{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# TypeSafe AI Jev: 桃太郎を使った利用例\n", "typesafe-sdk==0.6.0 / python-dotenv==1.2.3。作業ディレクトリの.envにTYPESAFE_API_KEYを設定してください。\n", "API出力は未保存です。全セルの実行で4回のAPIリクエストを送信します。\n" ], "id": "jev-example-00" }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 環境変数の読み込み" ], "id": "jev-example-01" }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from dotenv import load_dotenv\n", "\n", "load_dotenv(\".env\")" ], "id": "jev-example-02" }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 共通のstate" ], "id": "jev-example-03" }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "momotaro_story = (\n", " \"昔々、あるところにおじいさんとおばあさんが住んでいました。おばあさんが川で洗濯をしていると、\"\n", " \"大きな桃がどんぶらこと流れてきました。家に持ち帰り桃を割ると、中から元気な男の子が生まれ、\"\n", " \"二人は桃太郎と名付けて育てました。桃太郎はすくすくと成長し、ある日、鬼ヶ島の鬼たちが村人から\"\n", " \"金銀財宝を奪い、乱暴を働いていることを知ります。桃太郎は鬼退治を決意し、おばあさんが作ってくれた\"\n", " \"きび団子を持って旅に出ました。道中、犬・猿・雉に出会い、きび団子を分け与えることで仲間にします。\"\n", " \"桃太郎たちは力を合わせて鬼ヶ島に上陸し、鬼の大将と激しく戦った末に鬼たちを降伏させ、奪われていた\"\n", " \"財宝を取り戻しました。桃太郎たちは財宝を持って村に帰り、おじいさんとおばあさんや村人たちと共に\"\n", " \"平和な暮らしを取り戻しました。\"\n", ")" ], "id": "jev-example-04" }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Noul: 二択判定" ], "id": "jev-example-05" }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from typesafe_sdk import Noul, TypeSafeClient\n", "\n", "with TypeSafeClient() as client:\n", " response = client.system_one(\n", " model=\"jev-latest\",\n", " state=momotaro_story,\n", " questions={\n", " \"succeeded_with_animal_help\": Noul(\n", " instructions=\"桃太郎は犬・猿・雉など動物たちの協力を得て鬼退治に成功しましたか？\",\n", " criteria={\n", " \"true\": \"動物たちが仲間になり、鬼退治の成功に貢献している\",\n", " \"false\": \"動物の協力がない、または鬼退治に失敗している\",\n", " },\n", " ),\n", " },\n", " )\n", "\n", "print(response.nouls[\"succeeded_with_animal_help\"].noul)" ], "id": "jev-example-06" }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Choice: ジャンル分類" ], "id": "jev-example-07" }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from typesafe_sdk import Choice, TypeSafeClient\n", "\n", "with TypeSafeClient() as client:\n", " response = client.system_one(\n", " model=\"jev-latest\",\n", " state=momotaro_story,\n", " questions={\n", " \"genre\": Choice(\n", " instructions=\"この物語のジャンルとして最も適切なものはどれですか？\",\n", " criteria={\n", " \"昔話・おとぎ話\": \"口承で伝わる伝統的な民話、不思議な出来事や教訓を含む\",\n", " \"恋愛小説\": \"登場人物の恋愛関係が主題の物語\",\n", " \"推理小説\": \"謎解きや犯人捜しが主題の物語\",\n", " \"伝記\": \"実在した人物の生涯を記録した書物\",\n", " },\n", " ),\n", " },\n", " )\n", "\n", "result = response.choices[\"genre\"]\n", "print(result.choice)\n", "print(result.probabilities)\n", "print(result.confidence)" ], "id": "jev-example-08" }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Score: 脅威の評価" ], "id": "jev-example-09" }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from typesafe_sdk import Score, TypeSafeClient\n", "\n", "with TypeSafeClient() as client:\n", " response = client.system_one(\n", " model=\"jev-latest\",\n", " state=momotaro_story,\n", " questions={\n", " \"oni_threat_level\": Score(\n", " instructions=\"鬼ヶ島の鬼が村にもたらしていた脅威はどの程度深刻でしたか？\",\n", " criteria=[\n", " {\n", " # スコア値 0\n", " \"what\": \"村人に軽いいたずらをする程度\",\n", " \"examples\": [\"作物を少し荒らす\", \"夜中に物音を立てて驚かせる\"],\n", " },\n", " {\n", " # スコア値 1\n", " \"what\": \"村人の生活や財産に大きな被害を与えている\",\n", " \"examples\": [\"金銀財宝を略奪する\", \"家屋を破壊する\"],\n", " },\n", " {\n", " # スコア値 2\n", " \"what\": \"村人の命や安全を脅かすほど深刻な脅威となっている\",\n", " \"examples\": [\"村人に暴力を振るう\", \"村を占拠し支配する\"],\n", " },\n", " ],\n", " ),\n", " },\n", " )\n", "\n", "result = response.scores[\"oni_threat_level\"]\n", "print(result.score)\n", "print(result.legend)\n", "print(result.probabilities)\n", "print(result.confidence)" ], "id": "jev-example-10" }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 複合例: 3種類をまとめて評価" ], "id": "jev-example-11" }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from typesafe_sdk import Choice, Noul, Score, TypeSafeClient\n", "\n", "with TypeSafeClient() as client:\n", " response = client.system_one(\n", " model=\"jev-latest\",\n", " state=momotaro_story,\n", " questions={\n", " \"is_happy_ending\": Noul(\n", " instructions=\"この物語はハッピーエンドで終わりますか？\"\n", " ),\n", " \"protagonist_trait\": Choice(\n", " instructions=\"主人公の桃太郎の性格として最も近いものはどれですか？\",\n", " criteria={\"勇敢\": None, \"臆病\": None, \"狡猾\": None, \"怠惰\": None},\n", " ),\n", " \"moral_lesson_strength\": Score(\n", " instructions=\"この物語の教訓性（子供向けの教育的価値）はどの程度強いですか？\",\n", " criteria=[\n", " {\n", " # スコア値 0\n", " \"what\": \"教訓性はほとんどない\",\n", " \"examples\": [\"単なる出来事の羅列で教訓的な結論がない\"],\n", " },\n", " {\n", " # スコア値 1\n", " \"what\": \"ある程度の教訓が含まれている\",\n", " \"examples\": [\"努力や協力の大切さが部分的に読み取れる\"],\n", " },\n", " {\n", " # スコア値 2\n", " \"what\": \"勧善懲悪など明確な教訓が中心テーマとなっている\",\n", " \"examples\": [\"悪が裁かれ、善が報われる結末が明示されている\"],\n", " },\n", " ],\n", " ),\n", " },\n", " )\n", "\n", "score_result = response.scores[\"moral_lesson_strength\"]\n", "print(response.nouls[\"is_happy_ending\"].noul)\n", "print(response.choices[\"protagonist_trait\"].choice)\n", "print(score_result.score)\n", "print(score_result.legend)\n", "print(response.model)" ], "id": "jev-example-12" } ], "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "name": "python", "version": "3.13" } }, "nbformat": 4, "nbformat_minor": 5 }