Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
List<Prompt> -> List<Rollout>
Query an external optimized execution engine (like vLLM) asynchronously to generate rollout sequences and log-probabilities for reinforcement learning updates.
Problem it solves
Generating policy rollouts in-process during reinforcement learning (like GRPO) is slow and does not benefit from optimized batching/paging engines.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.