sh2orc commited on
Commit
a8066f7
·
verified ·
1 Parent(s): ed3564e

Upload 2 files

Browse files
Files changed (3) hide show
  1. .gitattributes +1 -0
  2. README.md +89 -3
  3. dataset.jsonl +3 -0
.gitattributes CHANGED
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ dataset.jsonl filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,89 @@
1
- ---
2
- license: mit
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Semantic Router Instruction Dataset
2
+
3
+ semantic-turn router 학습·평가용 한국어 instruction 데이터셋. 한 문장을 14개 축으로 동시에 분류하는 모델의 학습 데이터로 쓰기 위해 만들었고, 각 행의 `instruction`은 해당 축·값을 판별할 수 있는 내용 단서를 실제로 포함한다.
4
+
5
+ - **규모**: 600,000행 (14축 × 150값 전체 커버, 값별 행 수 고정)
6
+ - **언어**: 한국어 (일부 기술 용어·코드·수식은 원표기)
7
+ - **특성**: 전 행 장문 조합형 — 평균 140자, 80자 이상 92.8%, 120자 이상 69.5%
8
+
9
+ ## 파일 형식
10
+
11
+ `dataset.jsonl` — 한 줄에 JSON 객체 하나, UTF-8, 헤더 없음.
12
+
13
+ ```json
14
+ {"instruction":"신규 입사자 온보딩 자료를 만들다가 막혀서, …일정이 촉박해지면 한 페이지 요약본을 맨 앞에 먼저 놓아줘.","category":"presentationSkillId","value":"deck-redesign"}
15
+ ```
16
+
17
+ | 필드 | 타입 | 설명 |
18
+ |---|---|---|
19
+ | `instruction` | string | 사용자 발화(지시문). 유니크 보장(600k 전수 중복 0) |
20
+ | `category` | string | 14개 축 중 하나 |
21
+ | `value` | string | 해당 축의 값 (총 150값) |
22
+
23
+ ## 축·값·할당
24
+
25
+ | 축 | 값 수 | 행 수 | 값 |
26
+ |---|---:|---:|---|
27
+ | `service` | 16 | 96,000 | general, presentation, image, data_analysis, place_search, web_search, knowledge_search, calendar, memo, works, connector, artifact, coding, translation, skill, reasoning |
28
+ | `presentationSkillId` | 56 | 112,000 | executive-strategy, board-update, quarterly-business-review, sales-proposal, investor-pitch, training-explainer, technical-architecture, research-report, marketing-campaign, product-launch, product-roadmap, consulting-diagnosis, all-hands-update, thesis-research, crisis-communication, demo-day-pitch, compliance-training, lesson-lecture, customer-qbr, analyst-briefing, okr-review, esg-report, incident-postmortem, thesis-defense, grant-proposal, workshop-facilitation, competitive-analysis, investor-update, annual-report, pricing-proposal, analytics-readout, customer-onboarding, vendor-selection, market-entry, company-onboarding, case-study-analysis, model-un-position, lab-meeting, self-review, conference-talk, resume-portfolio, nonprofit-pitch, policy-briefing, earnings-brief, m-a-proposal, org-redesign, regulatory-response, course-syllabus, exam-prep, hobby-explainer, travel-itinerary, deck-redesign, deck-exec-polish, deck-board-upgrade, deck-content-cleanup, none |
29
+ | `contentDomain` | 15 | 60,000 | social, play, self_capability, affective_state, creative, support, explore, personal_scope, opinion, code_work, formal_reasoning, supplied_material, explicit_lookup, multi_step_research, unresolved |
30
+ | `action` | 14 | 56,000 | respond, create, revise, import, analyze, search, export, delete, translate, summarize, explain, compare, execute, none |
31
+ | `dialogueAct` | 10 | 40,000 | clarification_answer, option_selection, confirm, decline, correction, continue_output, preference_directive, meta_conversation, undo_action, none |
32
+ | `sourcePolicies` | 7 | 42,000 | web, internal, memory, attachments, conversation_only, auto, none |
33
+ | `memoryIntent` | 7 | 28,000 | informational, transactional, conversational, analytical, personal, navigational, none |
34
+ | `requestedOutputs` | 5 | 34,000 | chat, docx, xlsx, pptx, image |
35
+ | `diagramKind` | 4 | 24,000 | structure, visualization, plan, none |
36
+ | `presentationMode` | 4 | 24,000 | fast, professional, creative, none |
37
+ | `complexity` | 4 | 32,000 | simple, standard, complex, hard |
38
+ | `artifactKind` | 3 | 18,000 | diagram, interactive, none |
39
+ | `analysisMode` | 3 | 18,000 | focused, comprehensive, none |
40
+ | `artifactComplexity` | 2 | 16,000 | simple, complex |
41
+
42
+ 값별 행 수는 위 표의 (행 수 ÷ 값 수)로 균등 배분된다(예: service 값당 6,000행,
43
+ presentationSkillId 값당 2,000행). 한 `instruction`은 정확히 하나의 (category, value)
44
+ 쌍으로 라벨링된다 — 다중 라벨이 아니라, 축마다 별도 행으로 존재한다.
