id int64 100k 101k | query stringlengths 36 377 | tools stringlengths 451 3.96k | query_zhtw stringlengths 14 240 | think stringlengths 174 795 | answer stringlengths 85 673 | messages listlengths 3 3 |
|---|---|---|---|---|---|---|
100,003 | Assess DBR capacity for monthly income 230000, existing payments 20000, unsecured balance 1870000, new loan 2640000. | [{"name": "credit_capacity_dbr", "description": "Assess a borrower's unsecured lending capacity in Taiwan against the DBR 22x regulatory guidance and the monthly debt service ratio, given income, existing payments, outstanding unsecured debt and a proposed new loan.", "parameters": {"monthly_income": {"description": "A... | 我月收入 230000 元、現有月付 20000、無擔保負債 1870000,想再借 2640000,DBR 會超標嗎? | 使用者的問題是評估無擔保借款能力是否超過 DBR 上限。工具清單裡只提供了 `credit_capacity_dbr`,其用途與此需求相符,選它沒有問題。
參數的部分這樣對:平均月收入對應到 `monthly_income`,值是 230000。`existing_monthly_payment`(既有每月還款額)帶 20000。`unsecured_debt_balance` 設為 1870000,也就是題目說的無擔保負債餘額。`new_loan_amount`(擬申請的新貸款金額)帶 2640000。型別上 `existing_monthly_payment` 是 int,格式別弄錯。
另外要記住的規則:金管會規範無擔保債務... | [{"name": "credit_capacity_dbr", "arguments": {"monthly_income": 230000, "existing_monthly_payment": 20000, "unsecured_debt_balance": 1870000, "new_loan_amount": 2640000}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,004 | Check deposit insurance coverage for balances of 6620000 and 3750000 at 華南銀行. In addition, credit card interest math: balance 140000, apr 14.82, days 90, min payment 0.1. | [{"name": "labor_insurance_annuity", "description": "Calculate the Taiwan Labor Insurance old-age monthly annuity, taking the higher of formula A (avg insured salary x years x 0.775% + 3000) and formula B (avg insured salary x years x 1.55%), then applying the 4%-per-year early/deferred adjustment capped at 20%.", "par... | 同一家華南銀行放 6620000 元台幣加 3750000 元外幣,中央存保賠付上限是多少?同時,請算信用卡循環信用利息,未清償餘額 140000、年利率 14.82%、期間 90 天。 | 讓我理一下。需求是確認存款受存款保險保障的額度,共有 4 個候選函式。`labor_insurance_annuity`、`taiwan_regulation_lookup` 的功能與本題情境不符,正確的是 `deposit_insurance_coverage`,因為它就是用來查核存款保險保障額度的。問題其實包了兩個不同的需求;第二段要算信用卡循環信用利息,對應的是 `credit_card_revolving_interest`。
接下來確認每個參數要帶什麼:`bank` 就是要保機構名稱,這裡填 "華南銀行"。`ntd_deposit` 就是台幣存款餘額,這裡填 6620000。`fx_deposit_twd_equiv` ... | [{"name": "deposit_insurance_coverage", "arguments": {"bank": "華南銀行", "ntd_deposit": 6620000, "fx_deposit_twd_equiv": 3750000}}, {"name": "credit_card_revolving_interest", "arguments": {"balance": 140000, "apr": 14.82, "days": 90, "min_payment_rate": 0.1}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,005 | Compute insurance shortfall given income 1000000 and debts 15630000. | [{"name": "life_insurance_gap", "description": "Estimate a life insurance coverage gap using either the income replacement method or the needs-based method, accounting for dependants' support years, existing policies and outstanding liabilities.", "parameters": {"annual_income": {"description": "Annual income of the in... | 請幫我看壽險買夠了沒,年薪 1000000、負債 15630000、目前保額 9810000。 | 讓我理一下。需求是算壽險保額缺口,掃過 2 個工具後,`retirement_gap_analysis` 明顯屬於其他業務範疇;`life_insurance_gap` 才是負責評估壽險保障缺口的那一個,就用它。
參數的部分這樣對:`annual_income`(被保險人年收入)帶 1000000。`existing_coverage`(既有壽險保額)帶 9810000。`outstanding_liabilities` 就是未償負債,這裡填 15630000。注意 `existing_coverage` 的型別是 int,不要寫成字串。
檢查一遍沒有遺漏必要參數,就照 schema 回傳。 | [{"name": "life_insurance_gap", "arguments": {"annual_income": 1000000, "existing_coverage": 9810000, "outstanding_liabilities": 15630000}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,006 | Screen a cash_withdrawal of 15000000 USD to KH for a medium risk customer under Taiwan AML rules. | [{"name": "aml_transaction_screening", "description": "Screen a banking transaction against Taiwan anti-money-laundering rules, checking the NT$500,000 large cash transaction reporting threshold, high-risk jurisdiction flags and whether a suspicious transaction report is warranted.", "parameters": {"amount": {"descript... | 客戶要現金提領 15000000 元(USD),對方國別 KH,風險等級 medium,需要做大額通報嗎? | 先看清楚使用者要什麼。這題是要判斷一筆交易是否需要洗錢防制申報,共有 2 個候選函式。`taiwan_regulation_lookup` 的功能與本題情境不符,正確的是 `aml_transaction_screening`,因為它就是用來執行洗錢防制檢核的。
