# Speculative Decoding FT: Draft Model + EAGLE-2 + MEDUSA Head Training

> Source: https://sukruyusufkaya.com/en/learn/fine-tuning-cookbook/ftc-speculative-decoding-ft-eagle-medusa
> Updated: 2026-05-14T14:42:59.907Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part XIII — Custom Kernels & Performance Surgery
**TLDR:** FT version of speculative decoding: pair draft model with target, maximize accept rate. EAGLE-2 head training (Li et al. 2024, +94% throughput), MEDUSA multi-head training, training extra heads while target frozen. Llama 8B target + MEDUSA 4-head ~2-3h training on RTX 4090.