I'm diving deep into the intersection of hardware and AI, where the synergy between computational architecture and intelligent algorithms unlocks unprecedented efficiency and capabilities. Recently, I've been particularly drawn to the following areas:
- 1) LLM inference acceleration: Optimizing LLM inference speed via quantization, kernel optimization, and parallel computing for high-throughput, low-latency deployment across hardware.
- 2) Model high-performance optimization: Enhancing neural network efficiency end-to-end through algorithm adjustments (e.g., operator fusion) and hardware tuning, maximizing efficiency on resource-constrained devices.
- 3) Binary Neural Networks (BNN): Developing 1-bit neural networks to reduce computation/memory usage while preserving performance, for energy-efficient AI systems.
📙 Skills
- Python, C++, HLS, verilog, CUDA, Triton。