Component-Aware Self-Speculative Decoding in Hybrid Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Borobia, Hector, Seguí-Mas, Elies, Tormo-Carbó, Guillermina
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917454139097088
author Borobia, Hector
Seguí-Mas, Elies
Tormo-Carbó, Guillermina
author_facet Borobia, Hector
Seguí-Mas, Elies
Tormo-Carbó, Guillermina
contents Speculative decoding accelerates autoregressive inference by drafting candidate tokens with a fast model and verifying them in parallel with the target. Self-speculative methods avoid the need for an external drafter but have been studied exclusively in homogeneous Transformer architectures. We introduce component-aware self-speculative decoding, the first method to exploit the internal architectural heterogeneity of hybrid language models, isolating the SSM/linear-attention subgraph as a zero-cost internal draft. We evaluate this on two architecturally distinct hybrid families: Falcon-H1 (parallel: Mamba-2 + attention per layer) and Qwen3.5 (sequential: interleaved linear and attention layers), with a pure Transformer control (Qwen2.5). Parallel hybrids achieve acceptance rates of alpha = 0.68 at draft length k=2 under greedy decoding, while sequential hybrids yield only alpha = 0.038 -- an 18x gap attributable to how each architecture integrates its components. The property is scale-invariant: Falcon-H1 at 3B reproduces the rates observed at 0.5B. We further show that perplexity degradation from a companion ablation study predicts speculative viability without running speculative decoding: a 3.15x ratio (Falcon) maps to alpha = 0.37 at k=4, while 81.96x (Qwen) maps to alpha = 0.019. For sequential hybrids, generic LayerSkip achieves 12x higher acceptance rates than the component-aware strategy. The composition pattern of hybrid models -- not merely the presence of alternative components -- determines whether component-level self-speculation is viable.
format Preprint
id arxiv_https___arxiv_org_abs_2605_01106
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Component-Aware Self-Speculative Decoding in Hybrid Language Models
Borobia, Hector
Seguí-Mas, Elies
Tormo-Carbó, Guillermina
Computation and Language
Artificial Intelligence
I.2.7
Speculative decoding accelerates autoregressive inference by drafting candidate tokens with a fast model and verifying them in parallel with the target. Self-speculative methods avoid the need for an external drafter but have been studied exclusively in homogeneous Transformer architectures. We introduce component-aware self-speculative decoding, the first method to exploit the internal architectural heterogeneity of hybrid language models, isolating the SSM/linear-attention subgraph as a zero-cost internal draft. We evaluate this on two architecturally distinct hybrid families: Falcon-H1 (parallel: Mamba-2 + attention per layer) and Qwen3.5 (sequential: interleaved linear and attention layers), with a pure Transformer control (Qwen2.5). Parallel hybrids achieve acceptance rates of alpha = 0.68 at draft length k=2 under greedy decoding, while sequential hybrids yield only alpha = 0.038 -- an 18x gap attributable to how each architecture integrates its components. The property is scale-invariant: Falcon-H1 at 3B reproduces the rates observed at 0.5B. We further show that perplexity degradation from a companion ablation study predicts speculative viability without running speculative decoding: a 3.15x ratio (Falcon) maps to alpha = 0.37 at k=4, while 81.96x (Qwen) maps to alpha = 0.019. For sequential hybrids, generic LayerSkip achieves 12x higher acceptance rates than the component-aware strategy. The composition pattern of hybrid models -- not merely the presence of alternative components -- determines whether component-level self-speculation is viable.
title Component-Aware Self-Speculative Decoding in Hybrid Language Models
topic Computation and Language
Artificial Intelligence
I.2.7
url https://arxiv.org/abs/2605.01106