{ "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": "2610.13208", "title": "How Much Computation Does Every Forecast Need? Structured Block Skipping in Two Time-Series Foundation Models", "summary": "The method replaces full-depth forward passes in pretrained time-series foundation models with statically masked Transformer blocks, where whole blocks are bypassed using masks selected on calibration windows. It costs a calibration procedure per deployment setting and carries quality risk: aggressive skipping substantially degrades Traffic and Weather, and the measured latency gain is only 11-13% at the moderate sparsity that preserves accuracy. What can fail is generalization: dataset-specific masks transfer inconsistently, per-window optimal masks differ so no static mask captures the oracle opportunity, and Chronos representations do not support reliable routing to unseen domains.", "rationale": "A modest, workload-dependent latency gain from static block masks is cheap to validate on a single 24 GB GPU, but the paper explicitly shows the dynamic-routing upside is not deployable and mask transfer is unreliable, so adoption requires testing on the reader's own forecasting data before committing.", "technique": "Static structured Transformer block skipping at inference", "family": "esparsidade_e_poda", "recommendation": "testar", "published": "2026-10-09", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.13208", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces full-depth forward passes in pretrained time-series foundation models with statically masked Transformer blocks, where whole blocks are bypassed using masks selected on calibration windows. It costs a calibration procedure per deployment setting and carries quality risk: aggressive skipping substantially degrades Traffic and Weather, and the measured latency gain is only 11-13% at the moderate sparsity that preserves accuracy. What can fail is generalization: dataset-specific masks transfer inconsistently, per-window optimal masks differ so no static mask captures the oracle opportunity, and Chronos representations do not support reliable routing to unseen domains.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.13208", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Chronos-Bolt with 7/11 blocks and TimesFM 2.5 with 16/20 blocks reduce median A100 batch-1 latency by 11.2% and 13.4% respectively with little change in mean relative forecasting quality across four datasets.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.13208", "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": "" }