45
+
46
+ ## instruction 구성 (장문 조합형)
47
+
48
+ 모든 행은 아래 순서로 조합된다:
49
+
50
+ ```
51
+ [상황 오프너] (+ [배경절] 40%) + [그룹 고유 요청] + [제약 tail 1~4개] (+ [조건 분기 tail] 35%)
52
+ ```
53
+
54
+ - **상황 오프너**: "지난 3월에 착수한 멤버십 리뉴얼 건이 이번 주 안에 마감이라, " 같은 맥락 절. 업무형 30종 + 일상형 12종.
55
+ - **제약 tail**: "결론부터 먼저 보여주고 근거는 뒤에 붙여줘." 같은 출력 조건.
56
+ - **조건 분기 tail**: "만약 최신 수치를 못 찾으면 작년 3월 기준으로 대체하고 그 사실을 명시해줘." 같은 if/else 분기 — 긴 문장 안에 여러 갈래의 지시를 싣는다.
57
+
58
+ 축의 성격에 따라 6개 **작성 프로필**이 라우팅된다:
59
+
60
+ | 프로필 | 대상 축·값 | 특징 |
61
+ |---|---|---|
62
+ | general | service·sourcePolicies·artifact*·presentation*·requestedOutputs 등 | 업무 오프너 + 일반 tail + 분기 tail |
63
+ | formal | complexity::hard, contentDomain::formal_reasoning | 엄밀성 tail("생략 없이 모든 유도 과정을 보여줘") |
64
+ | code | contentDomain::code_work, service::coding | 엔지니어링 tail("테스트도 함께 붙여줘") |
65
+ | quick | complexity::simple | 간결 요청 + 간결 tail 1~2개 |
66
+ | personal | contentDomain::{social,support,opinion,explore,personal_scope}, memoryIntent::personal | 일상 오프너 + 일반 tail |
67
+ | micro | dialogueAct 전체, memoryIntent::{conversational,navigational,none}, contentDomain::{play,self_capability,affective_state,unresolved}, action::none | 일상 오프너 + 마이크로 발화 + 가벼운 대화 tail |
68
+
69
+ micro 프로필은 "응 맞아" 같은 짧은 발화 자체는 유지하되, 주변 맥락절로 전체 길이를 확보한다 — 라벨 의미(대화 행위)를 훼손하지 않으면서 장문 분포에 맞춘다.
70
+
71
+ ## 길이 분포
72
+
73
+ | 구간 | 행 수 | 비중 |
74
+ |---|---:|---:|
75
+ | 120–160자 | 217,365 | 36.2% |
76
+ | 160–220자 | 191,605 | 31.9% |
77
+ | 80–120자 | 139,221 | 23.2% |
78
+ | 220자+ | 8,534 | 1.4% |
79
+ | 40–80자 | 42,190 | 7.0% |
80
+ | 40자 미만 | 1,085 | 0.2% |
81
+
82
+ 평균 140.2자, 최단 25자(micro 발화), 최장 273자.
83
+
84
+
85
+ ## 주의사항
86
+
87
+ - 템플릿 조합 생성 데이터다. 개별 문장은 문법적으로 자연스럽지만 슬롯 조합의 의미적 어긋남(예: 업무 주제 × 일상 오프너)이 드물게 있을 수 있음.
88
+ 분류 라벨에는 영향이 없으나 자연성 평가에는 고려할 것.
89
+ - 특정 축의 행은 서로 다른 축의 행과 텍스트가 겹칠 수 있다(같은 주제가 여러 축에 등장). instruction은 전 데이터셋에서 전수 유니크 검증 완료.
dataset.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:18a048490960b4d46cff40055d532e641138933ba06f3e60f3fda105d3746bf1
3
+ size 242044342