About Me

I am a direct-track PhD student at Fudan University, enrolled in 2023. My research focuses on efficient large language models, especially compression, PEFT, adaptive inference, and LLM serving systems.

I work on the full path from model compression and parameter-efficient adaptation to post-training infrastructure, multi-tenant serving, evaluation, and multimodal generation.

News

  • 2026-04-30 AutoQRA was accepted to ICML 2026.
  • 2026-04-04 Deputy, QR-Adaptor, and CoRE were accepted to ACL 2026.
  • 2026-01-25 Large Language Model Compression with Global Rank and Sparsity Optimization was accepted to ICLR 2026.
  • 2025-05-01 BSLoRA was accepted to ICML 2025.
  • 2025-03-14 Enhancing Object Coherence in Layout-to-Image Synthesis was accepted to ICME 2025.
  • 2025-01-22 RankAdaptor and QPruner were accepted to NAACL 2025.
  • 2024-12-09 Dynamic Operator Optimization for Efficient Multi-Tenant LoRA Model Serving was accepted to AAAI 2025.

Notes

2026-06-16 When a Better Clock Is Not a Better Sampler A fixed-checkpoint consistency-model diagnostic where a post-hoc damage clock loses to the native skip rule. FID delta plot for the damage-clock diagnostic

Publications

Post-training、模型发布与训练服务

Main figure for PEFT scaling

On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters

MindLab

Thesis / Core Contributor

Ranked #2 on Hugging Face Daily Papers

面向 post-training 从单模型训练走向共享基座 + 海量个人 adapter 的趋势,提出 PEFT 的 Scale Up / Scale Down / Scale Out 视角;结合 1T base、百万 adapter 管理与 LoRA RL 实践,将 adapter 定位为个性化模型的持久状态与训练服务接口。

arXiv

LongStraw execution path

LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget

Changhai Zhou, Kieran Liu, Yuhua Zhou, Qian Qiao, Jun Gao, Harry Zhang, Irvine Lu, Nolan Ho, Lucian Li, Andrew Lei, Cleon Cheng, Steven Chiang, Yihang Zeng, Di Zhang, Rio Yang, Kaijie Chen, Andrew Chen, Pony Ma, Weizhong Zhang, Cheng Jin

Technical Report / 一作

Ranked #1 on Hugging Face Daily Papers

面向百万 token 长上下文 RL 后训练在固定 GPU 预算下的显存瓶颈,LongStraw 对共享 prompt 进行一次无梯度 prefill,仅保留后续 response 所需的模型状态,并逐条重放 response 与累积梯度;在 Qwen3.6-27B 和 GLM-5.2 上验证 2M token 级执行路径。

arXiv PDF

Official Macaron-V1 benchmark across chat, agent, coding, and GenUI tasks

Introducing Macaron-V1

MindLab

Model Release / Core Contributor

Macaron-V1 正式发布包含 748B Venti 与 35B Tall 两个变体,延续 Mixture-of-LoRA,以 L0 router 在 Chat、Agent、Coding、GenUI 四类专家间选择;Venti 基于 GLM-5.2,Tall 基于 Qwen3.6-35B-A3B,并通过 MinT + MindForge 对齐 production harness 下的工具使用、Coding 和 GenUI 工作流。

Project Venti Tall

Overview figure for Macaron-V1-Preview

Macaron-V1-Preview: A 749B Personal Agent Model with Mixture of LoRA

MindLab

Model Release / Core Contributor

参与 Macaron-V1-Preview,面向 Agent 原生优化的 749B 开源模型;通过 MinT 与 Mix-of-LoRA 实现高效训推,引入 R3 与 Harness Context Protocol 解决 RL、Agent 任务训练及 auto-research 中的训推/训服不一致,使模型在真实 harness 上优化;在生活类 Agent 与 Openclaw 类 Agent 任务上超越 GPT-5.4 和 Claude Opus 4.7,并在 Coding Agent 类任务上取得开源模型 SOTA。

