CUDA: faster dequantize kernels for Q4_0 and Q4_1#4938

ikawrakow merged 1 commit intomasterggml-org/llama.cpp:masterfrom

ik/cuda_faster_legacy_dequantizeggml-org/llama.cpp:ik/cuda_faster_legacy_dequantizeCopy head branch name to clipboard