{ "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": "2608.09291", "title": "UnionSparse: An Index-Efficient Sparsity Framework for Low-Bit Sparse LLM Inference on Edge", "summary": "UnionSparse replaces the sparse metadata formats and SpMM kernels in existing sparse-inference stacks (FlashLLM, SpInfer, CUTLASS/cuBLAS sparse paths) with an index-efficient bitmap encoding (IE-BME) plus a shared-memory parallel decoding kernel, targeting the payload-to-metadata bottleneck in W4A4 sparse decoding. The cost is adopting a custom kernel and encoding pipeline: weights must be re-encoded into IE-BME, integration work is required outside supported frameworks, and benefits presuppose a model that is already both 4-bit quantized and 30%-70% sparse. It can fail if the workload is dense or weight-only quantized without structured sparsity, if sparsity falls outside the evaluated range, if batch sizes grow beyond the small-batch regime the kernel optimizes, or if the hardware differs from the evaluated edge GPUs.", "rationale": "The speedup is real but conditional on an existing W4A4 sparse model and kernel-level integration, so the reader should validate it on their own model and GPU before adopting rather than treating it as a drop-in replacement.", "technique": "Index-efficient bitmap encoding for sparse low-bit SpMM", "family": "esparsidade_e_poda", "recommendation": "testar", "published": "2026-08-10", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2608.09291", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "UnionSparse replaces the sparse metadata formats and SpMM kernels in existing sparse-inference stacks (FlashLLM, SpInfer, CUTLASS/cuBLAS sparse paths) with an index-efficient bitmap encoding (IE-BME) plus a shared-memory parallel decoding kernel, targeting the payload-to-metadata bottleneck in W4A4 sparse decoding. The cost is adopting a custom kernel and encoding pipeline: weights must be re-encoded into IE-BME, integration work is required outside supported frameworks, and benefits presuppose a model that is already both 4-bit quantized and 30%-70% sparse. It can fail if the workload is dense or weight-only quantized without structured sparsity, if sparsity falls outside the evaluated range, if batch sizes grow beyond the small-batch regime the kernel optimizes, or if the hardware differs from the evaluated edge GPUs.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.09291", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Under W4A4 quantization and 30%-70% sparsity, UnionSparse outperforms FlashLLM and SpInfer by 2.30x and 1.43x, and CUTLASS and cuBLAS Tensor Core baselines by 1.56x and 3.46x respectively in SpMM decoding kernels.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.09291", "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": [ { "anchor": "S3.E1", "label": "(1)", "section": "III Gap Analysis and Motivation", "role": "metric", "latex": "\\small\\mathrm{PMR}=\\frac{B_{w}\\times M\\times K\\times(1-s)}{\\mathrm{Stor}_{\\text{meta}}}", "mathml": "PMR=Bw×M×K×(1−s)Stormeta", "context": "To quantify the efficiency of payload extraction, we define the Payload-to-Metadata Ratio (PMR) as", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "S3.E2", "label": "(2)", "section": "III Gap Analysis and Motivation", "role": "metric", "latex": "\\small{CI}_{\\text{SpMM}}=\\frac{2MN(1-s)}{B_{w}M(1-s)\\left(1+\\frac{1}{\\textsc{PMR}}\\right)+B_{w}N}", "mathml": "C​ISpMM=2​M​N​(1−s)Bw​M​(1−s)​(1+1PMR)+Bw​N", "context": "For sparse low-bit SpMM, metadata overhead reduces the effective compute intensity seen by the hardware. Using PMR, the effective CI can be written as", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "S5.E5", "label": "(5)", "section": "V Representation Optimization", "role": "proposed_method", "latex": "F(S)=\\min_{\\mathbf{r}\\subseteq S,\\;|\\mathbf{r}|=R_{A},\\;r\\in\\mathbf{r}}\\left(c_{t}(\\mathbf{r})+F(S\\setminus\\mathbf{r})\\right).", "mathml": "F⁡(S)=min𝐫⊆S,|𝐫|=RA,r∈𝐫⁡(ct​(𝐫)+F⁡(S∖𝐫)).", "context": "Grouped Subset Dynamic Programming. To support arbitrary RA and CA, UnionSparse solves the row-reordering problem with a subset dynamic programming (DP) procedure. Let S⊆{0,1,…,HL−1} denote the set of rows that have not yet been grouped, with |S| divisible by RA. The DP state value F⁡(S) is defined as the minimum grouped cost achievable on the remaining rows in S. For each state, we select the smallest row index r∈S, enumerate all row groups 𝐫⊆S of size RA that contain r, and minimize", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-09-17", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }