Tag

#model-efficiency

1 story tagged model-efficiency.

Horizontal bar chart showing CBO MMLU advantage over baselines: Llama-3.1-8B-Instruct at 6 percentage points above 40% compression, Qwen3-14B at 10 percentage points at 30% compression, and Llama-3.3-70B-Instruct at 23 percentage points at 50% compression. LLM block pruning via Ising glass optimization.

Ising glass LLM block pruning saves 23 MMLU points at 50%

LLM block pruning via Ising glass optimization preserves MMLU near 77 at 50% compression of Llama-3.3-70B-Instruct, beating baselines by 23 percentage points.

Lars Cornelissen · Sep 21, 2026