Gemini 3.1 Flash 高速推論APIの実装テクニック — ストリーミング・Function Calling・バッチ処理の最適化

// Node.js + @google/generative-ai で Gemini 3.1 Flash を使用
import { GoogleGenerativeAI } from "@google/generative-ai";
 
const client = new GoogleGenerativeAI(process.env.GEMINI_API_KEY);
const model = client.getGenerativeModel({
  model: "gemini-3.1-flash",
  generationConfig: {
    temperature: 0.7,
    topP: 0.95,
    topK: 40,
    maxOutputTokens: 2048,
  }
});
 
// トークン数の確認（コスト予測）
const countResponse = await model.countTokens(
  "Gemini 3.1 Flash についての長い質問文..."
);
console.log(`Input tokens: ${countResponse.totalTokens}`);

// Node.js + Fetch API でストリーミングを実装
async function streamGemini3Flash(prompt) {
  const response = await fetch(
    "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-flash:streamGenerateContent",
    {
      method: "POST",
      headers: {
        "Content-Type": "application/json",
        "x-goog-api-key": process.env.GEMINI_API_KEY,
      },
      body: JSON.stringify({
        contents: [
          {
            role: "user",
            parts: [{ text: prompt }],
          },
        ],
        generationConfig: {
          temperature: 0.7,
          maxOutputTokens: 2048,
        },
      }),
    }
  );
 
  if (!response.ok) {
    throw new Error(`API Error: ${response.statusText}`);
  }
 
  // ストリーミングレスポンスの処理
  const reader = response.body.getReader();
  const decoder = new TextDecoder();
  let fullText = "";
 
  try {
    while (true) {
      const { done, value } = await reader.read();
      if (done) break;
 
      const chunk = decoder.decode(value);
 
      // 行単位で JSON パースする（改行区切り）
      const lines = chunk.split("\n");
      for (const line of lines) {
        if (!line.trim()) continue;
 
        try {
          const parsed = JSON.parse(line);
          const text =
            parsed.candidates?.[0]?.content?.parts?.[0]?.text || "";
 
          if (text) {
            fullText += text;
            // ここでチャンク単位で処理（フロントエンドへの送信など）
            console.log("Stream chunk:", text);
          }
        } catch (e) {
          // JSON パース失敗は無視（不完全なチャンク対策）
        }
      }
    }
  } finally {
    reader.cancel();
  }
 
  return fullText;
}
 
// 使用例
const result = await streamGemini3Flash(
  "次のデータを JSON にして: 名前: 太郎, 年齢: 30, 職業: エンジニア"
);
console.log("Final result:", result);
// Expected output:
// {"名前": "太郎", "年齢": 30, "職業": "エンジニア"}

import { GoogleGenerativeAI } from "@google/generative-ai";
 
const client = new GoogleGenerativeAI(process.env.GEMINI_API_KEY);
const model = client.getGenerativeModel({ model: "gemini-3.1-flash" });
 
async function streamWithSDK(prompt) {
  // startStream() でストリーミングを開始
  const stream = await model.generateContentStream(prompt);
 
  // text プロパティでテキストチャンクを取得
  for await (const chunk of stream.stream) {
    const chunkText =
      chunk.candidates[0]?.content?.parts[0]?.text || "";
    if (chunkText) {
      process.stdout.write(chunkText); // リアルタイム出力
    }
  }
 
  // 最終レスポンスメタデータ
  const response = await stream.response;
  console.log(
    `\n\nTotal tokens used: ${response.usageMetadata.totalTokenCount}`
  );
}
 
await streamWithSDK("100 個の創造的な起業アイデアをリストアップしてください");

import { GoogleGenerativeAI } from "@google/generative-ai";
 
const client = new GoogleGenerativeAI(process.env.GEMINI_API_KEY);
const model = client.getGenerativeModel({ model: "gemini-3.1-flash" });
 
// Tool 定義
const tools = {
  googleSearch: {
    description: "Google 検索を実行",
    parameters: {
      type: "object",
      properties: {
        query: {
          type: "string",
          description: "検索クエリ"
        },
      },
      required: ["query"],
    },
  },
  fetchUrl: {
    description: "URL のコンテンツを取得",
    parameters: {
      type: "object",
      properties: {
        url: {
          type: "string",
          description: "フェッチ対象のURL"
        },
      },
      required: ["url"],
    },
  },
};
 
