{"paper":{"title":"UniComp: A Unified Evaluation of Large Language Model Compression via Pruning, Quantization and Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Compressed language models preserve factual recall but lose multi-step reasoning, multilingual ability, and instruction following.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andreas Geiger, Jonathan von Rad, Yong Cao","submitted_at":"2026-02-09T19:20:56Z","abstract_excerpt":"Model compression is increasingly essential for deploying large language models (LLMs), yet existing comparative studies largely focus on pruning and quantization evaluated primarily on knowledge-centric benchmarks. Thus, we introduce UniComp, a unified evaluation framework for comparing pruning, quantization, and knowledge distillation. UniComp evaluates compressed models along three dimensions: performance, reliability, and efficiency, using a diverse set of capability- and safety-oriented benchmarks together with a hardware-aware efficiency analysis. Through evaluation of six compression te"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Through evaluation of six compression techniques across 40 datasets, we observe (i) a consistent knowledge bias, where factual recall is largely preserved while multi-step reasoning, multilingual, and instruction-following capabilities degrade; (ii) a decoupling between performance and reliability, indicating that retained performance does not consistently imply preserved reliability; and (iii) that task-specific calibration can yield up to 50% relative improvement of reasoning performance in pruned models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The selected benchmarks and compression techniques are representative of general behavior, and the chosen reliability metrics accurately capture real-world safety and consistency issues without hidden selection effects.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"UniComp finds that LLM compression preserves factual recall but degrades multi-step reasoning, multilingual ability, and reliability, while task-specific calibration recovers up to 50% of lost reasoning performance in pruned models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Compressed language models preserve factual recall but lose multi-step reasoning, multilingual ability, and instruction following.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8d33721e483c030cc30b4a4092654f81218cfce599fe4257ed8a2d7d4fb3569e"},"source":{"id":"2602.09130","kind":"arxiv","version":5},"verdict":{"id":"ad7df9ec-4402-420e-b103-4ac03a2977b8","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T05:13:32.624478Z","strongest_claim":"Through evaluation of six compression techniques across 40 datasets, we observe (i) a consistent knowledge bias, where factual recall is largely preserved while multi-step reasoning, multilingual, and instruction-following capabilities degrade; (ii) a decoupling between performance and reliability, indicating that retained performance does not consistently imply preserved reliability; and (iii) that task-specific calibration can yield up to 50% relative improvement of reasoning performance in pruned models.","one_line_summary":"UniComp finds that LLM compression preserves factual recall but degrades multi-step reasoning, multilingual ability, and reliability, while task-specific calibration recovers up to 50% of lost reasoning performance in pruned models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The selected benchmarks and compression techniques are representative of general behavior, and the chosen reliability metrics accurately capture real-world safety and consistency issues without hidden selection effects.","pith_extraction_headline":"Compressed language models preserve factual recall but lose multi-step reasoning, multilingual ability, and instruction following."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.09130/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}