Publications

(2026). Broken Symmetry in BF16 Attention: Why FlashAttention Gradients Blow Up Late in Training. arXiv.

PDF Cite arXiv

(2026). Diamond Agent: Agentic Control of Federated HPC Resources as a Service. arXiv.

PDF Cite arXiv

(2026). A Comparative Study on LLM-enabled HPC User Support. in PEARC 2026.

Cite DOI

(2026). SCAPE: Accurate and Efficient LLM Training with Extreme Sparse Communication. arXiv.

PDF Cite arXiv

(2026). PHOENIX: Resilient LLM Training with Hot-Swapping via Zero-Overhead Checkpoint. arXiv.

PDF Cite arXiv

(2026). Quantifying Performance Variability in GPU Clusters. in IEEE TPDS.

Cite DOI

(2025). Efficient Fine-Grained GPU Performance Modeling for Distributed Deep Learning of LLM. in HiPC 2025.

PDF Cite DOI arXiv

(2025). COMPSO: Optimizing Gradient Compression for Distributed Training with Second-Order Optimizers. in PPoPP 2025.

Cite DOI

(2020). Constant Power Load stabilization Based on Trajectory Control in Dual-Active-Bridge DC-DC Converter. in IPEMC2020-ECCE Asia.

PDF Cite DOI

(2019). Open-Circuit Fault Diagnosis of Dual Active Bridge DC-DC Converter With Extended-Phase-Shift Control. In IEEE ACCESS.

PDF Cite DOI