张程皓
我是清华大学软件学院软件工程硕士研究生，预计 2027 年毕业，本科同样就读于清华大学软件学院。
我的研究与工程实践聚焦大模型后训练与检索增强，主要包括 Agentic RAG、SFT/GRPO、 大模型评测与误差分析，以及分布式训练系统优化。
曾在阿里巴巴千问事业部和快手基础大模型与应用部实习，开展长耗时 Agentic RAG、大模型后训练、 多模态推荐表征与生成式推荐相关工作，关注大模型系统的准确性、可靠性与运行效率。
教育经历
-
清华大学软件学院软件工程硕士2024年8月 - 预计2027年6月 -
清华大学软件学院软件工程学士2020年9月 - 2024年6月
经历
-
清华大学研究生会信息服务中心负责人2024 - 2025
荣誉与奖项
-
清华大学优秀学生干部2025
-
综合优秀奖学金2024-2025
-
科技创新奖学金2023
-
学业优秀奖学金2021-2022
-
清华大学挑战杯三等奖2023
实习经历
阿里巴巴｜千问事业部｜大模型算法实习生
2026.06 - 2026.08- 复杂知识题的长耗时 Agentic RAG： 面向单次 RAG 难以覆盖分散判据、错误证据易带偏推理的问题，负责长耗时解题 Agent 的 POC 验证与方案开发；将单轮生成重构为可自主检索、核验证据并主动停止的 Agent，设计 Source-Aware Agent Loop 管理工具调用、证据冲突与停止策略。在 Qwen3.6-35B-A3B 上，准确率由 85.23% 提升至 89.26%（+4.03pp）。
- Agent 后训练、评测与持续迭代： 定位直接 RL 引发的工具行为塌缩，以教师正确轨迹 SFT 恢复工具使用；针对 GRPO 信号稀疏，搭建“盲答筛题—多角度 Rollout 分层—失败归因与题目合成”闭环，并按正确性、工具行为、时延与 Bad Case 持续评测迭代。最终解题准确率达到 91.95%，Agent 时延由 19.49s 降至 12.23s。
- 异构 RL 训练系统优化： 针对策略更新与 Rollout 的负载差异，改造 verl fully-async 链路，实现 L20X/H20 按卡型分池调度与 NCCL 权重直传；32 卡（2+2 节点）下平均单步耗时由 280s 降至 116s，吞吐提升 2.41×。
快手｜基础大模型与应用部｜大模型算法实习生
2026.01 - 2026.06- 多模态推荐表征：领域适配与语义对齐（CREM，优秀 Launch Review）： 将 CREM backbone 从 Qwen2.5-VL-3B 升级至 Qwen2.5-Omni-3B，引入音频覆盖直播场景；以 LLM 生成千万级相关性软标签，结合高热负采样、in-batch hard negative mining、InfoNCE 与多模态 QA 联训完成推荐域对齐。负责数据 pipeline、分布式训练推理及线上 embedding 服务交付，实现离线 AUC +0.45pp，内循环大盘后验预期花费 +1.883%。
- 生成式推荐大模型：物品预测与用户画像统一建模（UBioRec）： 基于 Qwen2.5-7B 构造 60 万组三元组进行多任务 SFT，以 Adaptive Token Scheme 和共享前缀 KV cache 统一 item 预测与用户画像生成；画像推理加速 6.5×，万级检索 NDCG=0.2098，线上本地生活 GMV +2.29%、关注页零样本迁移时长 +0.91%。
动态
论文与手稿 (查看全部 )
SoBR: Scaling Annotation-Free Code Retriever Training Beyond the Memory Wall
Chenghao Zhang, et al.
AAAI Conference on Artificial Intelligence (AAAI), under review 2027 (first author)
Introduces Source Slots to compress candidate representations for annotation-free retriever training, together with a single replay after cross-device gradient aggregation to preserve full-batch gradient equivalence while releasing intermediate activations. The method scales training to an 8B retriever with a shared pool of 256 documents and reaches 4.5× the median training speed of activation checkpointing.
SoBR: Scaling Annotation-Free Code Retriever Training Beyond the Memory Wall
Chenghao Zhang, et al.
