Website · CV · Lab profile · Email

I'm Jingming Liang, a final-year Computer Science undergraduate at Nankai University and an incoming PhD student at the Nankai–Baidu Joint Lab, starting in 2027.

I work on efficient AI systems, with interests in LLM inference and serving, heterogeneous computing and GPU optimization, and systems for embodied AI.

• GPU Optimization for All-Pairs Shortest Path
  CCF-TCARCH 2025 National Champion · Individual competitor
  Optimized a blocked Floyd–Warshall pipeline through memory locality, kernel fusion, and stream scheduling. Reduced end-to-end runtime from 11.50s to 5.42s on the competition's official hidden datasets.

• Multimodal MoE Routing Bias — CMRD
  Developed a training-free probe for noun-driven visual routing bias and studied inference-time mitigation of object hallucination.

• K-LoRA++: Adaptive Weight Selection for Diffusion Models
  Challenge Cup 2025 · National Third Prize
  Extended K-LoRA with eight timestep-dependent scaling schedules for object–style fusion and contributed to an interactive demo.

• RaceRadar
  Built a competition discovery system with automated ingestion, deduplication, and ranking, paired with a native Swift iOS app.

More projects and research →

I contribute to StarryOS / tgoskits, with merged Linux syscall compatibility fixes covering seccomp validation, CPU affinity ABI, and ptrace request width.

I share CS and AI learning resources, and enjoy music, marathon running, and independent travel.