LoopCD: training-free contrastive decoding for Looped Transformers
Radar: the paper presents LoopCD, a training-free contrastive decoding method for recurrent Transformers that uses representations from earlier loops as weaker predictions to select tokens.
According to the paper logged in Radar on October 2, 2026, LoopCD is a training-free contrastive decoding method for Looped Transformers. It uses representations from earlier loops, which provide aligned weaker predictions, to guide token selection during generation.
For engineers working on inference with recurrent Transformers, the central point is evaluating a decoding method that draws on intermediate states without additional training. The source gives no numerical results, benchmarks or implementation details. To verify performance and applicability, consult the original paper using the title and date recorded here, and check its experiments and evaluation conditions.