Project

Main figure for MinT infrastructure

MinT: Managed Infrastructure for Training and Serving Millions of LLMs

MindLab

Technical Report / Core Contributor

Ranked #1 on Hugging Face Daily Papers

针对百万级 LoRA 从训练、版本、导出到上线的生命周期管理挑战,设计 shared-base + adapter revision 的托管基础设施;通过 adapter-only handoff 与一致的训练/推理路径降低发版、回滚和服务成本。

arXiv

LLM Serving 与推理加速

Main figure for Dynamic Operator Optimization

Dynamic Operator Optimization for Efficient Multi-Tenant LoRA Model Serving

Changhai Zhou, Yuhua Zhou, Shiyang Zhang, Yibin Wang, Zekai Liu

AAAI 2025 / CCF-A / 一作

针对多租户 LoRA serving 中 batch 形态、rank 组合和硬件特性动态变化导致 kernel 难以手工选择的问题,提出分层搜索空间与 cost-model-guided evolutionary search,将 SGMV operator 从经验调参转化为可搜索、可迁移的系统优化问题。

Paper

Making Activation Sparsity Executable in Grouped Low-Bit LLMs

Changhai Zhou et al.

NeurIPS 2026 在投 / CCF-A / 一作

针对分组低比特 LLM 中激活稀疏理论可省算、实际难加速的落差,提出 group-aligned sparse decoding,结合校准式稀疏规划、在线残差选择与融合低比特算子,把稀疏性落到端到端 decode 吞吐提升。

Deputy: Accelerating Large Language Model Inference with Dynamic Low-Rank Substitution

Yuhua Zhou*, Shichao Weng*, Changhai Zhou*, Yuhan Wu, Qian Qiao, Jun Gao, Fei Yang†, Aimin Pan†

ACL 2026 / CCF-A

面向 token 级计算冗余和固定近似策略难以兼顾质量/速度的问题,提出动态低秩替代推理框架,让 attention/FFN 在 full / low-rank / skip 路径间自适应切换,并以 hybrid KV cache 支撑低开销决策。

模型压缩与参数高效适配

Main figure for Large Language Model Compression with Global Rank and Sparsity Optimization

Large Language Model Compression with Global Rank and Sparsity Optimization

Changhai Zhou, Qian Qiao, Yuhua Zhou, Yuxin Wu, Shichao Weng, Weizhong Zhang, Cheng Jin

ICLR 2026 / CCF-A / 一作

针对 LLM 压缩中低秩与稀疏常被独立分配、忽略跨层冗余与二者交互的问题,提出 RPCA + Bernoulli global allocation 两阶段框架;在全局预算下联合分配 rank 与 sparsity,兼顾可解释性与可落地性,为训练无关压缩提供更系统的层级分配范式。

arXiv

Main figure for QR-Adaptor

Balancing Fidelity and Plasticity: Aligning Mixed-Precision Fine-Tuning with Linguistic Hierarchies

Changhai Zhou, Shiyang Zhang, Yuhua Zhou, Qian Qiao, Jun Gao, Shichao Weng, Weizhong Zhang, Cheng Jin

ACL 2026 / CCF-A / 一作

针对 4-bit 微调中量化 fidelity 与 LoRA plasticity 相互耦合但常被分开调参的问题,提出 QR-Adaptor;从语言层级出发联合分配 frozen weight bit-width 与 adapter rank,在固定显存预算下逼近 16-bit LoRA 基线,为低精度 PEFT 提供结构化分配依据。

arXiv

Main figure for AutoQRA

AutoQRA: Joint Optimization of Mixed-Precision Quantization and Low-rank Adapters for Efficient LLM Fine-Tuning

Changhai Zhou, Shiyang Zhang, Yuhua Zhou, Qian Qiao, Jun Gao, Cheng Jin, Kaizhou Qin, Weizhong Zhang

ICML 2026 / CCF-A / 一作

面对混合精度量化和 LoRA rank 组合空间巨大、人工规则难覆盖模型/任务差异的挑战,将每层 bit-width 与 rank 建模为联合离散搜索;用多保真进化搜索和信赖域贝叶斯优化降低搜索成本,给出自动化高效微调方案,减少从经验调参到部署可用之间的试错成本。

arXiv

Main figure for RankAdaptor

RankAdaptor: Hierarchical Dynamic Low-Rank Adaptation for Structural Pruned LLMs

Changhai Zhou, Shijie Han, Lining Yang, Yuhua Zhou, Xu Cheng, Yibin Wang, Hongguang Li

