On: Run llama3 locally with 1M token context

[tankado]

"Note: using a 256k context window requires at least 64GB of memory. Using a 1M+ context window requires significantly more (100GB+)."