接著逐一對應參數:`amount` 就是交易金額,這裡填 15000000。`transaction_type`(交易類型)帶 "cash_withdrawal"。幣別對應到 `currency`,值是 "USD"。`counterparty_country` 設為 "KH",也就是題目說的交易對手國別。`customer_risk_level`(客戶風險等級)帶 "me... | [{"name": "aml_transaction_screening", "arguments": {"amount": 15000000, "transaction_type": "cash_withdrawal", "currency": "USD", "counterparty_country": "KH", "customer_risk_level": "medium"}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,007 | Get a credit bureau report with a 12-month inquiry window. | [{"name": "jcic_credit_report", "description": "Request a Joint Credit Information Center credit report for a stated purpose, choosing the report scope and the look-back window for recent credit inquiries.", "parameters": {"query_purpose": {"description": "Purpose of the inquiry, e.g. 'mortgage_application'.", "type": ... | 調一份信用貸款用的聯合徵信中心信用報告(full),回溯 12 個月。 | 使用者的問題是申請聯徵信用報告。工具清單裡有 2 個函式,其中 `aml_transaction_screening` 分別處理法令遵循,跟這題無關;只有 `jcic_credit_report` 的描述對應調閱聯徵信用報告,所以選它。
參數的部分這樣對:`query_purpose` 設為 "personal_loan",也就是題目說的查詢目的。`report_type` 就是報告種類,這裡填 "full"。`recent_inquiries_months` 就是被查詢紀錄回溯月數,這裡填 12。
另外要記住的規則:聯徵中心每年提供 3 次免費查詢;短期內被銀行查詢次數過多會影響核貸與利率。
格式上以 tool_call 包... | [{"name": "jcic_credit_report", "arguments": {"query_purpose": "personal_loan", "report_type": "full", "recent_inquiries_months": 12}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,008 | What is the protected and unprotected amount at 台灣銀行 given deposits of 2430000 and 1920000? | [{"name": "twse_quote_lookup", "description": "Look up the daily quote (open, high, low, close, volume) of a Taiwan-listed security on TWSE or TPEx.", "parameters": {"symbol": {"description": "Taiwan stock ticker, e.g. '2330'.", "type": "str", "default": "2330"}, "date": {"description": "Trading date in YYYY-MM-DD.", "... | 請問我在 台灣銀行 的存款有多少受到存款保險保障?台幣 2430000、外幣折算 1920000。 | 使用者的問題是確認存款受存款保險保障的額度。掃過 3 個工具後,`twse_quote_lookup`、`fx_rate_lookup` 明顯屬於其他業務範疇;`deposit_insurance_coverage` 才是負責查核存款保險保障額度的那一個,就用它。
接下來確認每個參數要帶什麼:`bank` 就是要保機構名稱,這裡填 "台灣銀行"。台幣存款餘額對應到 `ntd_deposit`,值是 2430000。`fx_deposit_twd_equiv` 設為 1920000,也就是題目說的外幣存款折台幣金額。
另外要記住的規則:不同金融機構分別計算,因此分散存放可提高受保額度。
格式上以 tool_call 包住 JSO... | [{"name": "deposit_insurance_coverage", "arguments": {"bank": "台灣銀行", "ntd_deposit": 2430000, "fx_deposit_twd_equiv": 1920000}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,010 | How much does a 0.0081 TER cost me on 2600000 across 14 years? | [{"name": "expense_ratio_drag", "description": "Quantify how much a fund's total expense ratio erodes long-run returns by comparing gross and net compounded outcomes over a holding period.", "parameters": {"principal": {"description": "Initial investment in TWD.", "type": "int", "default": "1000000"}, "expense_ratio": ... | 請比較有無內扣費用的差異:本金 2600000、費用率 0.0081、年限 14、毛報酬 0.06。 | 好的,我來拆解這個需求。使用者想量化基金內扣費用對長期報酬的侵蝕,掃過 3 個工具後,`emergency_fund_target`、`jcic_credit_report` 明顯屬於其他業務範疇;`expense_ratio_drag` 才是負責評估內扣費用影響的那一個,就用它。
然後把使用者給的條件填進參數:`principal`(投入本金)帶 2600000。`expense_ratio`(總開銷(內扣)費用率)帶 0.0081。`years` 設為 14,也就是題目說的持有年數。扣費用前的年報酬率對應到 `gross_annual_return`,值是 0.06。
另外要記住的規則:內扣費用直接反映在淨值,不會另外扣款,... | [{"name": "expense_ratio_drag", "arguments": {"principal": 2600000, "expense_ratio": 0.0081, "years": 14, "gross_annual_return": 0.06}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,011 | What does 'TISA 台灣個人投資儲蓄帳戶' mean? Target audience: professional. | [{"name": "dividend_tax_compare", "description": "Compare the two Taiwan dividend taxation options - consolidated filing with an 8.5% dividend tax credit capped at NT$80,000 per household, versus separate taxation at a flat 28% - and report which one costs less.", "parameters": {"dividend_amount": {"description": "Tota... | 什麼是TISA 台灣個人投資儲蓄帳戶?我是從業人員,麻煩用繁體中文說明。 | 使用者的問題是解釋一個金融名詞。工具清單裡有 3 個函式,其中 `dividend_tax_compare`、`estate_tax_calc` 分別處理稅務試算,跟這題無關;只有 `financial_term_explain` 的描述對應解釋金融名詞,所以選它。
