{ "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": "2609.15528", "title": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "summary": "The method replaces a single shared multilingual vocabulary and its large embedding/output matrices with a large modular tokenizer from which per-language subtokenizers are extracted, so that training and inference use only the vocabulary subset relevant to the deployed languages. The cost is a full pretraining run with a custom batch-sampling strategy over subtokenizers, plus tokenizer training and nonstandard serving logic; it cannot be retrofitted onto existing pretrained checkpoints. It can fail when deployments shift across language subsets not anticipated at training time, when tooling assumes a fixed vocabulary, and when small per-language vocabularies degrade cross-lingual transfer or code-switching robustness.", "rationale": "The technique requires training a model from scratch with a specialized tokenizer-sampling regimen, which is outside the reader's constrained infrastructure, though the memory and inference benefits are worth tracking for future multilingual deployments.", "technique": "Modular language-specific subtokenizers for multilingual LLMs", "family": "arquitetura_eficiente", "recommendation": "observar", "published": "2026-09-14", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.15528", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces a single shared multilingual vocabulary and its large embedding/output matrices with a large modular tokenizer from which per-language subtokenizers are extracted, so that training and inference use only the vocabulary subset relevant to the deployed languages. The cost is a full pretraining run with a custom batch-sampling strategy over subtokenizers, plus tokenizer training and nonstandard serving logic; it cannot be retrofitted onto existing pretrained checkpoints. It can fail when deployments shift across language subsets not anticipated at training time, when tooling assumes a fixed vocabulary, and when small per-language vocabularies degrade cross-lingual transfer or code-switching robustness.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.15528", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "The abstract states that restricting predictions to relevant vocabulary subsets reduces memory usage and speeds up inference without sacrificing performance, with subtokenizer compression on par with monolingual tokenizers, but provides no quantified factor.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.15528", "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.EGx1", "label": "(1)", "section": "3 Inefficiencies of Multilingual Tokenizers", "role": "metric", "latex": "=\\frac{\\sum_{i=1}^{M}\\mathrm{length}\\big(\\mathcal{T}(\\mathcal{D}[i])\\big)}{\\sum_{i=1}^{M}\\mathrm{length}\\big(\\mathcal{T}_{\\text{ref}}^{l}(\\mathcal{D}[i])\\big)},", "mathml": "=∑i=1Mlength⁡(𝒯⁡(𝒟⁡[i]))∑i=1Mlength⁡(𝒯refl​(𝒟⁡[i])),", "context": "Multilingual LLMs require tokenizers with broad language coverage to handle diverse scripts and linguistic structures, but larger vocabularies increase the computational cost of the output layer, forcing tokenizers to operate under limited capacity. This constraint disproportionately affects under-represented languages, which tend to be poorly compressed (Zheng et al., 2021). To quantify the compression efficiency of a tokenizer on a given language ℒ, we use the Normalized Sequence Length (NSL) (Dagan et al., 2024), defined as the ratio of tokenized sequence length to that of a reference tokenizer. For a dataset 𝒟 with M examples in language ℒ, NSL is defined as:", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-10-07", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }