EXPLORE THE ARCHIVE

Bo Zheng

Publications and source records attributed to Bo Zheng.

5 recordsLinked to original sources

arXiv2026

Pailitao-MMSearch: Building Native E-Commerce Multimodal Search Foundation

The evolution of e-commerce has fundamentally transformed how users search for products, shifting from simple text-based keyword queries to complex multimodal interactions that seamlessly combine product images, natural language descriptions, and mixed-intent instructions. However, existing approaches face a critical dilemma: single-modal specialist models, deployed independently for text retrieval, visual search, and voice recognition, operate in isolation and cannot handle cross-modal queries, while general-purpose vision-language models lack the domain-specific knowledge necessary for fine-grained product understanding, user behavior modeling, and commercial intent reasoning. In this work, we present Pailitao-MMSearch, one native e-commerce multimodal search foundation model designed to bridge this gap. Our approach introduces three key innovations: (1)HybSID (Hybrid Semantic ID);(2)a two-stage continual pre-training strategy; and (3)a hybrid reasoning post-training pipeline. Built upon Qwen and deployed on Taobao's Pailitao multimodal search platform, Pailitao-MMSearch achieves substantial improvements in online A/B testing, including up to +13.61\% in Gross Merchandise Volume (GMV) and +8.21\% in transaction volume compared to traditional multi-modal search pipeline, demonstrating the effectiveness of our native e-commerce multimodal search large language models.

cs.AI↗

arXiv2026

SPARC: Sequence-aware Progressive Attribute Routing and Compression Framework for Generative Recommendation

Generative recommendation tokenizes items as discrete Semantic IDs (SIDs) and autoregressively generates target items from users' historical SID sequences. Although existing SIDs incorporate multimodal and structured information, they are typically statically assigned and independent of the current interaction context. In industrial scenarios, each behavior also contains heterogeneous attributes, such as category, brand, price, behavior type, and timestamp. Fully expanding these features greatly increases the input length, while directly compressing them into a single representation may prematurely discard context-relevant information. We propose \textbf{SPARC}, \uline{\textbf{S}}equence-aware \uline{\textbf{P}}rogressive \uline{\textbf{A}}ttribute \uline{\textbf{R}}outing and \uline{\textbf{C}}ompression Framework for Generative recommendation. SPARC first models the sequential dependencies of each field type to obtain context-aware field representations. It then routes the original, contextual, and identity representations of different fields into multiple slots to preserve complementary information under a fixed capacity. Finally, lightweight cross-item interaction integrates the intermediate tokens and compresses each historical item into a single token. Following the principle of contextualizing before compression, SPARC enriches user-history representations without increasing the input length of the generative backbone. Experiments on industrial Taobao and public Amazon datasets demonstrate that SPARC outperforms strong conventional and generative baselines. Further comparisons with static compression variants show that the improvement of SPARC comes from context-conditioned information retention rather than merely increasing the expressiveness of the compression module.

cs.IR↗

arXiv2026

On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability

We describe the architecture and ablations of Qwen3.8-Flash-Next, a sparse mixture-of-experts model with 125B parameters, 6B activated per token, and additional 51B parameters of n-gram embedding tables held off the accelerator. On fourteen pre-training benchmarks the model leads the 397B-A17B predecessor on eight and trails it on the rest by at most 2.6 points, at 1/3 the activated parameters, 1/3 the training tokens, and roughly 1/9 the training FLOPs. Token mixing uses a layer-wise hybrid of Gated DeltaNet (GDN) and global attention, with one full-attention layer in every four; at continued-pretraining time those full-attention layers are replaced by Qwen Sparse Attention (QSA), which scores context at micro-block granularity with a compressed lightweight indexer. The residual stream is widened to four branches and read through an elementwise gate, a design we call the Gated Residual (GR). Capacity is added outside the backbone by a single n-gram embedding layer whose tables are prefetched from host memory. We evaluate every candidate change along three axes: loss together with downstream benchmarks; the cost of the change in training, prefill and decode; and its effect on the optimal hyperparameters and training stability. Loss and downstream accuracy do not always move together: enlarging the n-gram vocabulary lowers loss monotonically while downstream accuracy saturates. The architecture and the Muon optimizer together shift the optimal learning rate and batch size upwards, render batch-size warmup unnecessary, and substantially improve stability under stress tests. Loss, benchmarks, efficiency and stability form one design problem. Solved jointly, they yield a recipe that is simultaneously more efficient, more capable and more stable.

cs.CL↗

arXiv2026

Self-OPD: On-Policy Distillation for Flow Matching Models without Teacher

On-policy distillation (OPD), which leverages a pre-trained, specialized teacher model to provide dense supervisory signals, has achieved significant success in Large Language Models (LLMs) and has recently been adapted to flow matching models. However, this paradigm suffers from two major issues: First, training a separate, task-specific teacher for every new objective incurs high computational costs. Second, the discrepancy between teacher and student distributions often leads to compounding errors along the generation trajectory. In this paper, we introduce \textbf{Self-OPD}, a teacher-free OPD framework for flow matching models that turns the student's own self-exploration into step-wise supervision. At each timestep, Self-OPD branches the deterministic next-state prediction into $K$ stochastic SDE candidates, rolls them out with the ODE sampler, and compares their rewards against a deterministic self-reference baseline to obtain normalized advantages. The velocity field is optimized with an all-branch pull-push objective, where high-advantage branches attract the student and low-advantage branches repel it under direction-aware attenuation and SDE-variance normalization. For multi-objective alignment, Self-OPD fuses normalized scores at the reward level, avoiding direct gradient conflict. Experiments on single and mixed reward benchmarks show that Self-OPD outperforms prior RL and OPD methods without task-specific teachers.

cs.CV↗

arXiv2026

Beyond Semantic IDs: Encoding Business-Value Ranking into Document Identifiers for Generative Retrieval

Generative Retrieval (GR) formulates retrieval as a sequence-to-sequence generation task, assigning each document a document identifier (DocID) and retrieving it through autoregressive decoding, making DocID design a critical factor in retrieval quality. However, existing schemes based on discrete representation learning suffer from inherent collision issues and create a mismatch between the DocID's encoding objective and the system's business optimization target. To address these limitations, we propose \textbf{Cluster-Ranked Identifier (CRID)}, which decouples DocID into \textit{semantic clustering} and \textit{business-value ranking}, yielding collision-free identifiers that support incremental updates via intra-cluster reranking. We further introduce an analytical framework that decomposes retrieval gains into \textit{personalized preference} and \textit{statistical prior} generalization, revealing how semantic cluster size governs the balance between the two components. Experiments on a Taobao e-commerce corpus of over 300M items show that CRID surpasses the strongest embedding-based retrieval baseline on top-K Hitrate, and delivers +1.06\% GMV in full-traffic deployment.

cs.IR↗