然後把使用者給的條件填進參數:要解釋的金融名詞對應到 `term`,值是 "TISA 台灣個人投資儲蓄帳戶"。`audience_level`(受眾程度)帶 "professional"。`language`(輸出語言)帶 "zh-TW"。
另外要記住的規則:解釋台灣金融名詞時應附上本地市場實例,避免直接套用美股或中國市場慣例。
確認 JSON 格式正確、引號與逗... | [{"name": "financial_term_explain", "arguments": {"term": "TISA 台灣個人投資儲蓄帳戶", "audience_level": "professional", "language": "zh-TW"}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,012 | Pull 期貨交易理論 exam items (medium) from the 期貨商業務員 question bank, 5 items. | [{"name": "license_exam_question_bank", "description": "Retrieve past examination questions from Taiwan financial licensing exams by licence type, subject topic, quantity and difficulty.", "parameters": {"license": {"description": "Name of the Taiwan financial licence exam.", "type": "str", "default": "證券商業務員"}, "topic... | 調 期貨商業務員 題庫:科目 期貨交易理論、題數 5、難度 medium。 | 好的,我來拆解這個需求。使用者想取得金融證照的歷屆試題,共有 3 個候選函式。`retirement_gap_analysis`、`financial_term_explain` 的功能與本題情境不符,正確的是 `license_exam_question_bank`,因為它就是用來調閱證照題庫的。
然後把使用者給的條件填進參數:`license` 就是證照名稱,這裡填 "期貨商業務員"。`topic` 就是科目主題,這裡填 "期貨交易理論"。題數對應到 `count`,值是 5。`difficulty`(難度)帶 "medium"。型別上 `count` 是 int,格式別弄錯。
另外要記住的規則:台灣金融證照題庫來源包含台灣... | [{"name": "license_exam_question_bank", "arguments": {"license": "期貨商業務員", "topic": "期貨交易理論", "count": 5, "difficulty": "medium"}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,013 | Compute interest on a 4560000 TWD time deposit at 1.717% for 12 months, simple. Also, time deposit math: principal 3980000, rate 1.293, months 6, compounding monthly_compound. Additionally, time deposit math: principal 3540000, rate 1.375, months 12, compounding simple. | [{"name": "time_deposit_interest", "description": "Compute the interest earned on a Taiwan bank time deposit for a given principal, annual rate, tenor in months, fixed or floating rate and simple or monthly-compounded interest.", "parameters": {"principal": {"description": "Deposit principal in TWD.", "type": "int", "d... | 我在台灣銀行存 456 萬定存,年利率 1.717%、12 個月、機動利率、單利存本取息,到期利息多少?另外,請計算定期存款利息,金額 3980000 元,年息 1.293%,存 6 個月。同時,請計算定期存款利息,金額 3540000 元,年息 1.375%,存 12 個月。 | 好的,我來拆解這個需求。使用者想算定期存款的到期利息,掃過 2 個工具後,`mortgage_amortization` 明顯屬於其他業務範疇;`time_deposit_interest` 才是負責計算定存利息的那一個,就用它。因為要對照 3 種情況,我會產生 3 個 tool_call,而不是把值塞在同一次呼叫裡。
參數的部分這樣對:`principal` 設為 4560000,也就是題目說的存款本金。`annual_rate`(年利率(%))帶 1.717。存期月數對應到 `months`,值是 12。`interest_type` 就是固定或機動利率,這裡填 "floating"。`compounding` 就是計息方式,... | [{"name": "time_deposit_interest", "arguments": {"principal": 4560000, "annual_rate": 1.717, "months": 12, "interest_type": "floating", "compounding": "simple"}}, {"name": "time_deposit_interest", "arguments": {"principal": 3980000, "annual_rate": 1.293, "months": 6, "compounding": "monthly_compound"}}, {"name": "time_... | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,015 | Retirement planning: age 31, retire 65, expense 50000, savings 5500000. Additionally, retrieve the spot_buy board rate of CNY at 華南銀行. | [{"name": "retirement_gap_analysis", "description": "Run a Taiwan retirement funding gap analysis: inflate today's living expenses to the retirement date, compute the required nest egg to age of life expectancy, compare with projected savings and report the shortfall and the monthly saving needed to close it.", "parame... | 幫我做退休缺口分析,31 歲、預計 65 歲退休、活到 85 歲、通膨 0.02。順便也幫我看:華南銀行的人民幣即期買入是多少?換算成 TWD。 | 先看清楚使用者要什麼。這題是要分析退休金缺口,可用函式只有 `retirement_gap_analysis` 一個,先確認它的描述確實涵蓋這個情境——是的,可以使用。問題其實包了兩個不同的需求;第二段要查銀行牌告匯率,對應的是 `fx_rate_lookup`。
