Mistral-Large-4
Mistral AIThe episode generation concurrency adapts following backpressure signals from generators, services, and the trainer, which limits staleness.
specific method Β· filed under post-training
Adaptive Rollout Concurrency assigns each data source a budget of concurrent rollouts based on that source's generation demand.1,2 It scales concurrency with the source's expected demand (target samples over acceptance rate) and rollout duration, so slow or heavily filtered sources get more concurrent rollouts.3
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
Adaptive Rollout Concurrency sets per-source budgetsβ MiMo V2 6 technical report
Slowersourcesneedmoreconcurrentrolloutstosustainthesame trainingcontributionβ MiMo V2 6 technical report
Theexpectedgenerationdemandis ππ = π΅π/ππβ MiMo V2 6 technical report
2 quotes from 2 sources, one card per model, each opening with its strongest.
The episode generation concurrency adapts following backpressure signals from generators, services, and the trainer, which limits staleness.
Slower sources need more concurrent rollouts to sustain the same training contribution.