in SortedRL's motivating RL setup, rollouts can consume about 70% of total training compute at a 16K maximum generation length

Name: in SortedRL's motivating RL setup, rollouts can consume about 70% of total training compute at a 16K maximum generation length
License: https://aidatacenterguide.com/terms#intellectual-property

~70%observed

Value kind	observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range.
Scope	Setup-specific result from SortedRL, not a universal RLVR compute share.
As of	2026-03-24
Source	SortedRL, OpenReview / ICLR 2026 submission · Section 2.2 and Figure 1, motivating setup at 16K maximum generation length
Review	checking…review by 2026-12-22 · standard cadence
Recorded changes	last 2026-08-24 · 2 revisions tracked
Claim id	of-compute-consumed-by-rollouts-at-16k-token

Where the guide uses it

1.4 Post-Training, Fine-Tuning & RL: The Hybrid Middle

← Full numbers register — every date-stamped figure in the guide, with revision history.