Pith. sign in

REVIEW 13 cited by

Hydra: Sequentially-Dependent Draft Heads for Medusa Decoding

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2402.05109 v2 pith:UR7ZX7AY submitted 2024-02-07 cs.LG

classification cs.LG
keywords draftheadsdecodinghydramodelcandidateheadmedusa
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

To combat the memory bandwidth-bound nature of autoregressive LLM inference, previous research has proposed the speculative decoding frame-work. To perform speculative decoding, a small draft model proposes candidate continuations of the input sequence that are then verified in parallel by the base model. One way to specify the draft model, as used in the recent Medusa decoding framework, is as a collection of lightweight heads, called draft heads, that operate on the base model's hidden states. To date, all existing draft heads have been sequentially independent, meaning that they speculate tokens in the candidate continuation independently of any preceding tokens in the candidate continuation. In this work, we propose Hydra heads: a sequentially-dependent drop-in replacement for standard draft heads that significantly improves the accuracy of draft head speculation. We further explore the design space of Hydra head training objectives and architectures, and propose a carefully tuned Hydra head recipe, which we call Hydra++, that improves decoding throughput by up to 1.31x and 2.70x compared to Medusa decoding and autoregressive de-coding respectively. Overall, Hydra heads are a simple and well-motivated intervention on standard draft heads that significantly improve the end-to-end speed of draft head-based speculative decoding. We make our code publicly available at https://github.com/zankner/Hydra.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 13 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Judge Decoding: Faster Speculative Sampling Requires Going Beyond Model Alignment

    cs.LG 2025-01 conditional novelty 7.0 of 10

    Judge decoding replaces strict token-alignment verification in speculative decoding with a learned correctness classifier on target embeddings, accepting about three times more tokens and enabling up to 9x speedup on ...

  2. SpecLA: Efficient Speculative Decoding for Linear-Attention Models

    cs.CL 2026-07 conditional novelty 6.0 of 10

    SpecLA speculatively decodes Gated DeltaNet linear-attention models with topology-aware verification and factor-buffer state recovery, reaching up to 1.70x end-to-end speedup over autoregressive decoding on GDN-1.3B.

  3. Unlocking Parallelism in Autoregressive Language Models via Speculative Decoding with Progressive Tree Drafting

    cs.CL 2026-07 accept novelty 6.0 of 10

    PTD accelerates autoregressive LLM decoding up to ~2× by guiding the target model to explore multiple coherent draft paths via a progressive, pruned tree in a single forward pass.

  4. Training-Free Tokenizer Transplantation via Orthogonal Matching Pursuit

    cs.CL 2025-06 conditional novelty 6.0 of 10

    OMP sparse coding of donor token embeddings, with coefficients transferred to the base embedding space, preserves LLM performance after tokenizer replacement better than published zero-shot baselines, though simple he...

  5. Jakiro: Boosting Speculative Decoding with Decoupled Multi-Head via MoE

    cs.CL 2025-02 conditional novelty 6.0 of 10

    Jakiro speeds up LLM inference by using MoE-based draft heads to decouple candidate predictions in speculative decoding trees, plus a contrastive parallel decoding stage for the last draft steps.

  6. Lossless Acceleration of Large Language Models with Hierarchical Drafting based on Temporal Locality in Speculative Decoding

    cs.CL 2025-02 conditional novelty 6.0 of 10

    Hierarchy Drafting accelerates LLM inference without retraining by retrieving draft tokens from three databases in order of temporal locality, beating prior database drafting methods on Spec-Bench.

  7. DeLS-Spec: Decoupled Long-Short Contexts for Parallel Speculative Drafting

    cs.CL 2026-07 conditional novelty 5.0 of 10

    DeLS-Spec improves block-parallel speculative decoding by fusing DFlash logits with an independently trained lightweight local head and a unigram prior correction.

  8. Your LLM Knows the Future: Uncovering Its Multi-Token Prediction Potential

    cs.CL 2025-07 conditional novelty 5.0 of 10

    A gated LoRA and a small sampler let an autoregressive LLM draft multiple future tokens per step, and self-speculative verification converts those drafts into up to roughly 5x fewer generation steps.

  9. Faster and Better LLMs via Latency-Aware Test-Time Scaling

    cs.CL 2025-05 conditional novelty 5.0 of 10

    Test-time scaling should be tuned for wall-clock latency rather than token count; branches plus speculative decoding win under time budgets.

  10. PLD+: Accelerating LLM inference by leveraging Language Model Artifacts

    cs.CL 2024-12 conditional novelty 5.0 of 10

    PLD+ accelerates LLM inference on input-guided tasks by ranking prompt-derived draft spans with hidden states or attention heads, beating tuning-free baselines and often surpassing the tuned EAGLE method.

  11. Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration

    cs.LG 2024-11 conditional novelty 5.0 of 10

    A CTC-based draft model with sequence-level training and a CTC transform for candidate cleaning improves accepted-token counts and inference speedup in speculative decoding of LLMs.

  12. S$^4$C: Speculative Sampling with Syntactic and Semantic Coherence for Efficient Inference of Large Language Models

    cs.CL 2025-06 conditional novelty 4.0 of 10

    S4C accelerates LLM generation by combining multi-head autoregressive draft heads with a continuous verification tree, measuring 2.26x to 2.60x speedups on Spec-bench.

  13. Closer Look at Efficient Inference Methods: A Survey of Speculative Decoding

    cs.CL 2024-11 conditional novelty 2.0 of 10

    A survey that categorizes speculative decoding methods into draft-centric and model-centric families and discusses deployment challenges.

Pith tools