Insights from the Yannic Kilcher episode “TiDAR: Think in Diffusion, Talk in Autoregression (Paper Analysis)”, published December 27, 2025.
Tidar, a hybrid architecture from Nvidia, leverages unused GPU capacity during autoregressive inference to run parallel diffusion-based drafts. By using these drafts as speculative proposals for the autoregressive model, the system achieves a 4-6x speed boost without sacrificing the output quality of standard autoregressive decoding.
Topics: LLM, Nvidia, Inference, Speculative Decoding, Diffusion Models
