{"slug": "when-compression-scores-cannot-decide-information-boundaries-for-group-robust", "title": "When Compression Scores Cannot Decide: Information Boundaries for Group-Robust LLM Pruning", "summary": "A new arXiv study (2608.02940v1) finds that a dense pruning score with 0.906 split-half reliability predicted a 16.1% gain but selected an endpoint 6.0% and 7.7% worse than two controls, modeling the gap via information interfaces. Across three dense LLMs, a coarse depth allocation cut worst-group perplexity inflation by 12.6–20.9% relative to balanced uniform allocation, and model-specific complete-mask endpoint selection improved by 2.7–8.0%. In OLMoE, router traces predicted singleton direction in 114/192 cases versus 81/192 under the strongest relabeling, with held-out worst-group KL reductions of 13.7% and 7.2%.", "body_md": "arXiv:2608.02940v1 Announce Type: new\nAbstract: A reproducible compression statistic can still select the wrong candidate. A dense pruning score with 0.906 split-half reliability predicted a 16.1% gain. Its selected endpoint was 6.0% and 7.7% worse than two controls. We model the gap through information interfaces that delimit which distinctions each statistic supports. For equal-weight groups, a conic law gives the exact pooling price for positive linear fixed-candidate damage, including diagonal and full PSD second moments. Three two-world constructions and an exact observation-fiber radius characterize what pooled moments, group-local moments, and reference-path curvature leave unresolved. A group-resolved diagonal recovers broad damage order (Spearman 0.9239) while fine order remains weak. Relative to balanced uniform allocation, a coarse depth allocation cuts worst-group perplexity inflation by 12.6--20.9% across three dense LLMs. Model-specific complete-mask endpoint selection improves over those references by 2.7--8.0%. In OLMoE, router traces predict singleton direction (114/192 versus 81/192 under the strongest relabeling). Finite-menu decisions on one layer yield held-out worst-group KL reductions of 13.7% and 7.2%. Local measurements construct candidates. Selection is licensed by complete candidate endpoints or a validated uniform guarantee, with uncertainty calibrated to every comparison.", "url": "https://wpnews.pro/news/when-compression-scores-cannot-decide-information-boundaries-for-group-robust", "canonical_source": "https://arxiv.org/abs/2608.02940", "published_at": "2026-08-05 04:00:00+00:00", "updated_at": "2026-08-05 04:09:55.177342+00:00", "lang": "en", "topics": ["large-language-models", "machine-learning", "ai-research"], "entities": ["arXiv", "OLMoE"], "alternates": {"html": "https://wpnews.pro/news/when-compression-scores-cannot-decide-information-boundaries-for-group-robust", "markdown": "https://wpnews.pro/news/when-compression-scores-cannot-decide-information-boundaries-for-group-robust.md", "text": "https://wpnews.pro/news/when-compression-scores-cannot-decide-information-boundaries-for-group-robust.txt", "jsonld": "https://wpnews.pro/news/when-compression-scores-cannot-decide-information-boundaries-for-group-robust.jsonld"}}