Single-Completion Training
An optimization approach that trains using only single-completion samples, which can reduce the computational overhead associated with multi-sample generation.
Timeline (last 90 days)
@k pressure can be fully realized through single-completion training, matching the performance of multi-completion @k optimizers at only a fraction of their cost.
Single-completion training means that at each policy update, only one completion is sampled per prompt, yet the learning pressure of the @k objective is simulated through reward transformation or regularization design.
All Facts (2)
@k pressure can be fully realized through single-completion training, matching the performance of multi-completion @k optimizers at only a fraction of their cost.
50%UnverifiedSingle-completion training means that at each policy update, only one completion is sampled per prompt, yet the learning pressure of the @k objective is simulated through reward transformation or regularization design.
50%