{ "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": "2610.12327", "title": "SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM Inference", "summary": "It replaces Hessian calibration on pre-collected natural text and SpMM-oriented pruning with calibration from dense-model autoregressive decoding activations plus an N:M sparse SpMV kernel. The cost is a pruning/calibration pipeline that must run the dense model generatively, a custom bitmask kernel, and dependence on N:M sparsity being efficient on the target GPU. It can fail through residual activation shift across domains or decoding settings, quality loss after pruning, limited benefit for prefill-heavy or batched serving, and no applicability to third-party API models.", "rationale": "For a 24 GB self-hosted setup this is plausible and directly targets decoding latency, but the reported gain is A100- and kernel-dependent, so it should be validated on the reader's exact model, workload, and hardware before production use.", "technique": "Decoding-aware N:M pruning for LLM inference", "family": "esparsidade_e_poda", "recommendation": "testar", "published": "2026-10-08", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.12327", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "It replaces Hessian calibration on pre-collected natural text and SpMM-oriented pruning with calibration from dense-model autoregressive decoding activations plus an N:M sparse SpMV kernel. The cost is a pruning/calibration pipeline that must run the dense model generatively, a custom bitmask kernel, and dependence on N:M sparsity being efficient on the target GPU. It can fail through residual activation shift across domains or decoding settings, quality loss after pruning, limited benefit for prefill-heavy or batched serving, and no applicability to third-party API models.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.12327", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "The paper reports up to 1.48x end-to-end wall-clock decoding speedup on A100 GPUs and better long-form generation accuracy than fixed-text calibration across Llama and Qwen models.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.12327", "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": "" }