id stringlengths 5 5 | kind stringclasses 1
value | language stringclasses 1
value | query stringlengths 7 160 | tools listlengths 5 5 | expected_tool_call dict | answer null | messages listlengths 3 3 | conversations listlengths 3 3 |
|---|---|---|---|---|---|---|---|---|
s1970 | tool | 繁體中文 | 我住在緯度 25.0478、經度 121.5170 附近，請問我在這邊可以飛無人機嗎？ | [
{
"type": "function",
"function": {
"name": "get_dataset",
"description": "取得資料集完整 metadata 與樣本資料列。\n\n參數：\n dataset_id: 從 search_datasets 取得的 dataset_id\n sample_rows: 樣本列數（v1.45.0: 預設 3，硬 cap 3）。設 0 跳過樣本\n\n回傳：metadata（含 schema.columns、schema.row_count、license、source.url、\n join_keys、... | {
"name": "dronegis_check_point",
"arguments": {
"lat": 25.0478,
"lon": 121.517
}
} | null | [
{
"role": "system",
"content": "You are a helpful assistant that can call tools when needed."
},
{
"role": "user",
"content": "我住在緯度 25.0478、經度 121.5170 附近，請問我在這邊可以飛無人機嗎？"
},
{
"role": "assistant",
"content": null,
"tool_calls": [
{
"id": "call_s1970",
"type... | [
{
"from": "system",
"value": "You are a helpful assistant with access to the following tools:\n\n[\n {\n \"name\": \"get_dataset\",\n \"description\": \"取得資料集完整 metadata 與樣本資料列。\\n\\n參數：\\n dataset_id: 從 search_datasets 取得的 dataset_id\\n sample_rows: 樣本列數（v1.45.0: 預設 3，硬 cap 3）。設 0 跳過樣本\\n\\n回傳... |
s1132 | tool | 繁體中文 | 我打算研究一下台北市的餐飲衛生相關資料，請幫我把這份名為 'tp-00001234' 的資料集先下載並轉換好，這樣我稍後查資料才不會因為要等待下載而變慢。 | [
{
"type": "function",
"function": {
"name": "list_domains",
"description": "列出全部 19 個 domain 標籤的定義（key、中文名、scope、典型問題、anchor 範例）。\n\n用於 LLM 在呼叫 search_datasets 之前確認該用哪個 domain 過濾。\n",
"parameters": {
"properties": {},
"title": "list_domainsArguments",
"type": "objec... | {
"name": "materialize_dataset",
"arguments": {
"dataset_id": "tp-00001234"
}
} | null | [
{
"role": "system",
"content": "You are a helpful assistant that can call tools when needed."
},
{
"role": "user",
"content": "我打算研究一下台北市的餐飲衛生相關資料，請幫我把這份名為 'tp-00001234' 的資料集先下載並轉換好，這樣我稍後查資料才不會因為要等待下載而變慢。"
},
{
"role": "assistant",
"content": null,
"tool_calls": [
{
... | [
{
"from": "system",
"value": "You are a helpful assistant with access to the following tools:\n\n[\n {\n \"name\": \"list_domains\",\n \"description\": \"列出全部 19 個 domain 標籤的定義（key、中文名、scope、典型問題、anchor 範例）。\\n\\n用於 LLM 在呼叫 search_datasets 之前確認該用哪個 domain 過濾。\\n\",\n \"parameters\": {\n \"pro... |
s1451 | tool | 繁體中文 | 我想知道全台灣有哪些雨量測站的位置清單，請問有什麼方法可以取得這些地點資訊嗎？ | [
{
"type": "function",
"function": {
"name": "search_ly_interpellations",
"description": "立法院質詢語意檢索 (~12k 質詢, 事由+說明合併向量).\n\nArgs:\n query: 自然語言 (e.g. \"長照經費\", \"兒少安置\")\n legislator_contains: 質詢委員字串 (e.g. \"王婉諭\")\n term: 屆別\n session: 會期\n limit: top-K (1..100)\n\nReturns:\n ... | {
"name": "rt_wra_rain_station_list",
"arguments": {}
} | null | [
{
"role": "system",
"content": "You are a helpful assistant that can call tools when needed."
