AI

Towards Efficient Whole Page Ranking with One-Pass Inference

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/11/2026, 05:03:24 UTC.

Briefings

A new framework called Multi-Item Scoring (MIS) proposes prefix-aware NAR-style inference for whole-page ranking, aiming to balance generation quality with efficiency. The study tested MIS on KuaiRand-1K and RL4RS datasets, showing it matches autoregressive CGA baseline quality while achieving 4.07x faster CPU decoding. On a social-video platform, MIS improved watch time by 0.058% and reduced ranking-service CPU consumption by 9.73%. The research addresses the gap between non-autoregressive efficiency and autoregressive quality in industrial applications.

Key points

• MIS is a prefix-aware NAR-style framework that integrates global candidate interactions, localized prefix conditioning, and position-aware scoring within a shared representation architecture.
• Experiments on KuaiRand-1K and RL4RS demonstrate that MIS approaches the generation quality of an autoregressive CGA baseline while achieving 4.07x and 3.54x faster CPU decoding, respectively.

Context

This research addresses the challenge in industrial whole-page ranking where existing non-autoregressive (NAR) approaches enable efficient page generation but typically score candidate items independently of previously selected items, limiting their ability to capture order-dependent interactions. Autoregressive (AR) methods address this limitation through sequential conditioning, but incur substantial inference overhead that challenges large-scale deployment. The study bridges this quality-efficiency gap using Multi-Item Scoring (MIS).

Limitations

The study was conducted on two specific datasets (KuaiRand-1K and RL4RS) and a social-video platform. Results are limited to these specific contexts and do not generalize to other ranking systems or platforms without further validation.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 05/11/2026, 05:18:48 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeic6bzgifzn2i2ycyf4wejenlcscgn7wzaxrng72l63lec6frhhn6y