Latent Reasoning with Normalizing Flows
Open paper
Benchmark
Model
Rank
Results
code-generation-on-humaneval
NF-CoT (Unified) + RL
–
Pass@1: 86.7
code-generation-on-mbpp
NF-CoT (Unified) + RL
–
Accuracy: 85.4
Rank counts only results with a code link.