The model has a context of 131,072, but I only have 48G of VRAM so I run it with a context of 32768.