analiza aiqbench
GDN Tree-Scan addresses the verification bottleneck in recurrent-hybrid language models by extending tree speculative decoding to handle Gated-DeltaNet architectures within vLLM. The approach achieves significant throughput gains over native multi-token prediction methods while maintaining near-native latency, offering a practical path forward for deploying hybrid model families at scale.
GDN Tree-Scan: Served Tree Verification for Recurrent-Hybrid Language Models
arXiv:2609.23900v1 Announce Type: new Abstract: Tree speculative decoding verifies multiple candidate continuations in one target forward pass. For attention-only transformers, the verifier mainly needs an ancestry mask. Recurrent-hybrid language models break this assumption: a candidate row must also carry the recurrent state that native sequential decode would have produced along its root-to-node path. Otherwise, a verifier can use a correct attention mask while still conditioning on an impossible recurrent history. We present GDN Tree-Scan, a served verifier for Gated-DeltaNet hybrid language models integrated into vLLM. The system combines FlashAttention-2 tree-bias attention, branch-local GDN scan/replay, device-side multidraft commitment, and accepted-chain-only state publication. On the public Qwen3.6-27B-FP8 checkpoint, in a clean batch-one (B=1) SWE/Codex decode gate at temperature 0.6, a six-node root-branch tree increases committed tokens/event by 17.2% at near-native verify-forward time and reaches 23.88 token-weighted decode tokens/s versus 18.80 for native five-step MTP (E5), a 27.0% token-weighted decode-throughput gain. The per-request-equal latency view is +4.0%, and end-to-end task wall time remains prefill-heavy. Empirical equivalence evidence is scoped to recurrent-oracle probability-rescore (p-rescore) closure within the observed native flip floor, not a full distribution-distance proof.
Abstract:Tree speculative decoding verifies multiple candidate continuations in one target forward pass. For attention-only transformers, the verifier mainly needs an ancestry mask. Recurrent-hybrid language models break this assumption: a candidate row must also carry the recurrent state that native sequential decode would have produced along its root-to-node path. Otherwise, a verifier can use a correct attention mask while still conditioning on an impossible recurrent history. We present GDN Tree-Scan, a served verifier for Gated-DeltaNet hybrid language models integrated into vLLM. The system combines FlashAttention-2 tree-bias attention, branch-local GDN scan/replay, device-side multidraft commitment, and accepted-chain-only state publication. On the public Qwen3.6-27B-FP8 checkpoint, in a clean batch-one (B=1) SWE/Codex decode gate at temperature 0.6, a six-node root-branch tree increases committed tokens/event by 17.2% at near-native verify-forward time and reaches 23.88 token-weighted decode tokens/s versus 18.80 for native five-step MTP (E5), a 27.0% token-weighted decode-throughput gain. The per-request-equal latency view is +4.0%, and end-to-end task wall time remains prefill-heavy. Empirical equivalence evidence is scoped to recurrent-oracle probability-rescore (p-rescore) closure within the observed native flip floor, not a full distribution-distance proof.
Machine Learning (cs.LG)
arXiv:2609.23900 [cs.LG]
(or arXiv:2609.23900v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.23900 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 271 cuvinte
Subiecte conexe
Bole: Efficient Tree Speculation for Hybrid-Attention Language Models
arXiv · 4 aug. 2026
TreeWY: Speculative Verification for Gated DeltaNet Hybrids
arXiv · 24 aug. 2026
H-Spec: Parallel Speculative Decoding Without a Drafter-Side KV Cache
arXiv · 22 sept. 2026
Lattice-Based Cryptography Explained
Quantum Zeitgeist · 15 aug. 2026