arXiv subjects

Zhaolong Xing

Publications and source records attributed to Zhaolong Xing.

7 recordsLinked to original sources

arXiv2026

Remember by Asking: Retrieval-Induced Memory Evolution for LLM Agents

Long-term memory is essential for language agents to maintain coherent and effective behavior over extended, multi-session interactions. Existing memory systems mainly use retrieval at read time, while write-time memory formation still relies on direct extraction or compression. However, when future information needs are unknown, compressing an entire interaction in one pass can overlook locally important details that may matter later. To this end, we introduce RIME, a retrieval-induced memory framework that shifts memory construction from monolithic compression toward evidence-centered integration. RIME uses generic self-questions to retrieve focused dialogue evidence and grounds memory formation in both the retrieved evidence and relevant historical memories, which are jointly reconciled into an evolving memory bank with temporal and provenance information. At inference time, compressed memory serves as the primary rather than the sole source of evidence: when it cannot support an answer, RIME retrieves relevant source dialogue together with its local context to recover information omitted during memory formation, without resorting to full-history processing. Extensive experiments on LoCoMo with Qwen3-235B-A22B and GPT-5.6 Sol show that RIME consistently achieves the best performance across all three quality metrics among the compared methods, while requiring substantially fewer query-time LLM tokens.

cs.CL↗

arXiv2026

Turning Off-Policy Tokens On-Policy: A Plug-in Approach for Improving LLM Alignment

Reinforcement learning (RL) post-training for large language models (LLMs) follows a efficient paradigm of "rollout then update", which inevitably results in off-policy training data. To resolve this, Importance sampling (IS) is proposed, while the token-level ratios compound over long sequences, causing severe variance exploded. A natural idea is "transferring" these off-policy token into on-policy token, so that the importance scores for correction are unnecessary. Following this idea, we propose Selective Importance Sampling (SIS), which is inspired by rejection sampling. Concretely, SIS implements by viewing off-policy model as proposal distribution, and implement a token-level rejection test: accepted tokens are viewed as on-policy, so that receive unit importance score, while rejected tokens retain the standard IS correction. Our proposed SIS is theoretically proved reducing the gap between token-level and sequence-level off-policy gradient estimators. The SIS acts as a plug-in that only modifies the importance ratio in the policy loss, adding negligible wall-clock overhead, and can be combine with a vast vary of RL post-training algorithms. Experiments on dense and MoE LLMs across math and agent benchmarks show that SIS consistently improves all objectives, while providing substantially stronger robustness under off-policy data.

cs.CL↗

arXiv2026

Averaged Evaluation Masks Capability Trade-Offs: Multi-Source Calibration for High-Sparsity LLM Pruning

Calibration data are often treated as a minor implementation detail in post-training LLM pruning because averaged evaluations suggest only modest effects. We show that this conclusion is an averaging artifact: at 60\% SparseGPT sparsity, calibration strategies separated by only 2.85 points in averaged commonsense accuracy differ by 51.9 points in Code retention. Across 15 sources, capability-decomposed analysis reveals an opposing pattern: calibration perplexity is positively associated with General retention but negatively associated with Math or Code retention, leaving no evaluated single source uniformly strong across capabilities. This finding motivates capability-balanced multi-source calibration. Under the same calibration budget, a balanced real-data mixture outperforms every evaluated single source on LLaMA-3.1-8B, beating C4 by 18.8 points; the advantage grows with sparsity and persists on LLaMA-3.1-70B. Because the original pretraining data of advanced LLMs are often inaccessible, we further introduce Information-Guided Self-Calibration for Pruning (IGSP). Using only the base model and evaluation taxonomy, IGSP generates capability-stratified pools and selects low-redundancy samples within capability-specific perplexity ranges, outperforming Self-Cal and SGS by up to 4.8 points. Together, these results recast calibration as a capability-coverage problem and identify multi-source design as a practical principle for preserving capabilities in high-sparsity LLM pruning.

cs.LG↗

arXiv2026

FlowTrain: Flow-Based Decoupled Training for Industrial-Grade Vision-Language Models

Industrial-grade distributed training of vision-language models (VLMs) remains far less efficient than that of unimodal LLMs. Existing solutions either follow a monolithic design that assigns uniform parallelism to heterogeneous modules or adopt a disaggregated deployment that separates modules while executing them as a batch-synchronized pipeline. In this paper, we highlight that the above solutions are still not sufficient, and VLM training can be further decoupled. To this end, we present FlowTrain, a flow-based decoupled training framework that reformulates VLM training as a producer-consumer dataflow coordinated through a unified memory pool. The encoder and backbone can progress independently over a global virtual address space. Since this execution decoupling fundamentally changes the optimization objective of allocation and scheduling, FlowTrain further introduces a heterogeneous parallel allocator that assigns module-specific parallelism strategies by solving a throughput matching problem. The dynamic packing scheduler is used to construct balanced microbatches at runtime according to the actual LLM-side computation cost. Extensive experiments on real-world workloads show that FlowTrain achieves over 50% MFU and up to 1.7x throughput improvement, narrowing the efficiency gap to LLM-only training.