把題目資訊映射到函式參數:`current_age`(目前年齡)帶 31。`retirement_age`(預計退休年齡)帶 65。`monthly_expense_today`(今日幣值的每月支出)帶 50000。`current_savings` 設為 5500000,也就是題目說的現有退休準備金。`inflation_rate` 就是假設通膨率,這裡填 0.02。... | [{"name": "retirement_gap_analysis", "arguments": {"current_age": 31, "retirement_age": 65, "monthly_expense_today": 50000, "current_savings": 5500000, "inflation_rate": 0.02, "life_expectancy": 85}}, {"name": "fx_rate_lookup", "arguments": {"base_currency": "CNY", "quote_currency": "TWD", "bank": "華南銀行", "rate_type": ... | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,017 | Look up the spot_buy rate for KRW against TWD at 彰化銀行. | [{"name": "gift_tax_calc", "description": "Calculate Taiwan gift tax for a donor in a given year using the NT$2.44 million annual exemption, prior gifts already made in the same year, and the 10%/15%/20% brackets.", "parameters": {"gift_amount": {"description": "Value of the current gift in TWD.", "type": "int", "defau... | 請查彰化銀行今天韓元(KRW)兌台幣的即期買入匯率。 | 收到,使用者需要的是查銀行牌告匯率。共有 4 個候選函式。`gift_tax_calc`、`time_deposit_interest`、`price_limit_range` 的功能與本題情境不符,正確的是 `fx_rate_lookup`,因為它就是用來查詢外匯牌告匯率的。
然後把使用者給的條件填進參數:外幣幣別對應到 `base_currency`,值是 "KRW"。計價幣別對應到 `quote_currency`,值是 "TWD"。`bank` 就是查詢的銀行,這裡填 "彰化銀行"。牌告匯率種類對應到 `rate_type`,值是 "spot_buy"。
檢查一遍沒有遺漏必要參數,就照 schema 回傳。 | [{"name": "fx_rate_lookup", "arguments": {"base_currency": "KRW", "quote_currency": "TWD", "bank": "彰化銀行", "rate_type": "spot_buy"}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,018 | What is the all-in cost of buying 2000 shares of 2382 at NT$974.8 (discount 0.28, day_trade False)? | [{"name": "stock_trade_cost", "description": "Calculate the total trading cost of a Taiwan stock order, including brokerage commission (0.1425% with a NT$20 minimum) and securities transaction tax (0.3%, halved to 0.15% for intraday day trading).", "parameters": {"price": {"description": "Execution price per share in T... | 成交價 974.8 元、2000 股、買進、手續費 0.28 折數、非當沖,交易成本是多少? | 讓我理一下。需求是試算一筆台股買賣的手續費與證交稅,掃過 2 個工具後,`bond_price_from_ytm` 明顯屬於其他業務範疇;`stock_trade_cost` 才是負責計算台股交易成本的那一個,就用它。
然後把使用者給的條件填進參數:每股成交價對應到 `price`,值是 974.8。`shares`(股數)帶 2000。`side` 就是買賣方向,這裡填 "buy"。手續費折扣倍數對應到 `discount`,值是 0.28。`day_trade`(是否為現股當沖)帶 false。`discount` 要用 float 型別傳入。
格式上以 tool_call 包住 JSON 物件,不要額外加解釋文字。 | [{"name": "stock_trade_cost", "arguments": {"price": 974.8, "shares": 2000, "side": "buy", "discount": 0.28, "day_trade": false}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,019 | Estimate inheritance tax for an estate worth 56360000 TWD. And also, how much estate tax is payable on 90540000 TWD with 3 children and spouse True? | [{"name": "etf_dividend_yield", "description": "Calculate the annualized cash dividend yield of a Taiwan-listed ETF from its market price, annual distribution per unit and distribution frequency.", "parameters": {"symbol": {"description": "ETF ticker, e.g. '0056'.", "type": "str", "default": "0056"}, "price": {"descrip... | 幫我算遺產稅,遺產總額 56360000,配偶 True,卑親屬 2 人。同時,請試算遺產稅:財產 90540000 元、配偶 True、子女 3 位、父母 1 位、喪葬費 1380000。 | 讓我理一下。需求是試算遺產稅,掃過 3 個工具後,`etf_dividend_yield`、`dividend_tax_compare` 明顯屬於其他業務範疇;`estate_tax_calc` 才是負責計算遺產稅的那一個,就用它。因為要對照 2 種情況,我會產生 2 個 tool_call,而不是把值塞在同一次呼叫裡。
參數的部分這樣對:遺產總額對應到 `gross_estate`,值是 56360000。`spouse` 設為 true,也就是題目說的是否有配偶。`lineal_descendants`(直系血親卑親屬人數)帶 2。
第二組條件則是:`gross_estate`(遺產總額)帶 90540000。是否有配偶對應... | [{"name": "estate_tax_calc", "arguments": {"gross_estate": 56360000, "spouse": true, "lineal_descendants": 2}}, {"name": "estate_tax_calc", "arguments": {"gross_estate": 90540000, "spouse": true, "lineal_descendants": 3, "parents": 1, "funeral_expense": 1380000}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,020 | Amortize a 9120000 loan at 2.458% for 20 years using equal_total. Also, calculate the securities transaction tax on selling 8210000 TWD of corporate_bond_etf, day_trade=True. | [{"name": "mortgage_amortization", "description": "Build a Taiwan mortgage amortization result: monthly payment during and after the interest-only grace period, total interest and total repayment, for either equal-total-payment or equal-principal repayment.", "parameters": {"principal": {"description": "Loan principal ... | 房貸試算:本金 9120000 元、年利率 2.458%、期間 20 年、寬限期 0 年、方式 equal_total。順便也幫我看:我賣出 821 萬元的債券型ETF,屬於當沖,證交稅要繳多少? | 好的,我來拆解這個需求。使用者想試算房貸月付金與總利息,可用函式只有 `mortgage_amortization` 一個,先確認它的描述確實涵蓋這個情境——是的,可以使用。使用者在後半段還問了另一件事——算賣出時要繳的證券交易稅,那要另外呼叫 `securities_transaction_tax`。
