I'm a Computer Science undergraduate at Ocean University of China, focused on AI infrastructure, ML systems, and LLM inference.

• 🔭 Building Tiny-LLM-Inference, a compact C++17/CUDA decoder-only inference runtime for systems learning and controlled experiments.
• ⚙️ Interested in LLM serving, scheduling, paged KV cache, prefill/decode optimization, GPU runtime performance, and benchmark methodology.
• 🌱 Strengthening my systems foundations through MIT 6.S081 / xv6.
• 🎯 Open to AI Infra, ML Systems, and LLM inference research or internship opportunities.

A single-process, single-device offline runtime inspired by modern LLM serving systems. It covers Hugging Face checkpoint loading, FCFS scheduling, chunked prefill/decode, tail preemption, scheduler-owned paged KV cache, CPU/CUDA execution, and reproducible correctness/performance evaluation.

Completed the Utilities, System Calls, Page Tables, Traps, and Copy-on-Write labs, including trace/sysinfo, USYSCALL/pgaccess, sigalarm/sigreturn, and COW fork with physical-page reference counting.

C++ · C · Python · CUDA · Linux · CMake · libtorch · Git · GoogleTest/CTest

LLM serving and benchmarking, vLLM/SGLang internals, multimodal inference systems, DiT, and non-autoregressive generation.