The Models You Can Run on a 16 GB GPU

For a batch size of 1 and a sequence length of 1024 tokens