然後把使用者給的條件填進參數:`principal` 就是貸款本金,這裡填 9120000。年利率(%)對應到 `annual_rate`,值是 2.458。`years` 設為 20,也就是題目說的貸款年限。`grace_period_years`(寬限期年數)帶 0。`method` 設為 "equal_total",也就是題... | [{"name": "mortgage_amortization", "arguments": {"principal": 9120000, "annual_rate": 2.458, "years": 20, "grace_period_years": 0, "method": "equal_total"}}, {"name": "securities_transaction_tax", "arguments": {"amount": 8210000, "security_type": "corporate_bond_etf", "day_trade": true}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,021 | Compute the 114 individual income tax bill: income 1600000, status single, deduction itemized. | [{"name": "income_tax_estimate", "description": "Estimate Taiwan individual consolidated income tax, applying the exemption, standard or itemized deduction, salary special deduction and the progressive bracket table for the given tax year.", "parameters": {"annual_income": {"description": "Total annual income in TWD.",... | 幫我算綜所稅:所得總額 1600000、申報身分 single、扶養人數 1、扣除方式 itemized、年度 114。 | 這個請求的重點在於試算個人綜合所得稅,共有 4 個候選函式。`labor_pension_fund_projection`、`license_exam_question_bank`、`gift_tax_calc` 的功能與本題情境不符,正確的是 `income_tax_estimate`,因為它就是用來估算綜所稅的。
參數的部分這樣對:`annual_income` 就是全年所得總額,這裡填 1600000。`filing_status`(申報身分)帶 "single"。扶養親屬人數對應到 `dependents`,值是 1。`deduction_type`(扣除額方式)帶 "itemized"。`tax_year`(民國年度)帶... | [{"name": "income_tax_estimate", "arguments": {"annual_income": 1600000, "filing_status": "single", "dependents": 1, "deduction_type": "itemized", "tax_year": 114}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,022 | Compute required retirement corpus and monthly savings for this profile. | [{"name": "etf_dividend_yield", "description": "Calculate the annualized cash dividend yield of a Taiwan-listed ETF from its market price, annual distribution per unit and distribution frequency.", "parameters": {"symbol": {"description": "ETF ticker, e.g. '0056'.", "type": "str", "default": "0056"}, "price": {"descrip... | 我 39 歲、62 歲退休(還有 23 年),每月要花 3 萬(今日幣值),手上 1680000 元,報酬抓 0.04、通膨 0.015、活到 85 歲,請算缺口與每月該存多少。 | 讓我理一下。需求是分析退休金缺口,先排除不相干的選項:`etf_dividend_yield`、`margin_trading_calc` 是台股交易、基金與 ETF用的,都不符合。剩下 `retirement_gap_analysis` 才是處理做退休缺口分析的函式。
把題目資訊映射到函式參數:`current_age` 設為 39,也就是題目說的目前年齡。`retirement_age` 設為 62,也就是題目說的預計退休年齡。`monthly_expense_today` 就是今日幣值的每月支出,這裡填 30000。`current_savings` 設為 1680000,也就是題目說的現有退休準備金。`annual_ret... | [{"name": "retirement_gap_analysis", "arguments": {"current_age": 39, "retirement_age": 62, "monthly_expense_today": 30000, "current_savings": 1680000, "annual_return": 0.04, "inflation_rate": 0.015, "life_expectancy": 85}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,023 | Is the 2.11% supplementary premium withheld on a 30000 TWD part_time_salary payment? | [{"name": "nhi_supplementary_premium", "description": "Calculate the Taiwan second-generation National Health Insurance supplementary premium at 2.11%, applying the NT$20,000 single-payment threshold (or the four-times-monthly-insured-salary threshold for bonuses) and the NT$10,000,000 single-payment cap.", "parameters... | 補充保費試算:所得類別 part_time_salary、金額 30000 元、投保薪資 40100。 | 收到,使用者需要的是算二代健保補充保費要扣多少。先排除不相干的選項:`etf_dividend_yield`、`labor_insurance_annuity` 是勞健保與退休金、基金與 ETF用的,都不符合。剩下 `nhi_supplementary_premium` 才是處理計算補充保費的函式。
參數的部分這樣對:所得類別對應到 `income_type`,值是 "part_time_salary"。`amount` 設為 30000,也就是題目說的單次給付金額。`monthly_insured_salary`(月投保金額)帶 40100。
另外要記住的規則:高額獎金以全年累計超過當月投保金額 4 倍的部分計收。
格式上以... | [{"name": "nhi_supplementary_premium", "arguments": {"income_type": "part_time_salary", "amount": 30000, "monthly_insured_salary": 40100}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,024 | Credit card interest math: balance 230000, apr 12.14, days 15, min payment 0.1. In addition, compute revolving credit interest on a 130000 TWD balance at 12.04% APR over 15 days. And also, how much interest accrues on 60000 of credit card revolving debt at 8.12% for 90 days? | [{"name": "credit_card_revolving_interest", "description": "Compute credit card revolving credit interest in Taiwan for a given outstanding balance, annual percentage rate and number of days, and flag any rate that exceeds the 15% statutory ceiling under Banking Act Article 47-1.", "parameters": {"balance": {"descripti... | 請算信用卡循環信用利息,未清償餘額 230000、年利率 12.14%、期間 15 天。順便也幫我看:我國泰世華信用卡欠 130000 元動用循環信用,年利率 12.04%,15 天的利息是多少?合法嗎?同時,循環利息試算:餘額 60000 元、循環利率 8.12%、90 天、最低應繳比率 0.1。 | 先看清楚使用者要什麼。這題是要算信用卡循環信用利息,共有 2 個候選函式。`income_tax_estimate` 的功能與本題情境不符,正確的是 `credit_card_revolving_interest`,因為它就是用來計算循環信用利息的。這題有 3 組條件要比較,必須分成 3 個獨立的函式呼叫。
