Fused Quantized GEMV Kernels for LLM Inference - 2.8x faster than cuBLAS - View it on GitHub