2 papers
cs.CL2025
RAD: Redundancy-Aware Distillation for Hybrid Models via Self-Speculative Decoding
Yuichiro Hoshino, Hideyuki Tachibana, Muneyoshi Inahara +1
Hybrid models combining Transformers and State Space Models (SSMs) are promising for balancing performance and efficiency. However, optimizing these hybrid models, particularly by…
stat.ML2021
Quasi-Taylor Samplers for Diffusion Generative Models based on Ideal Derivatives
Hideyuki Tachibana, Mocho Go, Muneyoshi Inahara +2
Diffusion generative models have emerged as a new challenger to popular deep neural generative models such as GANs, but have the drawback that they often require a huge number of n…