📧 tianyig@outlook.com | 🔗 github.com/tianyi-ge | 📝 zhihu/ge-tian-yi | 💼 linkedin/tianyi-ge
Currently interested in
- Agentic RL Frameworks: High-performance replay buffer for off-policy RL, Inference resource pooling, Triton basics
- High-Performance Networking: Application of UCX/RDMA, Zero-copy transport, Metadata management
- Hardware-Software Co-Design: Large-scale EDA tool, HPC, MPI, LLC cache-hit optimization
- Ascend/ray-ascend (Maintainer): Hccl collective, Ascend direct tensor transport
- openeuler/yuanrong-datasystem (Maintainer): RDMA support and optimization based on OpenUCX
- Ascend/TransferQueue (Committer): Zero-copy serialization, Performance benchmark
- ray-project/ray: Performance fixes, Observability features, Collective refactor
- cupy/cupy: NCCL
commSplitsupport
- C/C++ · Python · Cython · Go · Triton
- Kubernetes · Docker · Bazel
Distilling my problem-solving -> problem-finding -> problem-defining skills...
[1] Jiawei Zhang, Xiaochen Zhou, Tianyi Ge, et al. Joint task scheduling and containerizing for efficient edge computing. IEEE Transactions on Parallel and Distributed Systems (TPDS), 2021. DOI: 10.1109/TPDS.2021.3059687


