Fangcheng Fu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Fangcheng Fu has a total of 158 co-authors. Here are the top co-authors by number of shared publications:
- Bin Cui
Peking University
21 shared publications - Xiaonan Nie
Peking University, Beijing, China
7 shared publications - Xupeng Miao
Purdue University, West Lafayette, IN, USA
6 shared publications - Yujie Wang
Peking University, Beijing, China
6 shared publications - Hao Ge
Peking University, Beijing, China
5 shared publications - Haoyang Li
Peking University, Beijing, China
5 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38 (11) , 7332-7345
Collaborative-Guided Diffusion for Sequential Recommendation
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 38 (11) , 7332-7345
Collaborative-Guided Diffusion for Sequential Recommendation
- Open Access
- 28 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-17
Janus: Multi-LLM Serving at Production Scale
- 28 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-17
Open AccessJanus: Multi-LLM Serving at Production Scale
- Open Access
- 24 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 4 (4) , 1-29
StaleFlow: Staleness-Aware Data Management for Mitigating Data Skewness in Fully Disaggregated RL Post-Training
Reinforcement learning (RL) post-training has become pivotal for enhancing the capabilities of modern large models. A recent trend is to develop RL systems with a fully disaggregated architecture, which decouples the three RL phases (rollout, reward, and training) onto separate resources and executes them asynchronously. However, two critical data-level concerns arise: (1) asynchronous execution leads to data staleness in trajectories (the data generated by rollout) as the model parameters used in rollout may not be up to date, which impairs RL convergence; and (2) the length variation of trajectories introduces severe data skewness, leading to workload imbalance and degraded system performance. Existing systems fail to address these two concerns in a unified manner. Techniques that tightly control data staleness often constrain effective data skewness mitigation, while aggressive data skewness mitigation tends to exacerbate data staleness. As a result, systems are forced to trade off convergence for performance, or vice versa. To address this, we propose StaleFlow, an RL post-training system that jointly tackles data staleness and skewness. First, to control staleness, StaleFlow introduces a global consistency protocol that tracks the full lifecycle of each trajectory and constrains staleness. Second, to mitigate skewness, StaleFlow re-designs the RL system architecture by constructing data servers for trajectories and parameters to achieve flexible rollout coordination. Subsequently, we develop a suite of staleness-aware, throughput-oriented strategies to enhance system performance. Evaluations show that StaleFlow achieves up to 1.42-2.68× (1.18-1.91× on average) higher throughput than state-of-the-art systems, without compromising convergence. Our source code is available: https://github.com/psrl-project/psrl.
- 24 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 4 (4) , 1-29
Open AccessStaleFlow: Staleness-Aware Data Management for Mitigating Data Skewness in Fully Disaggregated RL Post-Training
Reinforcement learning (RL) post-training has become pivotal for enhancing the capabilities of modern large models. A recent trend is to develop RL systems with a fully disaggregated architecture, which decouples the three RL phases (rollout, reward, and training) onto separate resources and executes them asynchronously. However, two critical data-level concerns arise: (1) asynchronous execution leads to data staleness in trajectories (the data generated by rollout) as the model parameters used in rollout may not be up to date, which impairs RL convergence; and (2) the length variation of trajectories introduces severe data skewness, leading to workload imbalance and degraded system performance. Existing systems fail to address these two concerns in a unified manner. Techniques that tightly control data staleness often constrain effective data skewness mitigation, while aggressive data skewness mitigation tends to exacerbate data staleness. As a result, systems are forced to trade off convergence for performance, or vice versa. To address this, we propose StaleFlow, an RL post-training system that jointly tackles data staleness and skewness. First, to control staleness, StaleFlow introduces a global consistency protocol that tracks the full lifecycle of each trajectory and constrains staleness. Second, to mitigate skewness, StaleFlow re-designs the RL system architecture by constructing data servers for trajectories and parameters to achieve flexible rollout coordination. Subsequently, we develop a suite of staleness-aware, throughput-oriented strategies to enhance system performance. Evaluations show that StaleFlow achieves up to 1.42-2.68× (1.18-1.91× on average) higher throughput than state-of-the-art systems, without compromising convergence. Our source code is available: https://github.com/psrl-project/psrl.
