RL Is Bottlenecked by Inference. Scale It Independently
(
2 comments
| 8 points
| on HN
)
Re:RL Is Bottlenecked by Inference. Scale It Independently
by efiop on Wednesday August 05, 2026 @10:34AM
(#49183587)
what did gpu hours look like here? with 3 replicas for a 1.8x speedup, the cost tradeoff isn’t obvious.
[ Parent
| on HN ]
Re:RL Is Bottlenecked by Inference. Scale It Independently
by efiop on Wednesday August 05, 2026 @10:38AM
(#49183667)
ah, nevermind. 3 engines seem cheaper overall too: 7x661s vs 5x1200s of allocated H100 time per step. Nice.
[ Parent
| on HN ]