Topic

Depth pruning

Model compression that deletes whole transformer blocks, shortening the network so that each removed block takes one stage of computation out of inference.

Current clusters

build2 publishers

Multiverse's block-pruning method widens its MMLU lead over block-influence pruning as compression grows, reaching 22.9 points at 50%

Multiverse Computing ranks candidate prunings by the energy of an Ising glass, so a single GPU scored about 29 billion combinations in two days. The paper reports the older per-block heuristics do just as well at lighter compression.

Publishers:huggingface.coruntimewire.com

Reality

Evidence54
Adoption14
Hype gap+29
Incentives74
Confidence61