Hydra: Sequentially-Dependent Draft Heads for Medusa Decoding (arXiv:2402.05109)

Fields

id

sources/mkt-hydra-draft-heads

URL

https://arxiv.org/abs/2402.05109

Publisher

arXiv

Kind

paper

Published

2024-02-07

Accessed

2026-10-04

notes

arXiv v2 (Oct 2024). Proposes sequentially-dependent draft heads for Medusa-style decoding, improving over standard independent draft heads. Hydra++ improves decoding throughput by up to 1.31x and 2.70x compared to Medusa decoding and autoregressive decoding respectively. Code at https://github.com/zankner/Hydra. No venue listed on arXiv; likely preprint.

Notes

arXiv v2 (Oct 2024). Proposes sequentially-dependent draft heads for Medusa-style decoding, improving over standard independent draft heads. Hydra++ improves decoding throughput by up to 1.31x and 2.70x compared to Medusa decoding and autoregressive decoding respectively. Code at https://github.com/zankner/Hydra. No venue listed on arXiv; likely preprint.

Cross-references

This record references no other records.