NAACL 2025 / CCF-B / 一作

针对结构化剪枝后 LLM 各层容量受损不均、统一 rank 难以恢复性能的问题,提出分层动态 rank 分配和性能建模;在离线元学习与在线增量更新之间折中,把剪枝后的恢复微调从均匀补偿推进到按层诊断和按需适配。

arXiv

Main figure for QPruner

QPruner: Probabilistic Decision Quantization for Structured Pruning in Large Language Models

Changhai Zhou, Yuhua Zhou, Shijie Han, Qian Qiao, Hongguang Li

NAACL 2025 / CCF-B / 一作

针对结构化剪枝进一步叠加量化时精度损失难预测的问题,将 layer-wise mixed precision 与 Bayesian refinement 结合;把剪枝后模型的位宽分配从启发式选择转化为概率决策,使压缩率、显存和精度之间的取舍更可控,也更适合跨模型迁移。

arXiv

Main figure for BSLoRA

BSLoRA: Enhancing the Parameter Efficiency of LoRA with Intra-Layer and Inter-Layer Sharing

Yuhua Zhou*, Ruifeng Li*, Changhai Zhou, Fei Yang, Aimin Pan

ICML 2025 / CCF-A / 二作

针对 LoRA 参数效率仍受层内/层间重复表达限制的问题,引入 intra-layer 与 inter-layer sharing 机制;在尽量保持适配能力的同时复用低秩参数,说明 adapter 本身也存在可压缩结构,为 LoRA 结构设计提供轻量改造方向。

PMLR

LaRA: Layer-wise Rank Allocation for Efficient Fine-Tuning of Pruned Large Language Models

Yuhua Zhou, Changhai Zhou, Shiyang Zhang, Fei Yang, Yi Zhang, Aimin Pan

IP&M / 一区 top / 二作

面向剪枝 LLM 微调预算有限、各层恢复需求差异大的问题,提出 layer-wise rank allocation;将有限 LoRA 参数投向更关键层,帮助剪枝模型在训练成本受限时更有效恢复能力,也为后续剪枝模型的 rank 分配研究提供早期实证依据。

DOI

多模态生成模型

Main figure for SoulX-FlashHead

SoulX-FlashHead: Oracle-guided Generation of Infinite Real-time Streaming Talking Heads

Tan Yu, Qian Qiao, Le Shen, Ke Zhou, Jincheng Hu, Dian Sheng, Bo Hu, Haoming Qin, Jun Gao, Changhai Zhou, Shunshun Yin, Siyuan Liu

Technical Report / Core Contributor

面向长时实时数字人生成中音频驱动、身份一致性与长序列漂移难以兼顾的问题,参与 1.3B streaming talking-head 模型;结合 streaming-aware spatiotemporal pre-training、audio context caching 与 oracle-guided distillation,把多模态生成从离线短片段推进到低延迟连续交互场景。

arXiv

Main figure for Enhancing Object Coherence in Layout-to-Image Synthesis

Enhancing Object Coherence in Layout-to-Image Synthesis

Yibin Wang, Changhai Zhou, Honghui Xu

ICME 2025 / CCF-B / 二作

针对 layout-to-image 中对象语义错配、局部形变和物理关系不一致的问题,结合全局语义融合与自相似一致性注意力;让布局约束更稳定地传递到扩散生成过程,提升可控多模态生成的对象一致性,服务于更可编辑、更可靠的视觉内容生成。

arXiv

评测与代码推理

Main figure for CoRE code reasoning benchmark

CoRE: A Fine-Grained Code Reasoning Benchmark Beyond Output Prediction

Jun Gao, Yun Peng, Qian Qiao, Changhai Zhou, Yuhua Zhou, Shiyang Zhang, Shichao Weng, Zhenchang Xing, Xiaoxue Ren

ACL 2026 / CCF-A

针对代码推理评测过度依赖最终输出、难以区分真实推理与表面执行的问题,构建细粒度 benchmark;从实现不变性和过程透明性刻画模型行为,把评测从答案正确性推进到推理过程诊断,为代码 LLM 的训练迭代提供更可解释的信号。

arXiv

Contact

Feel free to reach out by email. You can also find my publication list on Google Scholar and code on GitHub.