{ "schema_version": 1, "builder_plan": { "scenario": "A narrow, repeated product task may not need your largest model.", "change": "Check whether a released student model fits the task before generating training data.", "baseline": "Your current model and an off-the-shelf small model on the same held-out tasks.", "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": "Student quality, serving cost, teacher calls, training cost, and failures on unusual inputs.", "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": "2609.29382", "title": "Decoupled Early Exits for Task-Dependent Compute Allocation in Flow-Matching VLAs", "summary": "The method replaces fixed full-depth execution of the VLM backbone and action expert with jointly configurable axes: backbone depth V, action-expert depth A, and denoising steps D, using lightweight Exit Transformers distilled from the final layer plus a KV-cache synthesis mechanism to handle skipped backbone layers. It costs per-exit distillation training (no retraining of the base policy), a 2.1-4.1% parameter increase per exit, and added complexity in cache management and per-task configuration selection. It can fail because the optimal (V,A,D) budget is task-dependent, so a configuration tuned on one task may degrade success rate on another, and gains are validated only on two VLAs and two simulation benchmarks, leaving real-robot and out-of-distribution behavior unverified.", "rationale": "The gains are substantial and cheap to attach, but adoption requires an existing pretrained flow-matching VLA policy and robotics control workload, and the task-dependent tuning burden makes it premature for a general constrained-infrastructure production stack.", "technique": "Decoupled early exits with KV-cache synthesis in flow-matching VLAs", "family": "destilacao", "recommendation": "observar", "published": "2026-09-24", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.29382", "independent_implementations": 2, "total_implementations": 3, "stars": 107, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces fixed full-depth execution of the VLM backbone and action expert with jointly configurable axes: backbone depth V, action-expert depth A, and denoising steps D, using lightweight Exit Transformers distilled from the final layer plus a KV-cache synthesis mechanism to handle skipped backbone layers. It costs per-exit distillation training (no retraining of the base policy), a 2.1-4.1% parameter increase per exit, and added complexity in cache management and per-task configuration selection. It can fail because the optimal (V,A,D) budget is task-dependent, so a configuration tuned on one task may degrade success rate on another, and gains are validated only on two VLAs and two simulation benchmarks, leaving real-robot and out-of-distribution behavior unverified.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.29382", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Joint (V,A,D) configurations reduce latency by 79.2% and FLOPs by 31.8% while improving mean success rate by 5.6 points on LIBERO and Meta-World with SmolVLA and pi_0.5; each exit adds 2.1% (SmolVLA) or 4.1% (pi_0.5) parameters.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.29382", "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": "" }