{ "schema_version": 1, "builder_plan": { "scenario": "A self-hosted model uses more compute or memory than your workload can justify.", "change": "Check for a usable pruned checkpoint and runtime support before attempting model surgery.", "baseline": "The original unpruned model at the same precision, context, and load.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Task quality, actual GPU memory, serving throughput, and preparation cost.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2609.38878", "title": "Audio Token Attention Is Predictable Before the Language Model Runs", "summary": "The method replaces full prefill of all 750-1,500 audio tokens per minute of speech with a closed-form linear predictor, fitted without labels, that ranks encoder outputs by their future attention and prunes before the language model runs, with a second cut at layer 2 for multiple-choice tasks. It costs a small offline fitting step and a bounded quality risk, mitigated by two label-free budgets that constrain deviation from the model's full-audio output, but adds a pruning component to the serving path. It can fail on models where the linear predictability assumption does not hold (2 of 13 LALMs fell below rho = .69), on audio domains far from the fitted distribution, or when the aggressive budget degrades accuracy on tasks beyond those evaluated.", "rationale": "The approach requires no training infrastructure and directly reduces prefill cost on a single GPU, but it is a single recent paper evaluated on a fixed set of LALMs, so production adoption should follow workload-specific validation of the quality budgets.", "technique": "Pre-LLM audio token pruning via linear attention prediction", "family": "esparsidade_e_poda", "recommendation": "testar", "published": "2026-09-30", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.38878", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces full prefill of all 750-1,500 audio tokens per minute of speech with a closed-form linear predictor, fitted without labels, that ranks encoder outputs by their future attention and prunes before the language model runs, with a second cut at layer 2 for multiple-choice tasks. It costs a small offline fitting step and a bounded quality risk, mitigated by two label-free budgets that constrain deviation from the model's full-audio output, but adds a pruning component to the serving path. It can fail on models where the linear predictability assumption does not hold (2 of 13 LALMs fell below rho = .69), on audio domains far from the fitted distribution, or when the aggressive budget degrades accuracy on tasks beyond those evaluated.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.38878", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "At its most compressive point, Triage lets one GPU serve 4x as many concurrent 5-minute audio streams of Qwen2.5-Omni-3B; at conservative settings quality stays within 0.04 WER/accuracy of full audio, and context capacity rises from 21.8 to about 62 minutes.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.38878", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }