Z zenodo.org Post-transformer inference: 224× compression of Llama-70B with improved accuracy 72 points anima-core 10 months ago 56 comments