AAAI Conference on Artificial Intelligence (AAAI), under review 2027 (first author)
Introduces Source Slots to compress candidate representations for annotation-free retriever training, together with a single replay after cross-device gradient aggregation to preserve full-batch gradient equivalence while releasing intermediate activations. The method scales training to an 8B retriever with a shared pool of 256 documents and reaches 4.5× the median training speed of activation checkpointing.
ClaimWeaver: Query-Conditioned Evidence Organization for Multi-Hop Retrieval-Augmented Generation
Chenghao Zhang, et al.
AAAI Conference on Artificial Intelligence (AAAI), under review 2027 (co-first author)
Introduces ClaimWeaver, a query-conditioned evidence organization framework that decomposes retrieved passages into source-grounded atomic claims, filters and consolidates evidence, and connects complementary claims through directed bridges for multi-hop reasoning.
ClaimWeaver: Query-Conditioned Evidence Organization for Multi-Hop Retrieval-Augmented Generation
Chenghao Zhang, et al.
AAAI Conference on Artificial Intelligence (AAAI), under review 2027 (co-first author)
Introduces ClaimWeaver, a query-conditioned evidence organization framework that decomposes retrieved passages into source-grounded atomic claims, filters and consolidates evidence, and connects complementary claims through directed bridges for multi-hop reasoning.
Disentangled Generation-Based Prototypical Alignment for Few-Shot Unsupervised Domain Adaptation in Graph-Level Anomaly Detection
Z. Ni, Chenghao Zhang, H. Wan, X. Zhao
AAAI Conference on Artificial Intelligence (AAAI) 2026
Introduces DGPA to mitigate performance degradation in cross-domain few-shot graph-level anomaly detection, improving average AUROC by 5.72pp over the strongest baseline.
Disentangled Generation-Based Prototypical Alignment for Few-Shot Unsupervised Domain Adaptation in Graph-Level Anomaly Detection
Z. Ni, Chenghao Zhang, H. Wan, X. Zhao
AAAI Conference on Artificial Intelligence (AAAI) 2026
Introduces DGPA to mitigate performance degradation in cross-domain few-shot graph-level anomaly detection, improving average AUROC by 5.72pp over the strongest baseline.
Factor-wise Disentangled Contrastive Learning for Cross-domain Few-shot Molecular Property Prediction
Z. Ni, Chenghao Zhang, H. Wan, X. Zhao
Frontiers of Computer Science 2025
Studies factor-wise disentangled contrastive learning for cross-domain few-shot molecular property prediction and improves average ROC-AUC by 1.53pp.
Factor-wise Disentangled Contrastive Learning for Cross-domain Few-shot Molecular Property Prediction
Z. Ni, Chenghao Zhang, H. Wan, X. Zhao
Frontiers of Computer Science 2025
Studies factor-wise disentangled contrastive learning for cross-domain few-shot molecular property prediction and improves average ROC-AUC by 1.53pp.
Geometry-Guided Domain Generalization for Monocular 3D Object Detection
F. Yang, H. Chen, Y. He, S. Zhao, Chenghao Zhang, K. Ni, G. Ding
AAAI Conference on Artificial Intelligence (AAAI) 2024
Uses geometry priors to improve cross-domain generalization for monocular 3D object detection. I contributed to implementing and integrating the attention module.
Geometry-Guided Domain Generalization for Monocular 3D Object Detection
F. Yang, H. Chen, Y. He, S. Zhao, Chenghao Zhang, K. Ni, G. Ding
AAAI Conference on Artificial Intelligence (AAAI) 2024
Uses geometry priors to improve cross-domain generalization for monocular 3D object detection. I contributed to implementing and integrating the attention module.
全部论文
研究主线
Agentic RAG
面向复杂知识任务构建可自主检索、核验证据、处理冲突并主动停止的长程 Agent。
大模型后训练
围绕 SFT/GRPO、轨迹构造、奖励设计、大模型评测与误差分析，优化模型能力与工具行为。
检索与推荐
通过对比学习与多任务目标，学习多模态推荐表征和生成式 user/item 模型。
分布式训练系统优化
优化分布式训练、异步 RL、Rollout 服务、通信、时延与吞吐，提升大模型研发效率。