1 BIT, POST TRAINING, QUANTIZATION

Summary from the TLDR AI newsletter

1 Bit language models are exciting. This work shows how to quantize the Linear layers of a language model without sacrificing performance. This can result in a 70B model running on consumer GPUs.