Configure Aspose.LLM for .NET for 128K-262K conTexts — pick the right preset, enable flash attention, quantize the KV cache, and manage memory.... Vision + text with many image turns. Prerequisites...cache size for a model with L layers, H heads of dim D , context...