VARG Deploys Generative Retrieval in Tmall Search With Value-Aware Item IDs
_reachsumit · x · 2026-09-15
VARG is a generative retrieval system for Tmall App search that admits generated item candidates straight into the final ranker, skipping pre-ranking:
- VARG-ID builds semantic prefixes with RQ-VAE, uses bidirectional query-item contrastive learning, and adds a value-ordered third token encoding fine-grained item addresses and business-value priors.
- Three-stage SFT learns item-to-identifier mapping, query-semantic retrieval, then personalized retrieval; LO-SFT learns local within-cluster ordering, and Prefix-GRPO aligns generation with business value via gated rewards.
- Daily coordinated product/model updates keep existing item addresses while incorporating new products and behavior feedback.
- Validated offline on tens of millions of products.
More from Research
- General Reasoning explains how to train agent swarms with RL and latency pressure — rosstaylor90 · 2026-11-02
- Mathematician Tony Feng: current-paradigm LLMs lack the diversity of thought research needs — littmath · 2026-11-02
- Measure Theory vs Probability: 'You're Confusing the Finger With the Moon' — LucaAmb · 2026-11-02
- DStill Distills Multi-Step Denoising Into One Pass, 6.45x Speedup for Diffusion Speculative Decoding — LucaAmb · 2026-11-02
- Radial gate objective cuts diffusion transformer FID from 16.24 to 11.83 — Junha Park · 2026-11-02
- REA: async brain-cerebellum runtime for robots where policies never wait on the LLM — Pengfei Zhou · 2026-11-02