arXiv subjects

Zhizhao Zeng

Publications and source records attributed to Zhizhao Zeng.

4 recordsLinked to original sources

arXiv2026

DivOPD: Spread Wide, Look Close for Asynchronous On-Policy Distillation of Multi-turn Agents

On-policy distillation (OPD) trains student agents through teacher supervision on their own interactions with an environment. However, in asynchronous multi-turn training, arrival-order batching can allow a few early or long rollouts to dominate learner updates while other valid rollouts become stale before being used, wasting already-generated experience. To address this problem, we introduce DivOPD, a simple learner-side batch-selection method that spreads a fixed turn budget across more rollouts and, within each rollout, prioritizes turns with larger cumulative teacher-student disagreement. Turns without usable teacher feedback are excluded. The per-turn loss and optimizer remain fixed; selection only changes which student-visited turns receive training weight. For no-progress rollouts, an optional extension briefly hands control to the teacher before returning it to the student. Across six teacher-student settings on the simulated ALFWorld, ScienceWorld, and WebShop benchmarks, with 1.5B-7B students, DivOPD raises cross-setting mean peak success rate from 77.4 to 84.4 and mean success over the last five evaluations from 71.5 to 78.6. It reaches all reported setting-specific targets with geometric-mean speedups of 1.84x in training tokens and 1.87x in learner GPU time relative to vanilla OPD. Teacher intervention further raises this last-five mean to 82.4 while retaining about 1.7x learner-GPU speedup over vanilla OPD. Code will be released at https://github.com/HanyangWang0418-oss/DivOPD.

cs.LG↗

arXiv2026

CuraWeb: Joint Optimization of Quality, Redundancy, and Diversity for Web-Scale Pretraining Data

Open-web corpora curated via highly selective filters, such as FineWeb-Edu and DCLM, constitute the core of LLM pretraining data and have significantly advanced LLM performance. However, these pipelines typically rely on singular optimization objectives, which inevitably narrows distributional diversity and marginalizes long-tail knowledge, thereby restricting data coverage and underutilizing the vast potential of the open web. To address this limitation, we propose a novel curation paradigm that shifts from linear pruning to the joint optimization of quality, redundancy, and diversity. This framework synergizes dual-track cleaning (rule-based and model-driven) with hybrid deduplication (n-gram and semantic), while employing a multi-objective sampler to balance informational quality with distributional breadth. Applying this framework to Common Crawl, we construct CuraWeb, a 2T-token English corpus. Unlike existing resources, CuraWeb establishes an industrial-grade standard for data curation by recovering a more holistic data distribution with enhanced diversity and minimal redundancy, achieving broader coverage of long-tail knowledge across diverse domains. Experimental evaluations at the 3B scale demonstrate that CuraWeb significantly outperforms state-of-the-art baselines, yielding an average performance gain of 1.8\% across a wide range of benchmarks, particularly in knowledge-intensive and reasoning tasks.

cs.AI↗

arXiv2026

BiPACE: Bisimulation-Guided Policy Optimization with Action Counterfactual Estimation for LLM Agents

Stepwise group-based RL is an attractive way to train long-horizon LLM agents without a learned critic: it reuses multiple sampled rollouts to estimate local advantages. Its weakness is less visible but more fundamental: every group-relative estimator assumes that the steps it compares are equivalent for credit assignment. We show that current agentic variants violate this assumption through a state-action credit mismatch. The observation-hash partition is overly fine on the state side, creating singleton groups with zero step-level signal, while a single within-group mean is too coarse on the action side, mixing state-value estimation with action-specific credit. We introduce BiPACE (Bisimulation-Guided Policy Optimization with Action Counterfactual Estimation), a drop-in advantage estimator that fixes both sides without adding a critic, auxiliary loss, or extra rollouts. BiGPO clusters steps by cosine distance in the actor's own hidden-state geometry, an empirical policy-induced proxy for bisimulation that substantially lowers the singleton rate left by observation hashing. PACE then recenters returns within each behavioral cluster using action-conditioned peer baselines; its Q-style instance estimates a local Q(s,a)-V(s) nonparametrically. On ALFWorld/Qwen2.5-7B, BiPACE_Q raises overall validation success from GiGPO's 90.8 to $97.1\pm0.9$ over three seeds, and crosses the 95% threshold on every seed, which GiGPO never does within the same budget. On Qwen2.5-1.5B it reaches $93.5\pm1.2$ versus GiGPO's 86.7, and on WebShop and TextCraft it improves over GRPO and GiGPO at both model scales. The measured BiPACE-specific overhead is 11.3% of a single training-step wall time. Yet it changes the estimator's comparison unit from surface identity to approximate behavioral equivalence plus action-side counterfactuals. The code is available at https://github.com/TianxiangZhao/BiPACE.

cs.CL↗

arXiv2025

Rectify Evaluation Preference: Improving LLMs' Critique on Math Reasoning via Perplexity-aware Reinforcement Learning

To improve Multi-step Mathematical Reasoning (MsMR) of Large Language Models (LLMs), it is crucial to obtain scalable supervision from the corpus by automatically critiquing mistakes in the reasoning process of MsMR and rendering a final verdict of the problem-solution. Most existing methods rely on crafting high-quality supervised fine-tuning demonstrations for critiquing capability enhancement and pay little attention to delving into the underlying reason for the poor critiquing performance of LLMs. In this paper, we orthogonally quantify and investigate the potential reason -- imbalanced evaluation preference, and conduct a statistical preference analysis. Motivated by the analysis of the reason, a novel perplexity-aware reinforcement learning algorithm is proposed to rectify the evaluation preference, elevating the critiquing capability. Specifically, to probe into LLMs' critiquing characteristics, a One-to-many Problem-Solution (OPS) benchmark is meticulously constructed to quantify the behavior difference of LLMs when evaluating the problem solutions generated by itself and others. Then, to investigate the behavior difference in depth, we conduct a statistical preference analysis oriented on perplexity and find an intriguing phenomenon -- ``LLMs incline to judge solutions with lower perplexity as correct'', which is dubbed as \textit{imbalanced evaluation preference}. To rectify this preference, we regard perplexity as the baton in the algorithm of Group Relative Policy Optimization, supporting the LLMs to explore trajectories that judge lower perplexity as wrong and higher perplexity as correct. Extensive experimental results on our built OPS and existing available critic benchmarks demonstrate the validity of our method.

cs.CL↗