arXiv subjects

Chengzhi Yu

Publications and source records attributed to Chengzhi Yu.

3 recordsLinked to original sources

arXiv2026

Mismatch Matters: On-Policy Distillation Beyond Token Agreement

On-policy distillation (OPD) has emerged as a core component of modern LLM post-training pipelines, yet we reveal a failure mode: degenerate agreement, where students exploit repetitive loops to achieve near-perfect token agreement with the teacher despite globally flawed responses. We therefore shift our focus from agreement to teacher-student mismatch, and find that mismatch tokens can be mainly categorized into two types: student-excess tokens and student-deficit tokens. Student-excess tokens are generated by the student but assigned near-zero probability by the teacher; their log-ratio corrections grow unbounded and destabilize the update. Student-deficit tokens, in contrast, are preferred by the teacher but rarely sampled by the student; their absence blocks the transfer of the teacher's reasoning patterns. To tackle these mismatch directions, we propose TIDE (Token-level Independent Deficit-Excess correction), which applies bounded Hellinger shaping to suppress the most severe sampled excesses and an analytic teacher top-$K$ injection to restore deficient probability mass without requiring deficit tokens to be sampled. Across mathematical reasoning benchmarks with multiple Qwen3 teacher-student pairs, TIDE consistently outperforms standard OPD and recent token-selection and reward-shaping baselines. Moreover, the gains of TIDE are more pronounced under strong teacher-student mismatch, where it improves Avg@8 from 6.9% to 20.3%, reduces average response length by a factor of 3.6, and substantially reduces formatting failures. Code is available at https://github.com/yzc-666/TIDE

cs.AI↗

arXiv2025

Optimizing LVLMs with On-Policy Data for Effective Hallucination Mitigation

Recently, large vision-language models (LVLMs) have risen to be a promising approach for multimodal tasks. However, principled hallucination mitigation remains a critical challenge.In this work, we first analyze the data generation process in LVLM hallucination mitigation and affirm that on-policy data significantly outperforms off-policy data, which thus calls for efficient and reliable preference annotation of on-policy data. We then point out that, existing annotation methods introduce additional hallucination in training samples, which may enhance the model's hallucination patterns, to address this problem, we propose training a hallucination classifier giving binary annotations, which guarantee clean chosen samples for the subsequent alignment. To further harness of the power of on-policy data, we design a robust iterative direct preference optimization (DPO) algorithm adopting a dynamic sample reweighting scheme. We conduct comprehensive experiments on three benchmarks with comparison to 8 state-of-the-art baselines. In particular, our approach reduces the hallucination rate of LLaVA-1.5-7B on MMHalBench by 50.8% and the average hallucination rate on Object HalBench by 79.5%; more significantly, our method fully taps into the potential of open-source models, enabling LLaVA-1.5-13B to even surpass the performance of GPT-4V.

cs.CV↗

arXiv2025

Variations of saturation vapor pressure and evaporation rate of liquids with their vaporization enthalpy

The phase transition enthalpy of condensed materials can be altered by factors such as electric fields, and such variations in turn affect physical and chemical behaviors including phase equilibrium. However, due to the neglect of non-expansion work, the Clapeyron equation does not account for the effect of changes in phase transition enthalpy on equilibrium. In this paper, by analyzing the electric-field-induced changes in phase transition enthalpy and incorporating the non-expansion work performed on the system, we extended both the Clapeyron and Clausius-Clapeyron equations to explicitly include variations in phase transition enthalpy. Building upon these extensions, analytical expressions for the vapor pressure of liquids and for the total evaporation rate of sessile liquid droplets as functions of the change in vaporization enthalpy have been derived, showing that an approximately 1% decrease in vaporization enthalpy of a liquid can cause an increase of about 20% in its vapor pressure and an almost twofold increase in its evaporation rate. The theoretical predictions by the present equations were compared and found to be in good agreement with the experimental observations.

cond-mat.soft↗