把題目資訊映射到函式參數:`balance` 設為 230000,也就是題目說的循環信用未清償餘額。`apr`(循環信用年利率(%))帶 12.14。`days`(計息天數)帶 15。最低應繳比率對應到 `min_payment_rate`,值是 0.1。
第二組條件則是:循環信用未清償餘額對應到 `balance`,值是 ... | [{"name": "credit_card_revolving_interest", "arguments": {"balance": 230000, "apr": 12.14, "days": 15, "min_payment_rate": 0.1}}, {"name": "credit_card_revolving_interest", "arguments": {"balance": 130000, "apr": 12.04, "days": 15}}, {"name": "credit_card_revolving_interest", "arguments": {"balance": 60000, "apr": 8.12... | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,027 | What does 信託業法 Article 24 say? Include amendment history: True. | [{"name": "taiwan_regulation_lookup", "description": "Retrieve the full text and latest amendment date of a specific article of a Taiwan financial statute from the national law database.", "parameters": {"law_name": {"description": "Name of the statute in Traditional Chinese, e.g. '銀行法'.", "type": "str", "default": "銀行... | 信託業法 第 24 條在講什麼?麻煩調全國法規資料庫。 | 先看清楚使用者要什麼。這題是要調閱某條金融法規的條文,目前只有一個可用函式 `taiwan_regulation_lookup`,功能描述正好對應這個需求,所以直接用它。
參數的部分這樣對:法規名稱對應到 `law_name`,值是 "信託業法"。`article`(條號)帶 24。`include_amendments`(是否包含修正沿革)帶 true。
另外要記住的規則:應以全國法規資料庫最新版本為準,並註明最近修正日期。
最後只要輸出純 JSON 的函式呼叫,不加任何多餘說明或 Markdown。 | [{"name": "taiwan_regulation_lookup", "arguments": {"law_name": "信託業法", "article": 24, "include_amendments": true}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,028 | Look up TAIEX_TR data for 2026-01-15, fields ['open', 'high', 'low', 'close', 'volume']. | [{"name": "financial_term_explain", "description": "Explain a Taiwan financial or investment term at a requested audience level, with a local market example.", "parameters": {"term": {"description": "The financial term to explain.", "type": "str", "default": "本益比"}, "audience_level": {"description": "'beginner', 'inter... | 請查 發行量加權股價報酬指數(TAIEX_TR)在 2026-01-15 的open、high、low、close、volume。 | 好的,我來拆解這個需求。使用者想查台灣股市指數的當日數據,工具清單裡有 2 個函式,其中 `financial_term_explain` 分別處理金融知識查詢,跟這題無關;只有 `market_index_lookup` 的描述對應查詢市場指數,所以選它。
接著逐一對應參數:`index_name`(指數代碼)帶 "TAIEX_TR"。`date` 設為 "2026-01-15",也就是題目說的查詢日期。`fields` 就是要回傳的欄位,這裡填 ["open", "high", "low", "close", "volume"]。注意 `fields` 的型別是 list,不要寫成字串。
另外要記住的規則:加權指數不含股利,... | [{"name": "market_index_lookup", "arguments": {"index_name": "TAIEX_TR", "date": "2026-01-15", "fields": ["open", "high", "low", "close", "volume"]}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,029 | Compute the Taiwan labor insurance old-age annuity for average insured salary 45800, 19 years of seniority, claim age offset -5. | [{"name": "bond_price_from_ytm", "description": "Price a fixed-coupon bond from its yield to maturity, reporting clean price, premium or discount status and total coupon cash flow.", "parameters": {"face_value": {"description": "Face value in TWD.", "type": "int", "default": "1000000"}, "coupon_rate": {"description": "... | 我平均月投保薪資 45800 元、年資 19 年,提前 5 年請領,勞保老年年金每月能領多少? | 先看清楚使用者要什麼。這題是要試算勞保老年年金月領金額,掃過 3 個工具後,`bond_price_from_ytm`、`license_exam_question_bank` 明顯屬於其他業務範疇;`labor_insurance_annuity` 才是負責計算勞保年金給付的那一個,就用它。
接下來確認每個參數要帶什麼:平均月投保薪資對應到 `avg_insured_salary`,值是 45800。投保年資對應到 `service_years`,值是 19。`claim_age_offset` 就是相對法定請領年齡的年數差,這裡填 -5。
