在prompt內文可以要求AI寫程式並執行
分享一個組prompt同時呼叫Code Interpreter 讓LLM幫你寫程式並執行(另外要收費), 可叫LLM幫你產檔或畫圖,
上行範例如下:
{
"model": "gpt-5-mini",
"input": [
{
"role": "user",
"content": "以下是一組現金流（日期 + 金額），請使用 python tool 計算 XIRR，並產生一張現金流分佈的直方圖, x軸為月份,y軸為金額, 要小心中文字不要變亂碼，最後回傳計算結果並附圖：\n\n2026/1/1 -5000\n2026/2/1 -8000\n2026/3/1 -4000\n2026/4/1 -3000\n2026/5/1 -4000\n2026/6/1 -9000\n2026/7/1 -12345\n2026/8/1 -6000\n2026/9/1 -70000\n2026/10/1 -8562\n2026/11/1 -5555\n2026/12/1 140000"
}
],
"tools": [
{
"type": "code_interpreter",
"container": {
"type": "auto"
}
}
],
"tool_choice": "auto"
}
回傳結果:
依以下json結果下載的圖片:
{
...略...
"output": [
{
"id": "rs_0d2794e1ebce33f3006a3e3cc611548196a361905aa3016302",
"type": "reasoning",
"summary": []
},
{
"id": "ci_0d2794e1ebce33f3006a3e3ccf82c48196ab45f2a3ba058490",
"type": "code_interpreter_call",
"status": "completed",
"code": "# 計算 XIRR 並繪製現金流每月分佈圖 (中文標籤處理)\r\nimport numpy as np\r\nimport matplotlib.pyplot as plt\r\nfrom matplotlib import font_manager as fm\r\nfrom datetime import datetime\r\nimport os\r\n\r\n# 現金流資料 (日期, 金額)\r\ndata = [\r\n (\"2026/1/1\", -5000),\r\n (\"2026/2/1\", -8000),\r\n (\"2026/3/1\", -4000),\r\n (\"2026/4/1\", -3000),\r\n (\"2026/5/1\", -4000),\r\n (\"2026/6/1\", -9000),\r\n (\"2026/7/1\", -12345),\r\n (\"2026/8/1\", -6000),\r\n (\"2026/9/1\", -70000),\r\n (\"2026/10/1\", -8562),\r\n (\"2026/11/1\", -5555),\r\n (\"2026/12/1\", 140000),\r\n]\r\n\r\n# 解析日期與金額\r\ndates = [datetime.strptime(d, \"%Y/%m/%d\") for d, _ in data]\r\namounts = np.array([a for _, a in data], dtype=float)\r\n\r\n# XIRR 函數實作 (使用 Newton-Raphson 與 bisection 作為 fallback)\r\ndef xirr(dates, amounts, guess=0.1, tol=1e-8, maxiter=200):\r\n # 以第一個日期為基準計算天數差 (年為單位)\r\n days = np.array([(d - dates[0]).days for d in dates], dtype=float)\r\n def f(r):\r\n return np.sum(amounts / ((1.0 + r) ** (days / 365.0)))\r\n def df(r):\r\n return np.sum(- (days / 365.0) * amounts / ((1.0 + r) ** (days / 365.0 + 1.0)))\r\n \r\n # 初始 guess，避免 r <= -1\r\n r = guess\r\n if r <= -0.999999:\r\n r = 0.1\r\n # Newton-Raphson\r\n for i in range(maxiter):\r\n try:\r\n fr = f(r)\r\n dfr = df(r)\r\n except Exception:\r\n break\r\n if abs(fr) < tol:\r\n return r\r\n # 若導數太小或會造成數值問題，跳出使用二分法\r\n if dfr == 0 or not np.isfinite(dfr):\r\n break\r\n r_new = r - fr / dfr\r\n if not np.isfinite(r_new):\r\n break\r\n # 若新 r 趨近於 -1，避免無效值\r\n if r_new <= -0.999999:\r\n break\r\n # 收斂\r\n if abs(r_new - r) < tol:\r\n return r_new\r\n r = r_new\r\n \r\n # 二分法尋找根 (先找 bracket)\r\n lo, hi = -0.999999, 10.0\r\n flo, fhi = f(lo + 1e-12), f(hi) # avoid exactly -1\r\n # 嘗試擴展 hi 直到與 lo 異號或達上限\r\n attempts = 0\r\n while flo * fhi > 0 and attempts < 100:\r\n hi *= 2\r\n fhi = f(hi)\r\n attempts += 1\r\n if flo * fhi > 0:\r\n # 無法找到符號改變，回傳 nan\r\n return float('nan')\r\n # bisection\r\n for i in range(200):\r\n mid = (lo + hi) / 2.0\r\n fmid = f(mid)\r\n if abs(fmid) < tol:\r\n return mid\r\n if flo * fmid < 0:\r\n hi = mid\r\n fhi = fmid\r\n else:\r\n lo = mid\r\n flo = fmid\r\n return (lo + hi) / 2.0\r\n\r\nresult = xirr(dates, amounts, guess=0.1)\r\n# 轉換為年化百分率\r\nif np.isfinite(result):\r\n