- Open Access
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1055-1072
LAER-MoE: Load-Adaptive Expert Re-layout for Efficient Mixture-of-Experts Training
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1055-1072
Open AccessLAER-MoE: Load-Adaptive Expert Re-layout for Efficient Mixture-of-Experts Training
- Open Access
- 28 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 398-412
Elastor: Elastic and Efficient Model Partitioning and Checkpointing for Fault-Tolerant Distributed Training
- 28 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 398-412
Open AccessElastor: Elastic and Efficient Model Partitioning and Checkpointing for Fault-Tolerant Distributed Training
- Open Access
- 4 December 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (6) , 1-30
Hydraulis: Balancing Large Transformer Model Training via Co-designing Parallel Strategies and Data Assignment
To optimize large Transformer model training, both efficient parallel computing and advanced data management are indispensable. However, current methods often assume a stable and uniform training workload, neglecting data-induced imbalances-arising from both sampling and packing processes-which can impede training performance. Specifically, data sampling imbalance arises from uneven sequence length distribution of the training data, while data packing imbalance stems from the discrepancy between the linear memory complexity and quadratic time complexity of the attention mechanism. To address these imbalance issues, we develop Hydraulis, which jointly optimizes the parallel strategies and data assignment. For one thing, we introduce large model training with dynamic heterogeneous parallel strategies in response to the sequence length variations within and across training iterations. For another, we devise a two-stage data assignment approach, which strikes a good balance in terms of the training workloads both within and across model replicas. Empirical results demonstrate that Hydraulis outperforms existing systems by 1.32-2.66×. Our source code is available: https://github.com/PKU-DAIR/Hetu.
- 4 December 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (6) , 1-30
Open AccessHydraulis: Balancing Large Transformer Model Training via Co-designing Parallel Strategies and Data Assignment
To optimize large Transformer model training, both efficient parallel computing and advanced data management are indispensable. However, current methods often assume a stable and uniform training workload, neglecting data-induced imbalances-arising from both sampling and packing processes-which can impede training performance. Specifically, data sampling imbalance arises from uneven sequence length distribution of the training data, while data packing imbalance stems from the discrepancy between the linear memory complexity and quadratic time complexity of the attention mechanism. To address these imbalance issues, we develop Hydraulis, which jointly optimizes the parallel strategies and data assignment. For one thing, we introduce large model training with dynamic heterogeneous parallel strategies in response to the sequence length variations within and across training iterations. For another, we devise a two-stage data assignment approach, which strikes a good balance in terms of the training workloads both within and across model replicas. Empirical results demonstrate that Hydraulis outperforms existing systems by 1.32-2.66×. Our source code is available: https://github.com/PKU-DAIR/Hetu.
- Open Access
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 8, 7002-7010
Model Rake: A Defense Against Stealing Attacks in Split Learning
Split learning is a prominent framework for vertical federated learning, where multiple clients collaborate with a central server for model training by exchanging intermediate embeddings. Recently, it is shown that an adversarial server can exploit the intermediate embeddings to train surrogate models to replace the bottom models on the clients (i.e., model stealing). The surrogate models can also be used to reconstruct private training data of the clients (i.e., data stealing). To defend against these stealing attacks, we propose Model Rake (i.e., Rake), which runs two bottom models on each client and differentiates their output spaces to make the two models distinct. Rake hinders the stealing attacks because it is difficult for a surrogate model to approximate two distinct bottom models. We prove that, under some assumptions, the surrogate model converges to the average of the two bottom models and thus will be inaccurate. Extensive experiments show that Rake is much more effective than existing methods in defending against both model and data stealing attacks, and the accuracy of normal model training is not affected.
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 8, 7002-7010
Open AccessModel Rake: A Defense Against Stealing Attacks in Split Learning
Split learning is a prominent framework for vertical federated learning, where multiple clients collaborate with a central server for model training by exchanging intermediate embeddings. Recently, it is shown that an adversarial server can exploit the intermediate embeddings to train surrogate models to replace the bottom models on the clients (i.e., model stealing). The surrogate models can also be used to reconstruct private training data of the clients (i.e., data stealing). To defend against these stealing attacks, we propose Model Rake (i.e., Rake), which runs two bottom models on each client and differentiates their output spaces to make the two models distinct. Rake hinders the stealing attacks because it is difficult for a surrogate model to approximate two distinct bottom models. We prove that, under some assumptions, the surrogate model converges to the average of the two bottom models and thus will be inaccurate. Extensive experiments show that Rake is much more effective than existing methods in defending against both model and data stealing attacks, and the accuracy of normal model training is not affected.