},
{
"role": "user",
"content": "我想知道全台灣有哪些雨量測站的位置清單，請問有什麼方法可以取得這些地點資訊嗎？"
},
{
"role": "assistant",
"content": null,
"tool_calls": [
{
"id": "call_s1451",
"type": "... | [
{
"from": "system",
"value": "You are a helpful assistant with access to the following tools:\n\n[\n {\n \"name\": \"search_ly_interpellations\",\n \"description\": \"立法院質詢語意檢索 (~12k 質詢, 事由+說明合併向量).\\n\\nArgs:\\n query: 自然語言 (e.g. \\\"長照經費\\\", \\\"兒少安置\\\")\\n legislator_contains: 質詢委員字串 (e.g.... |
s0576 | tool | 繁體中文 | 我正在規劃亞洲藝術史的遠距工作專題，想找一下歷史博物館館藏中，由張大千先生創作、屬於繪畫類的相關作品，可以幫我列出前 5 筆嗎？ | [
{
"type": "function",
"function": {
"name": "search_drug_label",
"description": "搜 twinkle-ai/tw-drug-labels-vision (CC-BY-4.0, ~72k 藥品仿單) 結構化欄位.\n\n對打 Taiwan-Health-MCP search_drug 的「依適應症 / 禁忌 / 副作用 / 警語」\n關鍵字精準搜尋. 比 search_drug(indication=...) 更全面 (包含禁忌/警語/\n副作用/交互作用 等仿單欄位, 不只適應症).\n\nArgs:\n ... | {
"name": "search_nmh_artifacts",
"arguments": {
"query": "張大千繪畫作品",
"category": "繪畫",
"creator_contains": "張大千",
"limit": 5
}
} | null | [
{
"role": "system",
"content": "You are a helpful assistant that can call tools when needed."
},
{
"role": "user",
"content": "我正在規劃亞洲藝術史的遠距工作專題，想找一下歷史博物館館藏中，由張大千先生創作、屬於繪畫類的相關作品，可以幫我列出前 5 筆嗎？"
},
{
"role": "assistant",
"content": null,
"tool_calls": [
{
"id": "call_... | [
{
"from": "system",
"value": "You are a helpful assistant with access to the following tools:\n\n[\n {\n \"name\": \"search_drug_label\",\n \"description\": \"搜 twinkle-ai/tw-drug-labels-vision (CC-BY-4.0, ~72k 藥品仿單) 結構化欄位.\\n\\n對打 Taiwan-Health-MCP search_drug 的「依適應症 / 禁忌 / 副作用 / 警語」\\n關鍵字精準搜尋. 比 se... |
s1777 | tool | 繁體中文 | 請幫我查申請案號 113105522 的專利詳細內容 | [
{
"type": "function",
"function": {
"name": "get_patent_body",
"description": "取得單一專利的完整 description body (技術領域 / 先前技術 / 實施方式)\n+ 全部 claims。從 TIPO ftps 即時下載解析 (非 cache, 200-500 ms)。\n\n參數:\n appl_no: 申請案號 (例 \"112108163\"), 通常從 search_patents hit 取得\n\n回傳:\n {\n appl_no, title_zh, abs... | {
"name": "get_patent_body",
"arguments": {
"appl_no": "113105522"
}
} | null | [
{
"role": "system",
"content": "You are a helpful assistant that can call tools when needed."
