Extreme quantization and sparsity of large language models by representing parameters as probaility distributions.
See the report: https://api.wandb.ai/links/aklein4/f54zip18
-
Create VM with version:
tpu-ubuntu2204-base -
Run command:
git clone https://github.com/aklein4/pBit.git -
Run command:
cd ~/pBit && . setup_vm.sh <HF_TOKEN> <WANDB_TOKEN>