xirr_pct = result * 100\r\nelse:\r\n xirr_pct = None\r\n\r\n# 嘗試找到系統中可用的中文字型以避免亂碼\r\ndef find_chinese_font():\r\n # 常見的中文字體關鍵字\r\n keywords = [\"Noto\", \"NotoSansCJK\", \"Source Han\", \"SimHei\", \"SimSun\", \"Microsoft YaHei\", \"Arial Unicode\", \"PingFang\", \"Heiti\", \"WenQuanYi\", \"DejaVu Sans\"]\r\n found = []\r\n for f in fm.findSystemFonts(fontpaths=None, fontext='ttf'):\r\n try:\r\n name = fm.FontProperties(fname=f).get_name()\r\n for kw in keywords:\r\n if kw.lower() in name.lower() or kw.lower() in f.lower():\r\n found.append(f)\r\n break\r\n except Exception:\r\n continue\r\n return found[0] if found else None\r\n\r\nchinese_font = find_chinese_font()\r\nif chinese_font:\r\n plt.rcParams['font.family'] = fm.FontProperties(fname=chinese_font).get_name()\r\nelse:\r\n # 若找不到，仍設定為 sans-serif，matplotlib 會嘗試 fallback\r\n plt.rcParams['font.family'] = 'sans-serif'\r\n\r\n# 繪製每月金額直方圖 (使用 bar chart 以月份為 x 軸)\r\nmonths = [d.strftime(\"%m月\") for d in dates]\r\nfig, ax = plt.subplots(figsize=(10,5))\r\nbars = ax.bar(months, amounts, color=['tab:blue' if a<0 else 'tab:orange' for a in amounts])\r\nax.set_xlabel(\"月份\", fontsize=12)\r\nax.set_ylabel(\"金額\", fontsize=12)\r\nax.set_title(\"每月現金流分佈\", fontsize=14)\r\nax.axhline(0, color='black', linewidth=0.5)\r\n# 在 bar 上顯示數值\r\nfor bar, amt in zip(bars, amounts):\r\n height = bar.get_height()\r\n va = 'bottom' if height >= 0 else 'top'\r\n ax.text(bar.get_x() + bar.get_width()/2, height, f\"{amt:,.0f}\", ha='center', va=va, fontsize=9)\r\n\r\nplt.tight_layout()\r\n\r\n# 儲存圖檔\r\noutput_path = \"/mnt/data/cashflow_hist.png\"\r\nfig.savefig(output_path, dpi=150)\r\n\r\n# 列印結果與圖檔路徑\r\nprint(f\"計算完成：XIRR = {xirr_pct:.6f}%（年化）\" if xirr_pct is not None else \"XIRR 無法計算\")\r\nprint(f\"圖檔已儲存：{output_path}\")\r\n\r\n# 顯示圖形\r\nplt.show()",
"container_id": "cntr_6a3e3cc58dcc8190b4465d8b4759ef0600d7077fd1d958bb",
"outputs": null
},
{
"id": "msg_0d2794e1ebce33f3006a3e3cdf6f8081968dd16200bed18b5c",
"type": "message",
"status": "completed",
"content": [
{
"type": "output_text",
"annotations": [
{
"type": "container_file_citation",
"container_id": "cntr_6a3e3cc58dcc8190b4465d8b4759ef0600d7077fd1d958bb",
"end_index": 0,
"file_id": "cfile_6a3e3cdedd348190ad4de82b9250aec3",
"filename": "cfile_6a3e3cdedd348190ad4de82b9250aec3.png",
"start_index": 0
},
{
"type": "container_file_citation",
"container_id": "cntr_6a3e3cc58dcc8190b4465d8b4759ef0600d7077fd1d958bb",
"end_index": 129,
"file_id": "cfile_6a3e3ce0d85881908b251c6263d30862",
"filename": "cashflow_hist.png",
"start_index": 94
}
],
"logprobs": [],
"text": "我已計算並繪製完畢。\n\n計算結果：\n- XIRR = 9.358533%（年化）\n\n已產生的圖檔（每月現金流分佈，x 軸：月份，y 軸：金額）：\n[Download the image](sandbox:/mnt/data/cashflow_hist.png)\n\n如果你要我改成其他格式（例如 PDF 或 SVG）、改變字型、或改為累積現金流圖/折線圖，請告訴我。"
}
],
"role": "assistant"
}
],
...略...
}annotations內會列出產生的檔案, 通常會產出多餘的圖, cfile開頭的圖可以不理它, 只要下載其餘的圖即可,
下載方式是查看annotations內資訊, 用 get method呼叫: https://你的服務名稱.azure.com/openai/v1/containers/{container id}/files/{檔案 id}/content
內容裡也有程式碼, 如果要複用, 也能把程式碼拿來在自己的電腦跑, 比較省錢
Taiwan is a country. 臺灣是我的國家