Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules
Open paper
Benchmark
Model
Rank
Results
sentence-completion-on-hellaswag
Dream (Instruct) with SchED (Exp k=2, τhigh=7.5, τlow=2.5)
#74
Accuracy: 80.21
arithmetic-reasoning-on-gsm8k
Dream (Base) with SchED (Exp k=2, τhigh=7.5, τlow=2.5)
#171
Accuracy: 77.71
Rank counts only results with a code link.