quantize linux bin to convert models, not sure if it's useful

GGML 4bit version.. its pretty fast

no act orders or groups, just 4bit

Upload GPT-J-FourChannel-g128-4bit.safetensors

Commit the basics.

Update README.md

initial commit