Estimate of VRAM needs based on context length and quantization

It would really help to know what is the VRAM necessary to load and run the models that are available on the Ollama.com site. The needs are enormous when larger context window is set with the num_ctx parameter. In addition, this also depends on quantization of the model. Just a few examples of VRAM needs would be helpful. For example, 2k tokens, 8k, 32k, 128k. Thank you!