},
{
"role": "user",
"content": "請幫我查申請案號 113105522 的專利詳細內容"
},
{
"role": "assistant",
"content": null,
"tool_calls": [
{
"id": "call_s1777",
"type": "function",
... | [
{
"from": "system",
"value": "You are a helpful assistant with access to the following tools:\n\n[\n {\n \"name\": \"get_patent_body\",\n \"description\": \"取得單一專利的完整 description body (技術領域 / 先前技術 / 實施方式)\\n+ 全部 claims。從 TIPO ftps 即時下載解析 (非 cache, 200-500 ms)。\\n\\n參數:\\n appl_no: 申請案號 (例 \\\"1121... |
s1435 | tool | 繁體中文 | 我現在在台北市的經緯度 25.0410, 121.5641 這個位置，想拍一些餐廳的空拍畫面，請問這邊可以用無人機嗎？ | [
{
"type": "function",
"function": {
"name": "get_teacher_recruit_paper",
"description": "取得單一教甄試卷的全部選擇題 + 標準答案.\n\n參數:\n paper_id: 從 search_teacher_recruit(_questions) hit 取得\n (格式 `tjse_{county}_{year}_{stage}_{subject}`,\n e.g. \"tjse_taipei_113_es_音樂\")\n incl... | {
"name": "dronegis_check_point",
"arguments": {
"lat": 25.041,
"lon": 121.5641
}
} | null | [
{
"role": "system",
"content": "You are a helpful assistant that can call tools when needed."
},
{
"role": "user",
"content": "我現在在台北市的經緯度 25.0410, 121.5641 這個位置，想拍一些餐廳的空拍畫面，請問這邊可以用無人機嗎？"
},
{
"role": "assistant",
"content": null,
"tool_calls": [
{
"id": "call_s1435... | [
{
"from": "system",
"value": "You are a helpful assistant with access to the following tools:\n\n[\n {\n \"name\": \"get_teacher_recruit_paper\",\n \"description\": \"取得單一教甄試卷的全部選擇題 + 標準答案.\\n\\n參數:\\n paper_id: 從 search_teacher_recruit(_questions) hit 取得\\n (格式 `tjse_{county}_{year}_... |
s1645 | tool | 繁體中文 | "你好，我正在整理氣象資料，請幫我查詢一下目前「臺中」自動氣象站的觀(...TRUNCATED) | [{"type":"function","function":{"name":"rt_cwa_realtime_rainfall","description":"即時自動氣象(...TRUNCATED) | {
"name": "rt_cwa_realtime_rainfall",
"arguments": {
"locationName": "臺中",
"limit": 10
}
} | null | [{"role":"system","content":"You are a helpful assistant that can call tools when needed."},{"role":(...TRUNCATED) | [{"from":"system","value":"You are a helpful assistant with access to the following tools:\n\n[\n {(...TRUNCATED) |
s1228 | tool | 繁體中文 | "我這週末在嘗試調整飲食，想知道每 100 克的鮭魚裡面大約含有多少毫克的(...TRUNCATED) | [{"type":"function","function":{"name":"get_exam_paper","description":"取得單一國考試卷的(...TRUNCATED) | {
"name": "query_food_nutrition",
"arguments": {
"food_name": "鮭魚",
"nutrient": "膽固醇"
}
} | null | [{"role":"system","content":"You are a helpful assistant that can call tools when needed."},{"role":(...TRUNCATED) | [{"from":"system","value":"You are a helpful assistant with access to the following tools:\n\n[\n {(...TRUNCATED) |
s0869 | tool | 繁體中文 | "我們團隊正在分析各縣市教師甄選的命題趨勢，請幫我找一下臺北市在 2024 (...TRUNCATED) | [{"type":"function","function":{"name":"search_fsc_bank_law_articles","description":"銀行局法規(...TRUNCATED) | {
"name": "search_teacher_recruit",
"arguments": {
"county": "taipei",
"stage": "es",
"year_from": 2024
}
