{ "schema": "nuosu_task_catalog/1.0", "version": "2026.08.04", "author": "Wuhe Axi", "affiliation": "Xi'an Jiaotong University", "dataset_id": "NiceAsiv/nuosu-corpus", "organization_rule": "Training objective is encoded by format; corpus purpose is encoded by task; source and processing method stay in metadata.", "training_entrypoints": { "sft": "ready_sft.jsonl", "sft_validation": "validation_sft.jsonl", "continued_pretraining": "ready_cpt.jsonl", "research_test": "research_test_eval.jsonl" }, "task_definitions": { "translation": { "format": "sft", "description": "Sentence, phrase, or lexical translation between Nuosu and another language.", "growth_priority": "high" }, "single_turn_qa": { "format": "sft", "description": "A single user request followed by one grounded Nuosu answer.", "growth_priority": "highest" }, "multi_turn_dialogue": { "format": "sft", "description": "Multi-turn natural dialogue with stable speaker context.", "growth_priority": "highest" }, "pronunciation_orthography": { "format": "sft", "description": "Standard Yi character, standard pinyin, and IPA alignment.", "growth_priority": "medium" }, "terminology_translation": { "format": "sft", "description": "Standard terminology and user-interface label translation.", "growth_priority": "medium" }, "monolingual_language_modeling": { "format": "cpt", "description": "Continuous Nuosu text for continued pretraining.", "growth_priority": "high" }, "orthography_reference": { "format": "reference", "description": "Normative Unicode inventory used for validation and tokenizer coverage.", "growth_priority": "complete_reference" } }, "task_stage_counts": { "monolingual_language_modeling": { "excluded": 55, "ready": 4238 }, "orthography_reference": { "ready": 1165, "reference_only": 55 }, "pronunciation_orthography": { "ready": 1770 }, "single_turn_qa": { "ready": 365 }, "terminology_translation": { "ready": 180 }, "translation": { "ready": 199441, "validation": 10839 } }, "format_stage_counts": { "cpt": { "excluded": 55, "ready": 4238 }, "reference": { "ready": 1165, "reference_only": 55 }, "sft": { "ready": 201756, "validation": 10839 } }, "known_gaps": { "single_turn_qa": "365 Wiktionary-derived records are ready; native-authored and source-grounded general QA is still missing.", "multi_turn_dialogue": "0 records; prioritize natural multi-turn conversations.", "continued_pretraining_ready": "4,238 usable GT, OCR, PDF, web, and romanized records are ready.", "sentence_translation": "Most translation volume is dictionary phrases; collect complete parallel sentences." }, "omitted_representations": { "research_gated/ocr/pages_all.jsonl": "not copied because each OCR page already has exactly one routed task record", "yidir": "excluded by project scope", "NuosuBench internal_test.jsonl": "research_test_eval.jsonl is the canonical prompt/reference projection" }, "training_policy": { "benchmark_overlap": "audit metadata only; never excludes an otherwise usable record", "excluded_only_when": [ "no Standard Yi content", "empty text" ] }, "derived_text_sources": { "nuosubburma_ocr_ground_truth": { "records": 1030, "ready": 1030, "same_work_overlap_flagged": 142, "images_copied": 0, "usage": "human-verified GT text reused as an LLM CPT candidate" }, "nuosu_bench_academic_research_split": { "source_dataset": "TianYeZ1214/NuosuBench", "source_revision": "0709c1fd4f6eaabd4b058ea419027ccc42731dee", "split_method": "sha256(instruction,input,output)", "train": 88487, "validation": 10839, "research_test": 11171, "policy": "train participates in gradients; validation selects checkpoints; research_test remains excluded" } }, "files": { "excluded.jsonl": 55, "manual_review.jsonl": 0, "normative_reference.jsonl": 1220, "quarantine_cpt.jsonl": 0, "quarantine_sft.jsonl": 0, "ready_cpt.jsonl": 4238, "ready_sft.jsonl": 201756, "review_cpt.jsonl": 0, "review_sft.jsonl": 0, "validation_sft.jsonl": 10839, "research_test_eval.jsonl": 11171 }, "unique_records": 218108, "validation_errors": [], "passed": true }