.
- Baselines — scripted and learned reference policies
- Self-Play & Elo — population tournament and rating ladder
- Run — commands: throughput, competition eval, full sweeps, env spec
Headline throughput¶
JAX vectorized backend: 270,795 env-steps/s (1,083,181 agent-steps/s) at batch 4096 on CPU. See Run for the full table and how to reproduce it.