cs.LG↗

arXiv2026

NestPipe: Large-Scale Recommendation Training on 1,500+ Accelerators via Nested Pipelining

Modern recommendation models have increased to trillions of parameters. As cluster scales expand to O(1k), distributed training bottlenecks shift from computation and memory to data movement, especially lookup and communication latency associated with embeddings. Existing solutions either optimize only one bottleneck or improve throughput by sacrificing training consistency. This paper presents NestPipe, a large-scale decentralized embedding training framework that tackles both bottlenecks while preserving synchronous training semantics. NestPipe exploits two hierarchical sparse parallelism opportunities through nested pipelining. At the inter-batch level, Dual-Buffer Pipelining (DBP) constructs a staleness-free five-stage pipeline through dual-buffer synchronization, mitigating lookup bottlenecks without embedding staleness. At the intra-batch level, we identify the embedding freezing phenomenon, which inspires Frozen-Window Pipelining (FWP) to overlap All2All communication with dense computation via coordinated stream scheduling and key-centric sample clustering. Experiments on production GPU and NPU clusters with 1,536 workers demonstrate that NestPipe achieves up to 3.06x speedup and 94.07% scaling efficiency.

cs.DC↗

arXiv2026

Rollout-Training Co-Design for Efficient LLM-Based Multi-Agent Reinforcement Learning

Despite algorithm-level innovations for multi-agent reinforcement learning (MARL), the underlying networked infrastructure for large-scale MARL training remains underexplored. Existing training frameworks primarily optimize for single-agent scenarios and fail to address the unique system-level challenges of MARL, including rollout-training synchronization barriers, rollout load imbalance, and training resource underutilization. To bridge this gap, we propose FlexMARL, the first end-to-end training framework that holistically optimizes rollout, training, and their orchestration for large-scale LLM-based MARL. Specifically, FlexMARL introduces the joint orchestrator to manage data flow under the rollout-training disaggregated architecture. Building upon the experience store, a novel micro-batch driven asynchronous pipeline eliminates the synchronization barriers while providing strong consistency guarantees. Rollout engine adopts a parallel sampling scheme combined with hierarchical load balancing, which adapts to skewed inter/intra-agent request patterns. Training engine achieves on-demand hardware binding through agent-centric resource allocation. The training states of different agents are swapped via unified and location-agnostic communication. Empirical results on a large-scale production cluster demonstrate that FlexMARL achieves up to 7.3x speedup and improves hardware utilization by up to 5.6x compared to existing frameworks.

cs.LG↗

arXiv2025

OxygenREC: An Instruction-Following Generative Framework for E-commerce Recommendation

Traditional recommendation systems suffer from inconsistency in multi-stage optimization objectives. Generative Recommendation (GR) mitigates them through an end-to-end framework; however, existing methods still rely on matching mechanisms based on inductive patterns. Although responsive, they lack the ability to uncover complex user intents that require deductive reasoning based on world knowledge. Meanwhile, LLMs show strong deep reasoning capabilities, but their latency and computational costs remain challenging for industrial applications. More critically, there are performance bottlenecks in multi-scenario scalability: as shown in Figure 1, existing solutions require independent training and deployment for each scenario, leading to low resource utilization and high maintenance costs-a challenge unaddressed in GR literature. To address these, we present OxygenREC, an industrial recommendation system that leverages Fast-Slow Thinking to deliver deep reasoning with strict latency and multi-scenario requirements of real-world environments. First, we adopt a Fast-Slow Thinking architecture. Slow thinking uses a near-line LLM pipeline to synthesize Contextual Reasoning Instructions, while fast thinking employs a high-efficiency encoder-decoder backbone for real-time generation. Second, to ensure reasoning instructions effectively enhance recommendation generation, we introduce a semantic alignment mechanism with Instruction-Guided Retrieval (IGR) to filter intent-relevant historical behaviors and use a Query-to-Item (Q2I) loss for instruction-item consistency. Finally, to resolve multi-scenario scalability, we transform scenario information into controllable instructions, using unified reward mapping and Soft Adaptive Group Clip Policy Optimization (SA-GCPO) to align policies with diverse business objectives, realizing a train-once-deploy-everywhere paradigm.

cs.IR↗