{ "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": "2610.21383", "title": "Calibration Language Shapes Block-Influence Pruning in Multilingual Language Models", "summary": "It replaces English-only calibration of Block-Influence scores for depth pruning with equal-budget multilingual calibration when pruning small models for Indian-language workloads. The cost is obtaining disjoint in-language calibration data and recomputing BI per language before averaging; no extra inference-time memory or latency is added, and pruning itself reduces deployment footprint. It can fail per architecture: on Llama-3.2-1B the balanced calibration severely harms held-out Tamil at two removed blocks, and at larger pruning budgets the selected blocks converge and perplexity collapses regardless of calibration.", "rationale": "The reader can run this cheaply on a single 24 GB GPU if they prune multilingual models, but the paper itself frames the result as a diagnostic requiring per-architecture and per-budget validation, so adoption without workload-specific testing is not justified.", "technique": "Multilingual calibration for Block-Influence depth pruning", "family": "esparsidade_e_poda", "recommendation": "testar", "published": "2026-10-15", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.21383", "independent_implementations": 0, "total_implementations": 2, "stars": 123, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "It replaces English-only calibration of Block-Influence scores for depth pruning with equal-budget multilingual calibration when pruning small models for Indian-language workloads. The cost is obtaining disjoint in-language calibration data and recomputing BI per language before averaging; no extra inference-time memory or latency is added, and pruning itself reduces deployment footprint. It can fail per architecture: on Llama-3.2-1B the balanced calibration severely harms held-out Tamil at two removed blocks, and at larger pruning budgets the selected blocks converge and perplexity collapses regardless of calibration.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.21383", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Late-block BI is 20.7-55.5% higher for Hindi, Marathi, and Tamil than for English; equal-budget English-Hindi-Marathi calibration mitigates post-pruning perplexity damage for Qwen at six removed blocks and for Gemma at all tested depths, including held-out Tamil, but the effect is architecture- and budget-dependent rather than a universal factor.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.21383", "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": "" }