// Function Calling の呼び出し
async function callWithTools(userPrompt) {
  const response = await model.generateContent({
    contents: [
      {
        role: "user",
        parts: [{ text: userPrompt }],
      },
    ],
    tools: [
      {
        functionDeclarations: Object.entries(tools).map(([name, spec]) => ({
          name,
          description: spec.description,
          parameters: spec.parameters,
        })),
      },
    ],
  });
 
  // レスポンスをチェック
  const candidates = response.candidates || [];
  for (const candidate of candidates) {
    const functionCalls =
      candidate.content?.parts?.filter(p => p.functionCall) || [];
 
    for (const part of functionCalls) {
      const { name, args } = part.functionCall;
      console.log(`📞 Function call: ${name}`, args);
 
      // ここで実際のツール実行を行う
      // 今回は仮の結果を返す
    }
  }
 
  return response;
}
 
// 使用例
await callWithTools(
  "最新の Google Cloud Platform 価格情報を検索して、短くまとめてください"
);

// Batch API リクエストの例（JSON Lines 形式）
// requests.jsonl
/*
{"custom_id":"req-1","params":{"model":"gemini-3.1-flash","contents":[{"role":"user","parts":[{"text":"Python で FizzBuzz を実装"}]}]}}
{"custom_id":"req-2","params":{"model":"gemini-3.1-flash","contents":[{"role":"user","parts":[{"text":"JavaScript で クイックソート を実装"}]}]}}
{"custom_id":"req-3","params":{"model":"gemini-3.1-flash","contents":[{"role":"user","parts":[{"text":"Go で ファイアウォール を実装"}]}]}}
*/
 
import fs from "fs";
import axios from "axios";
 
async function submitBatchRequest() {
  const batchContent = `{"custom_id":"req-1","params":{"model":"gemini-3.1-flash","contents":[{"role":"user","parts":[{"text":"Python で FizzBuzz を実装"}]}]}}
{"custom_id":"req-2","params":{"model":"gemini-3.1-flash","contents":[{"role":"user","parts":[{"text":"JavaScript で クイックソート を実装"}]}]}}
{"custom_id":"req-3","params":{"model":"gemini-3.1-flash","contents":[{"role":"user","parts":[{"text":"Go で ファイアウォール を実装"}]}]}}`;
 
  const formData = new FormData();
  const blob = new Blob([batchContent], { type: "text/plain" });
  formData.append("file", blob, "requests.jsonl");
 
  const response = await axios.post(
    "https://generativelanguage.googleapis.com/v1beta/batches",
    formData,
    {
      headers: {
        "x-goog-api-key": process.env.GEMINI_API_KEY,
        ...formData.getHeaders(),
      },
    }
  );
 
  console.log("Batch submitted:", response.data);
  return response.data.name; // batchId
}
 
// バッチ実行結果の確認
async function checkBatchStatus(batchId) {
  const response = await axios.get(
    `https://generativelanguage.googleapis.com/v1beta/${batchId}`,
    {
      headers: { "x-goog-api-key": process.env.GEMINI_API_KEY },
    }
  );
 
  console.log("Batch status:", response.data.state);
  // state: "PROCESSING" / "COMPLETED" / "FAILED"
 
  if (response.data.state === "COMPLETED") {
    // 結果ファイルの URL から結果を取得
    const resultsUrl = response.data.outputFile;
    console.log("Results available at:", resultsUrl);
  }
 
  return response.data;
}
 
const batchId = await submitBatchRequest();
await new Promise(resolve => setTimeout(resolve, 10000)); // 10秒待機
await checkBatchStatus(batchId);

async function callWithRetry(prompt, maxRetries = 3) {
  for (let attempt = 1; attempt <= maxRetries; attempt++) {
    try {
      const response = await model.generateContent(prompt);
      return response;
    } catch (error) {
      if (error.status === 429) {
        // Rate limit exceeded
        const waitTime = Math.pow(2, attempt) * 1000; // Exponential backoff
        console.log(`Rate limited. Waiting ${waitTime}ms...`);
        await new Promise(resolve => setTimeout(resolve, waitTime));
        continue;
      }
      throw error;
    }
  }
  throw new Error("Max retries exceeded");
}

function sanitizePrompt(userInput) {
  // SQL インジェクション・プロンプトインジェクション対策
  const forbidden = [
    "system_prompt",
    "__init__",
    "eval(",
    "exec(",
  ];
 
  for (const pattern of forbidden) {
    if (userInput.toLowerCase().includes(pattern.toLowerCase())) {
      throw new Error("Invalid input detected");
    }
  }
 
  return userInput.trim().substring(0, 4000); // Length limit
}