Lookahead Decoding
A draft-model-free parallel LLM decoding method that applies Jacobi decoding principles to generate and verify multiple candidate tokens in parallel, thereby accelerating autoregressive inference.
Timeline (last 90 days)
Lookahead Decoding is based on the parallel decoding idea of Jacobi iteration, reformulating autoregressive sequence generation as the problem of solving a system of nonlinear equations.
Eagle improves acceptance rates through feature-level prediction, while Lookahead Decoding avoids using an independent draft model through Jacobi iteration.
Speculative decoding has multiple variants, including Medusa, EAGLE/EAGLE-2, Lookahead Decoding, and SpecInfer, each with different trade-offs in acceptance rate, complexity, VRAM overhead, and architectural invasiveness.
All Facts (3)
Speculative decoding has multiple variants, including Medusa, EAGLE/EAGLE-2, Lookahead Decoding, and SpecInfer, each with different trade-offs in acceptance rate, complexity, VRAM overhead, and architectural invasiveness.
70%UnverifiedLookahead Decoding is based on the parallel decoding idea of Jacobi iteration, reformulating autoregressive sequence generation as the problem of solving a system of nonlinear equations.
50%UnverifiedEagle improves acceptance rates through feature-level prediction, while Lookahead Decoding avoids using an independent draft model through Jacobi iteration.
50%