- Open Access
- 27 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 963-978
ByteScale: Communication-Efficient Scaling of LLM Training with a 2048K Context Length on 16384 GPUs
- 27 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 963-978
Open AccessByteScale: Communication-Efficient Scaling of LLM Training with a 2048K Context Length on 16384 GPUs
- Open Access
- 17 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (3) , 1-28
Malleus: Straggler-Resilient Hybrid Parallel Training of Large-scale Models via Malleable Data and Model Parallelization
As the scale of models and training data continues to grow, there is an expanding reliance on more GPUs to train large-scale models, which inevitably increases the likelihood of encountering dynamic stragglers that some devices lag behind in performance occasionally. However, hybrid parallel training, one of the de facto paradigms to train large models, is typically sensitive to the stragglers. This paper presents Malleus , a straggler-resilient hybrid parallel training framework for large-scale models. Malleus quantifies the stragglers at the nuanced, per-GPU granularity during training, and develops a novel planning algorithm to deduce the optimal parallelization of GPU devices, pipeline stages, model layers, and training data, maximizing training efficiency when stragglers exist. In addition, once a shift in the straggler situation is detected, Malleus adaptively adjusts the parallelization via a re-planning process, and seamlessly and efficiently migrates the model states on the fly, without sacrificing the stability of the training tasks. Empirical results on large language models with up to 110B parameters show that Malleus consistently outperforms existing parallel training frameworks under various straggler situations, delivering on average 2.63-5.28x of efficiency improvement.
- 17 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (3) , 1-28
Open AccessMalleus: Straggler-Resilient Hybrid Parallel Training of Large-scale Models via Malleable Data and Model Parallelization
As the scale of models and training data continues to grow, there is an expanding reliance on more GPUs to train large-scale models, which inevitably increases the likelihood of encountering dynamic stragglers that some devices lag behind in performance occasionally. However, hybrid parallel training, one of the de facto paradigms to train large models, is typically sensitive to the stragglers. This paper presents Malleus , a straggler-resilient hybrid parallel training framework for large-scale models. Malleus quantifies the stragglers at the nuanced, per-GPU granularity during training, and develops a novel planning algorithm to deduce the optimal parallelization of GPU devices, pipeline stages, model layers, and training data, maximizing training efficiency when stragglers exist. In addition, once a shift in the straggler situation is detected, Malleus adaptively adjusts the parallelization via a re-planning process, and seamlessly and efficiently migrates the model states on the fly, without sacrificing the stability of the training tasks. Empirical results on large language models with up to 110B parameters show that Malleus consistently outperforms existing parallel training frameworks under various straggler situations, delivering on average 2.63-5.28x of efficiency improvement.
- Top Cited PapersOpen Access
- 17 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (3) , 1-30
PQCache: Product Quantization-based KVCache for Long Context LLM Inference
As the field of Large Language Models (LLMs) continues to evolve, the context length in inference is steadily growing. Key-Value Cache (KVCache), the intermediate representations of tokens within LLM inference, has now become the primary memory bottleneck due to limited GPU memory. Current methods selectively determine suitable keys and values for self-attention computation in LLMs to address the issue. However, they either fall short in maintaining model quality or result in high serving latency. Drawing inspiration from advanced embedding retrieval techniques prevalent in the data management community, we consider the storage and retrieval of KVCache as a typical embedding retrieval problem. We propose PQCache , which employs Product Quantization (PQ) to manage KVCache, maintaining model quality while ensuring low serving latency. During the prefilling phase, we apply PQ to tokens' keys for each LLM layer and head. During the autoregressive decoding phase, we use PQ codes and centroids to approximately identify important preceding tokens, then fetch the corresponding key-value pairs for self-attention computation. Through meticulous design of overlapping and caching, we minimize any additional computation and communication overhead during both phases. Extensive experiments demonstrate that PQCache achieves both effectiveness and efficiency, with 4.60% score improvement over existing methods on InfiniteBench and low system latency in both prefilling and decoding.
- 17 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (3) , 1-30
Top Cited PapersOpen AccessPQCache: Product Quantization-based KVCache for Long Context LLM Inference
As the field of Large Language Models (LLMs) continues to evolve, the context length in inference is steadily growing. Key-Value Cache (KVCache), the intermediate representations of tokens within LLM inference, has now become the primary memory bottleneck due to limited GPU memory. Current methods selectively determine suitable keys and values for self-attention computation in LLMs to address the issue. However, they either fall short in maintaining model quality or result in high serving latency. Drawing inspiration from advanced embedding retrieval techniques prevalent in the data management community, we consider the storage and retrieval of KVCache as a typical embedding retrieval problem. We propose PQCache , which employs Product Quantization (PQ) to manage KVCache, maintaining model quality while ensuring low serving latency. During the prefilling phase, we apply PQ to tokens' keys for each LLM layer and head. During the autoregressive decoding phase, we use PQ codes and centroids to approximately identify important preceding tokens, then fetch the corresponding key-value pairs for self-attention computation. Through meticulous design of overlapping and caching, we minimize any additional computation and communication overhead during both phases. Extensive experiments demonstrate that PQCache achieves both effectiveness and efficiency, with 4.60% score improvement over existing methods on InfiniteBench and low system latency in both prefilling and decoding.