另外要記住的規則:展延每延後 1 年增給 4%(上限 20%),減給每提前 1 年減 ... | [{"name": "labor_insurance_annuity", "arguments": {"avg_insured_salary": 45800, "service_years": 19, "claim_age_offset": -5}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,030 | Which dividend tax option is cheaper: 730000 dividends, 2810000 other income, married_joint? | [{"name": "dividend_tax_compare", "description": "Compare the two Taiwan dividend taxation options - consolidated filing with an 8.5% dividend tax credit capped at NT$80,000 per household, versus separate taxation at a flat 28% - and report which one costs less.", "parameters": {"dividend_amount": {"description": "Tota... | 股利 730000 元、其他所得 2810000 元、夫妻合併申報,請比較兩種股利課稅方式哪個划算。 | 好的,我來拆解這個需求。使用者想比較股利合併計稅與分開計稅哪個省,共有 3 個候選函式。`expense_ratio_drag`、`emergency_fund_target` 的功能與本題情境不符,正確的是 `dividend_tax_compare`,因為它就是用來比較兩種股利課稅方式的。
參數的部分這樣對:`dividend_amount`(全年股利所得)帶 730000。股利以外的所得對應到 `other_income`,值是 2810000。申報身分對應到 `filing_status`,值是 "married_joint"。
另外要記住的規則:邊際稅率 20% 以下通常合併計稅較省,30% 以上多半分開計稅有利。
... | [{"name": "dividend_tax_compare", "arguments": {"dividend_amount": 730000, "other_income": 2810000, "filing_status": "married_joint"}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,031 | Compute debt burden ratio: income 210000, debt 280000, new loan 2940000. Additionally, check unsecured borrowing capacity for this applicant profile. | [{"name": "jcic_credit_report", "description": "Request a Joint Credit Information Center credit report for a stated purpose, choosing the report scope and the look-back window for recent credit inquiries.", "parameters": {"query_purpose": {"description": "Purpose of the inquiry, e.g. 'mortgage_application'.", "type": ... | DBR 22 倍試算:收入 210000、負債 280000、新增貸款 2940000。同時,月入 16 萬要申請 2930000 元信貸,目前無擔保負債 2710000,過得了嗎? | 讓我理一下。需求是評估無擔保借款能力是否超過 DBR 上限,共有 2 個候選函式。`jcic_credit_report` 的功能與本題情境不符,正確的是 `credit_capacity_dbr`,因為它就是用來評估授信負債比的。因為要對照 2 種情況,我會產生 2 個 tool_call,而不是把值塞在同一次呼叫裡。
把題目資訊映射到函式參數:平均月收入對應到 `monthly_income`,值是 210000。`unsecured_debt_balance` 就是無擔保負債餘額,這裡填 280000。`new_loan_amount`(擬申請的新貸款金額)帶 2940000。
第二組條件則是:平均月收入對應到 `mont... | [{"name": "credit_capacity_dbr", "arguments": {"monthly_income": 210000, "unsecured_debt_balance": 280000, "new_loan_amount": 2940000}}, {"name": "credit_capacity_dbr", "arguments": {"monthly_income": 160000, "unsecured_debt_balance": 2710000, "new_loan_amount": 2930000}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,032 | Retrieve the spot_buy board rate of USD at 星展銀行(台灣). Additionally, determine the reserve gap given liquid assets of 80000. | [{"name": "emergency_fund_target", "description": "Determine the recommended emergency fund size from monthly expenses, desired months of coverage and job stability, and compare it with current liquid assets.", "parameters": {"monthly_expense": {"description": "Monthly household expense in TWD.", "type": "int", "defaul... | 星展銀行(台灣)的美元即期買入是多少?換算成 TWD。另外,工作穩定的話緊急預備金要抓幾個月?我月花 100000,現金 80000。 | 先看清楚使用者要什麼。這題是要查銀行牌告匯率,先排除不相干的選項:`estate_tax_calc`、`foreign_currency_deposit_return` 是外匯、稅務試算用的,都不符合。剩下 `fx_rate_lookup` 才是處理查詢外匯牌告匯率的函式。問題其實包了兩個不同的需求;第二段要算緊急預備金該準備多少,對應的是 `emergency_fund_target`。
然後把使用者給的條件填進參數:外幣幣別對應到 `base_currency`,值是 "USD"。`quote_currency`(計價幣別)帶 "TWD"。`bank` 就是查詢的銀行,這裡填 "星展銀行(台灣)"。牌告匯率種類對應到 `rat... | [{"name": "fx_rate_lookup", "arguments": {"base_currency": "USD", "quote_currency": "TWD", "bank": "星展銀行(台灣)", "rate_type": "spot_buy"}}, {"name": "emergency_fund_target", "arguments": {"monthly_expense": 100000, "job_stability": "stable", "current_liquid_assets": 80000}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
100,033 | Compute future value of monthly contributions of 15000 over 12 years. | [{"name": "retirement_gap_analysis", "description": "Run a Taiwan retirement funding gap analysis: inflate today's living expenses to the retirement date, compute the required nest egg to age of life expectancy, compare with projected savings and report the shortfall and the monthly saving needed to close it.", "parame... | 每個月扣 15000 元投資 00679B,投 12 年、年報酬 7.0%,總值和總投入各多少? | 讓我理一下。需求是推估定期定額的終值,共有 2 個候選函式。`retirement_gap_analysis` 的功能與本題情境不符,正確的是 `regular_investment_projection`,因為它就是用來推算定期定額累積結果的。