} | null | [{"role":"system","content":"You are a helpful assistant that can call tools when needed."},{"role":(...TRUNCATED) | [{"from":"system","value":"You are a helpful assistant with access to the following tools:\n\n[\n {(...TRUNCATED) |
s1183 | tool | 繁體中文 | "我正在做一個週末整理藥物知識的專案，想找出仿單中提到有關於「失眠」(...TRUNCATED) | [{"type":"function","function":{"name":"search_datasets","description":"**Search Taiwan government o(...TRUNCATED) | {
"name": "search_drug_label",
"arguments": {
"keyword": "失眠",
"field": "adverse_reactions",
"limit": 30
}
} | null | [{"role":"system","content":"You are a helpful assistant that can call tools when needed."},{"role":(...TRUNCATED) | [{"from":"system","value":"You are a helpful assistant with access to the following tools:\n\n[\n {(...TRUNCATED) |
End of preview. Expand in Data Studio
twinkle_hub_finetune_dataset
MCP tool-calling SFT 資料集，由 Agent Tools Fine-Tuning Platform 以「反向生成 + teacher solver 驗證」流程產生。
- 語言：繁體中文
- 工具（來自 MCP server）：
search_datasets,get_dataset,query_rows,materialize_dataset,search_patents,get_patent_body,search_exam,search_exam_questions,get_exam_paper,search_teacher_exam,search_teacher_exam_questions,get_teacher_exam_paper,search_teacher_recruit,search_teacher_recruit_questions,get_teacher_recruit_paper,search_taiwan_md,get_taiwan_md_article,search_judicial,get_judicial_full,search_ly_bills,search_ly_interpellations,search_ly_gazette_agendas,search_fsc_regulations,search_fsc_bank_law_articles,search_fsc_bank_interpretations,search_cultural_heritage,search_legal_interpretations,search_nmh_artifacts,search_niche_corpora,lookup_icd10,search_drug,get_drug_details,query_food_nutrition,search_health_supplements,search_drug_label,check_drug_interaction,search_foods_by_nutrient,analyze_meal_nutrition,list_domains,dronegis_check_point,dronegis_nearby,rt_cwa_realtime_rainfall,rt_cwa_24h_rainfall,rt_cwa_typhoon_status,rt_cwa_radar_image_url,rt_cwa_36h_forecast,rt_ncdr_active_alerts,rt_reservoir_live,rt_wra_water_level,rt_wra_groundwater_level,rt_wra_river_station_list,rt_wra_rain_station_list - Train / Test：1002 / 251
- 負樣本比例：0.0
- Teacher 資料品質：函式名正確率
0.9834、名稱+參數全對率0.6296
檔案
| 檔案 | 筆數 |
|---|---|
train.jsonl |
1002 |
test.jsonl |
251 |
分層切分（tool / no-tool 各自切），load_dataset 會自動對應成 train / test split。
欄位
| 欄位 | 說明 |
|---|---|
id |
樣本 ID |
kind |
tool（需呼叫工具）/ no_tool（不需呼叫） |
language |
生成語言 |
query |
使用者請求 |
tools |
可用工具（OpenAI function schema 陣列） |
expected_tool_call |
標準答案的工具呼叫；no_tool 為 null |
answer |
no_tool 樣本的文字回答；tool 為 null |
messages |
OpenAI 對話格式（assistant 帶 tool_calls） |
conversations |
ShareGPT 格式（system 含工具清單、gpt 以 <tool_call> 輸出） |
生成方式
先從 MCP 工具 schema 抽出「目標工具 + 合法參數」作為 ground truth，再由 teacher model 生成對應的自然語言使用者問題；另由 teacher 以 solver 身分重新解題做品質過濾（只保留還原正確者）。因此 expected_tool_call 可作為可精確比對的標準答案，用來計算有工具呼叫率 / 函式名正確率 / 名稱+參數全對率。
載入
from datasets import load_dataset
ds = load_dataset("Simon-Liu/twinkle_hub_finetune_dataset")
print(ds["train"][0])
- Downloads last month
- 29