Environment for running Falcon-40B on NixOS
-
Download TheBloke's experimental 3-bit AutoGPTQ https://huggingface.co/TheBloke/falcon-40b-instruct-3bit-GPTQ
-
Everything should be nixified so on NixOS just do:
nix runBy default, the
falcontest app in this repo expects to find the dataset in../falcon-40b-instruct-3bit-GPTQ
Using Geforce RTX 4090, expect it to take few minutes to generate a lama story.