inference/research
verifiedhigh · 0.88updated 2026-10-11

arXiv 2605.10453: SlimSpec: Low-Rank Draft LM-Head for Accelerated Speculative Decoding

paper

Fields

id
sources/w15-pap-slimspec
Publisher
arXiv
Kind
paper
Published
2026-05-11
Accessed
2026-10-11
notes
Paper by Anton Plaksin, Sergei Krutikov, Sergei Skvortsov, Alexander Samarin (Nebius). Investigates draft LM-head projection bottleneck in auxiliary-head drafters (EAGLE-3). Evaluates low-rank LM head compression across batch 1 and batch 64 regimes on LLaMA 3.1 8B, GPT-OSS 20B, and Qwen3-30B-A3B.

Notes

Paper by Anton Plaksin, Sergei Krutikov, Sergei Skvortsov, Alexander Samarin (Nebius). Investigates draft LM-head projection bottleneck in auxiliary-head drafters (EAGLE-3). Evaluates low-rank LM head compression across batch 1 and batch 64 regimes on LLaMA 3.1 8B, GPT-OSS 20B, and Qwen3-30B-A3B.

Cross-references

This record references no other records.