116efee0ee
llama: enable K-shift for quantized KV cache It will fail on unsupported backends or quant types.
llama: enable K-shift for quantized KV cache It will fail on unsupported backends or quant types.