{"id":"W4416219021","doi":"10.48550/arxiv.2508.15212","title":"SparK: Query-Aware Unstructured Sparsity with Recoverable KV Cache Channel Pruning","year":2025,"lang":"en","type":"preprint","venue":"ArXiv.org","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Strong; Institute for Catastrophic Loss Reduction","keywords":"Cache; Computation; Robustness (evolution); Pruning; Channel (broadcasting); Byte; Quantization (signal processing); SPARK (programming language); Inference","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001251363,0.001561679,0.00131747,0.0007975211,0.0006939992,0.001301948,0.004951633,0.001325721,0.005884655],"category_scores_gemma":[0.007184391,0.0008645728,0.001241532,0.0007931273,0.001044059,0.003117987,0.003325986,0.002541104,0.003443545],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008465307,"about_ca_system_score_gemma":0.003110176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01070082,"about_ca_topic_score_gemma":0.02687794,"domain_scores_codex":[0.9991108,0.0001441504,0.00006188998,0.0002676907,0.0002860815,0.0001294065],"domain_scores_gemma":[0.9980305,0.0008536251,0.00009207131,0.0005625307,0.0002732039,0.0001881166],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002068116,0.0007206813,0.007761881,0.0008819873,0.0003590957,0.0008704537,0.001142342,0.1625589,0.04300734,0.02157736,0.1487771,0.6102748],"study_design_scores_gemma":[0.0001553016,0.00007816259,0.0002963897,0.00001541397,0.00002160685,0.0001009285,0.00008256677,0.9742172,0.009318127,0.01047627,0.005208266,0.00002983944],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03519422,0.0006518986,0.7870973,0.0006211034,0.0003190969,0.0002900285,0.002646828,0.1699862,0.003193406],"genre_scores_gemma":[0.3488569,0.0003182907,0.6272639,0.000776353,0.000206694,0.0005419218,0.008086718,0.008741356,0.005207843],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01070082,"threshold_uncertainty_score":0.02127707,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02914784969433545,"score_gpt":0.2631101394685396,"score_spread":0.2339622897742041,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}