參數的部分這樣對:`monthly_amount`(每月扣款金額)帶 15000。假設年化報酬率對應到 `annual_return`,值是 0.07。投資年數對應到 `years`,值是 12。累積標的代號對應到 `symbol`,值是 "00679B"。型別上 `annual_return` 是 float,格式別弄錯。
另外要記住的規則:定期定額終值採期末年金複利公式;台灣券商多... | [{"name": "regular_investment_projection", "arguments": {"monthly_amount": 15000, "annual_return": 0.07, "years": 12, "symbol": "00679B"}}] | [
{
"content": "You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.\n\n<tools>\n[{\"type\": \"function\", \"function\": {\... |
tw-finance-function-call-reasoning
台灣金融場景的繁體中文 function-calling + 推理鏈微調資料集。
欄位規格對齊 twinkle-ai/tw-function-call-reasoning-10k。
⚠️ 使用限制:僅供研究,不得商業使用
本資料集以 CC BY-NC 4.0 授權釋出,僅供學術研究、模型能力探索與方法驗證之用。
請務必理解以下事項後再使用:
不得作商業用途。 包含但不限於:訓練用於對外營利的模型、包裝為付費產品或服務、 作為商業交付物的一部分。若有商業需求,請自行重新建置資料並取得合規來源。
這不是財務、稅務、法律或投資建議。 資料中的稅率、費率、法規門檻雖依 2026 年 (民國 115 年)台灣公開資訊整理,但可能已經過時或有誤。任何實際決策前, 請以主管機關公告為準(財政部、金管會、勞動部、衛福部、中央銀行、全國法規資料庫)。
內容為程式化合成,非真實考題逐字收錄。 題目由模板與參數取樣組合而成, 並非任何證照考試的原始試題。工具(function)的 schema 為虛構的 API 介面設計, 不對應任何真實可呼叫的服務。
不得用於自動化的金融決策或客戶服務,除非已由具備資格的專業人員完整審閱與驗證。
資料中的人名、金額、帳戶情境均為虛構,不含任何真實個人資料。
一句話總結:這是一份研究用的方法論展示資料集,不是可以直接上線的產品資產。
資料集概覽
| 項目 | 內容 |
|---|---|
| 語言 | 繁體中文(query_zhtw / think)+ 英文(query) |
| 領域 | 11 類台灣金融場景 |
| 工具數 | 35 個 function schema |
| 樣本數 | 1,200(train 1,080 / test 120,比例 9:1) |
| 額外 | challenge config:40 題驗收挑戰集(欄位不同,見下) |
| 授權 | CC BY-NC 4.0(僅研究、非商業) |
領域分佈
securities 台股交易|tax 稅務|pension 勞健保與退休金|banking 銀行存放款|
funds 基金與 ETF|fx 外匯|bond 債券與利率|compliance 法令遵循|
credit 信用與授信|planning 財務規劃|knowledge 金融知識查詢
欄位規格
與參考資料集完全一致的 7 個欄位:
| 欄位 | 型別 | 說明 |
|---|---|---|
id |
int64 |
樣本編號 |
query |
string |
英文指令 |
tools |
string |
可用函式清單(JSON 字串,含干擾項) |
query_zhtw |
string |
繁體中文指令(實際訓練用的 user turn) |
think |
string |
繁體中文推理鏈 |
answer |
string |
期望的函式呼叫(JSON 字串) |
messages |
list[{role, content}] |
長度 3 的 system / user / assistant,可直接 SFT |
messages[2] 的格式為 <think>...</think> 後接一至多個 <tool_call>...</tool_call>。
範例
{
"id": 100004,
"query_zhtw": "同一家華南銀行放 6620000 元台幣加 3750000 元外幣,中央存保賠付上限是多少?同時,請算信用卡循環信用利息,未清償餘額 140000、年利率 14.82%、期間 90 天。",
"answer": "[{\"name\": \"deposit_insurance_coverage\", \"arguments\": {\"bank\": \"華南銀行\", \"ntd_deposit\": 6620000, \"fx_deposit_twd_equiv\": 3750000}}, {\"name\": \"credit_card_revolving_interest\", \"arguments\": {\"balance\": 140000, \"apr\": 14.82, \"days\": 90}}]"
}
品質指標
由 scripts/validate_dataset.py 產出:
| 指標 | 數值 |
|---|---|
| train / test | 1,080 / 120(90.0% : 10.0%) |
| 函式呼叫總數 | 1,493(平均每題 1.24 次) |
| 平均提供工具數 | 2.51(含干擾項) |
| 參數溯源率 | **99.98%**(5,716 / 5,717) |
| train/test 完全重複 | 0 筆 |
| 結構性錯誤 | 0 |
「參數溯源率」指 answer 中每個參數值都能在題目文字裡找到來源。
題目沒提到的參數一律不會出現在 answer(該參數有 default 可用),
以符合 system prompt「不得臆測參數值」的要求。
(唯一未通過的 1 筆是驗證器對整數型 enum 的邊界案例,非資料缺陷。)
challenge config:驗收挑戰集
40 題,欄位與 train/test 不同:id / phenomenon / query_zhtw / tools /
expected_behavior / scoring / rubric。
涵蓋 8 類 IID 測試集在結構上量不到的現象,每類 5 題: 口語與非標準輸入、參數不足需追問、無適用工具需拒答、錯誤前提、 多輪上下文、時效性衝突、單位與量詞陷阱、近似工具混淆。
其中 8 題的正確行為是「不要呼叫任何工具」(scoring: must_not_call),
32 題需要人工或 LLM-judge 依 rubric 評分。這正是重點:
驗收評測無法用 exact match 完成,必須有領域方定義的評分準則。
已知限制
- 合成資料的同構性:
think由模板組合生成,句式多樣性有限。 以此微調的模型可能學到「推理鏈的外觀」而非推理能力本身。 - IID 切分的樂觀偏誤:test 是從同一個生成器隨機切出的。實測 test 的 query 模板 100% 在 train 出現過,與訓練集最相似樣本的字元 5-gram Jaccard 平均達 0.42。 在這個 test set 上的高分不代表真實場景的表現。
- 工具為虛構 schema:不對應任何可呼叫的真實 API。
- 法規時效:以 2026-09 的公開資訊為準,之後的修法未反映。
重現方式
python3 src/build_dataset.py --n 1200 --seed 20260909 # 產生 train/test
python3 scripts/build_challenge_set.py # 產生挑戰集
python3 scripts/validate_dataset.py # 品質驗證
python3 scripts/leakage_report.py # 樂觀偏誤分析
資料來源
制度參數整理自台灣公開資訊,逐項出處見 docs/SOURCES.md。
欄位規格參考 twinkle-ai/tw-function-call-reasoning-10k(CC-BY-4.0)——
僅參考其 schema 設計,未使用其任何資料內容。
引用
@misc{tw_finance_function_call_reasoning_2026,
title = {tw-finance-function-call-reasoning: A Research-Only Traditional Chinese
Function-Calling Dataset for Taiwan Financial Scenarios},
author = {Simon Liu},
year = {2026},
note = {CC BY-NC 4.0. Research use only, not for commercial use.},
url = {https://huggingface.co/datasets/Simon-Liu/tw-finance-function-call-reasoning}
}
- Downloads last month
- 43