Topic
Pruning
-
News AI
LLM weight pruning degrades fine-grained skills that survive quantization in 35-model audit
Researchers evaluated 35 compressed checkpoints across six open-weight language models, finding that pruning causes capability collapse on complex tasks.