# Speculative Decoding: Preserving Exact Greedy Output

## The Core Idea
Speculative decoding uses a small, fast **draft model** to propose several tokens ahead, then verifies them in parallel with the large **target model**. The key question: how does this produce *identical* results to running the target model alone?

## Why It's Exact (Greedy Case)

**1. Verification is cheap and exact.**
After the draft model proposes k tokens, the target model runs a single forward pass over the prompt + draft tokens. Because of causal attention, this one pass yields the target model's logits at