← 研踪 PaperTrailarXiv:2611.22309 · v1
Efficient Wide-Beam Generative Retrieval with Model-Native Trie Decoding and Shared-Context Attention
首次提交 2026-11-17 15:07
版本更新 2026-11-17 15:07
雷达观测 2026-11-18 19:43
摘要
Generative retrieval (GR) predicts catalog entries as short semantic identifiers (SIDs). Trie-constrained beam search keeps these identifiers valid, but wide beams expose two serving costs: vocabulary-wide score tensors that are mostly discarded, and repeated attention over the same prompt. We present Trie-Head, a model-native output interface that stores the identifier trie as validated GPU tensors. Compared with a Python custom logit processor (CLP), it reduces CPU--GPU coordination and memory traffic by avoiding Python constraint callbacks and vocabulary-wide candidate buffers. It scores compact codebook candidates and streams the full-vocabulary normalizer, preserving the original model's valid-token scores in real arithmetic without retraining. Shared-context attention complements this output path by reusing prompt key/value blocks across beams while preserving their private decode tails. We implement both mechanisms in SGLang and evaluate candidate sourcing for LinkedIn Hiring Assistant and qualification retrieval for the Qualification Retrieval Platform. The combined system improves closed-loop throughput by up to 5.57x over the CLP baseline and up to 2.63x over BF16 FlashRec at a common beam width of 200 on the Qualification Retrieval Platform workload.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-17 15:07 · 最新观测版本
Efficient Wide-Beam Generative Retrieval with Model-Native Trie Decoding and Shared-Context Attention
Generative retrieval (GR) predicts catalog entries as short semantic identifiers (SIDs). Trie-constrained beam search keeps these identifiers valid, but wide beams expose two serving costs: vocabulary-wide score tensors that are mostly discarded, and repeated attention over the same prompt. We present Trie-Head, a model-native output interface that stores the identifier trie as validated GPU tensors. Compared with a Python custom logit processor (CLP), it reduces CPU--GPU coordination and memory traffic by avoiding Python constraint callbacks and vocabulary-wide candidate buffers. It scores compact codebook candidates and streams the full-vocabulary normalizer, preserving the original model's valid-token scores in real arithmetic without retraining. Shared-context attention complements this output path by reusing prompt key/value blocks across beams while preserving their private decode tails. We implement both mechanisms in SGLang and evaluate candidate sourcing for LinkedIn Hiring Assistant and qualification retrieval for the Qualification Retrieval Platform. The combined system improves closed-loop throughput by up to 5.57x over the CLP baseline and up to 2.63x over BF16 FlashRec at a common beam width of 200 on the Qualification Retrieval Platform workload.
查看此版本解读 →查看此版本原文 ↗