A standalone GEMM kernel for fp16 activation and quantized weight, extracted from FasterTransformer - View it on GitHub