{ "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": "2610.01434", "title": "MWOP: Modality-aware Width-wise Operation Pruning for Efficient MLLMs", "summary": "MWOP replaces dense attention and FFN execution in multimodal LLMs by independently pruning V2V, T2V, and T2T attention paths and selecting FFN channels separately per modality, while preserving the token sequence, using path-sparse Triton kernels for actual acceleration. It costs a one-time pruning pipeline with first-order Taylor importance estimation, FFN re-evaluation, LoRA recovery training, and dependence on custom Triton kernels that must be maintained per model. It can fail if the custom kernels are not ported to new architectures or serving stacks, if accuracy degrades on visual tasks underrepresented in the 12-benchmark average, or if the reader's workloads run on third-party APIs where weights cannot be modified.", "rationale": "The reported prefill speedup is meaningful on a single 24GB GPU for self-hosted 7B MLLMs, but adoption requires running the pruning and LoRA recovery pipeline plus custom Triton kernels, so workload-specific validation on the reader's exact model and tasks is needed before production use.", "technique": "Modality-aware width-wise operation pruning for MLLMs", "family": "esparsidade_e_poda", "recommendation": "testar", "published": "2026-10-01", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.01434", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "MWOP replaces dense attention and FFN execution in multimodal LLMs by independently pruning V2V, T2V, and T2T attention paths and selecting FFN channels separately per modality, while preserving the token sequence, using path-sparse Triton kernels for actual acceleration. It costs a one-time pruning pipeline with first-order Taylor importance estimation, FFN re-evaluation, LoRA recovery training, and dependence on custom Triton kernels that must be maintained per model. It can fail if the custom kernels are not ported to new architectures or serving stacks, if accuracy degrades on visual tasks underrepresented in the 12-benchmark average, or if the reader's workloads run on third-party APIs where weights cannot be modified.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.01434", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "1.6x prefill speedup with 99.7% average performance retention on LLaVA-OneVision-7B across 12 benchmarks; combined with token compression, prefill speedups rise from 2.0x and 1.9x to 2.9x and 2.7x; also validated on Qwen2.5-VL-7B.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.01434", "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": "" }