Estimate the VRAM requirements of a model and context size. Optionally provide a Hugging Face token for gated models.