Model compression

Chinese Open-Source LLMs Are Now Slimmable: Qwen 27B Trimmed to 23B, No Retraining

A Reddit dev shrank Qwen 27B to ~23B by stripping layers — no retraining. Open-source Chinese LLMs are now "hackable" for local deployment.