{ "schema_version": 1, "builder_plan": { "scenario": "Your workload might fit a smaller or differently structured model.", "change": "Check for available weights and a compatible runtime before planning a product trial.", "baseline": "Your current model on identical held-out tasks and the same resource budget.", "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 success, latency, memory, context limits, and migration 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": "2610.26747", "title": "recAMPLIFY: Large-Scale Pretraining of Recursive Protein Language Models", "summary": "recAMPLIFY replaces parameter scaling in masked protein language models with a weight-tied recursive Transformer core that refines hidden states over a variable number of cycles, optionally with a Gated Mixture Recurrence decoder for adaptive depth. The recipe requires large-scale pretraining (2T tokens), which exceeds the reader's infrastructure and budget; only released checkpoints, if published, would make GMR's adaptive-depth inference usable, and GMR's accuracy trade-off is task-dependent and can be negative. Recursive weight-tying can become unstable beyond the trained depth (the paper's own gating analysis addresses this), and the method is specific to masked protein encoders, not general LLM workloads.", "rationale": "The technique is a pretraining recipe for protein language models requiring 2T tokens of compute, so it is relevant to efficiency-aware architecture design but cannot be adopted or meaningfully tested on the reader's constrained infrastructure.", "technique": "Recursive weight-tied pretraining for protein language models", "family": "arquitetura_eficiente", "recommendation": "observar", "published": "2026-10-19", "as_of": "2026-10-20", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.26747", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "recAMPLIFY replaces parameter scaling in masked protein language models with a weight-tied recursive Transformer core that refines hidden states over a variable number of cycles, optionally with a Gated Mixture Recurrence decoder for adaptive depth. The recipe requires large-scale pretraining (2T tokens), which exceeds the reader's infrastructure and budget; only released checkpoints, if published, would make GMR's adaptive-depth inference usable, and GMR's accuracy trade-off is task-dependent and can be negative. Recursive weight-tying can become unstable beyond the trained depth (the paper's own gating analysis addresses this), and the method is specific to masked protein encoders, not general LLM workloads.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.26747", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "155M unique parameters (56% fewer than AMPLIFY-350M) trained on the same 2T tokens improves CASP contact recovery and 3 of 4 supervised tasks, raising fold accuracy by 22.5%; GMR reduces expected layer depth by up to 37% with task-dependent accuracy trade-offs.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.26747", "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": "" }