Cong L.

"The only thing better than free, is fast." Hi 🖐, I'm The Cong, a Senior ML Engineer. I write low-level CUDA/HIP GPU kernel for Ampere, Ada, Hopper, Blackwell, CDNA3. To optimize clients’ AI systems in both training and inference. I specialize in applying various techniques such as quantization (FP16, BF16, FP8), distillation, I/O aware, max out hardware utilization and hardware accelerators on the latest Nvidia and AMD GPUs, leading to massive cost saving and blazing fast runtime. For application specific, I have hands-on experience with algorithms such as FlashAttention, KV Caching, vLLM, Implicit GEMM (for CNN). My inference tech stacks include Low level C/C++, Nvidia CUDA/PTX, TensorRT, CUTLASS, cuBLAS, AMD ROCm HIP/Assembly coupled with profiling skills using Nsight Compute, Nsight System, Rocprof and deployment using Docker containerization. About my qualities. I'm proactive, curious and love learning about new advancement in the field of Deep learning,... I work diligently and creatively to bring satisfaction to my clients.