ByteDance's OneTrans-GSR unifies reranking and mix-ranking with one AR transformer
_reachsumit · x · 2026-11-05
- ByteDance researchers propose OneTrans-GSR, reformulating the separate reranking and mix-ranking stages (mixing product cards, short videos, livestreams, and ad cards) as a single generative slate-ranking paradigm.
- Rule-based rerankers and low-capacity specialized models constrain the combinatorial slate space and require manual objective tuning; OneTrans-GSR uses a single decoder-only policy, extending the deployed OneTrans backbone, to generate the exposure slate end-to-end.
- A reward model evaluates complete slates during training and guides reinforcement learning toward slate-level optimization, making performance scalable with compute.
More from Research
- ProteinMPNN memorizes transmembrane folds: noise may trick it, devs speculate — sokrypton · 2026-11-05
- Four generations of protein design models chart the road to fully programmable protein design — AllThingsApx · 2026-11-05
- ProteinMPNN likely memorized transmembrane folds; 0.3 noise breaks the recognition — sokrypton · 2026-11-05
- 10 AI agents in a virtual town wipe each other out in 4 days; 3 house-share agents just melt into puddles — repligate · 2026-11-05
- lm-eval Pitfall: Chatty Models Break Letter-Only Prompts, Regex Grabs Wrong Answers — mancubus77 · 2026-11-05
- Evolution can't copy homework: why species improve slower than iPhones — louisvarge · 2026-11-05