Generation Meets Verification: Accelerating Large Language Model Inference with Smart Parallel Auto-Correct Decoding

Fields

id

sources/sd-space-paper

URL

https://arxiv.org/abs/2402.11809

Publisher

arXiv (Yi et al.)

Kind

paper

Published

2024-02

Accessed

2026-10-04

notes

SPACE (Smart Parallel Auto-Correct dEcoding) integrates semi-autoregressive inference with speculative decoding. Speedup 2.7x-4.0x on HumanEval-X. Uses specialized semi-autoregressive SFT to enable simultaneous multi-token prediction, plus auto-correct decoding for parallel generation and verification. Accepted at ACL 2024 Findings.

Notes

SPACE (Smart Parallel Auto-Correct dEcoding) integrates semi-autoregressive inference with speculative decoding. Speedup 2.7x-4.0x on HumanEval-X. Uses specialized semi-autoregressive SFT to enable simultaneous multi-token prediction, plus auto-correct decoding for parallel generation and verification. Accepted at ACL 2024 Findings.

Cross-references

This record references no other records.