使用 Nova 產生結構化輸出

結構化輸出對於machine-to-machine的通訊使用案例至關重要，因為這可讓下游使用案例更有效地使用和處理產生的輸出。無論是從文件擷取資訊、建立從 APIs，還是開發採取動作的客服人員，這些任務都需要基礎模型來產生特定結構化格式的輸出。

Nova Models 利用限制解碼來確保所產生輸出中的高模型可靠性，並允許模型輕鬆處理複雜的結構描述。限制解碼倚賴文法「限制」模型在每個步驟中可以輸出的可能字符。這與過去使用的提示技術不同，因為這會變更模型在產生輸出時可以選擇的實際字符。例如，關閉 JSON 物件時，模型將限制為僅選取 } 個字符。每次傳遞工具組態時，都會利用限制解碼。由於工具使用已提供我們特定的結構描述，因此我們能夠使用它，根據開發人員所需的結構描述動態產生文法。限制解碼可防止模型產生無效的金鑰，並根據定義的結構描述強制執行正確的資料類型。

若要利用工具搭配結構化輸出使用，主要步驟是定義輸出所需的 JSON 結構描述。以下是tool_config定義中定義的 JSON 結構描述範例：

  tool_config = {
  "tools": [
    {
      "toolSpec": {
        "name": "ProductAnalysis",
        "description": "Analyze product information from text.",
        "inputSchema": {
          "json": {
            "type": "object",
            "properties": {
              "name": {
                "type": "string",
                "description": "Product name"
              },
              "rating": {
                "maximum": 5,
                "description": "Customer rating 1-5",
                "type": [
                  "number",
                  "null"
                ],
                "minimum": 1
              },
              "features": {
                "description": "Key product features",
                "type": "array",
                "items": {
                  "type": "string"
                }
              },
              "category": {
                "type": "string",
                "description": "Product category"
              },
              "price": {
                "type": "number",
                "description": "Price in USD"
              }
            },
            "required": [
              "name",
              "category",
              "price",
              "features"
            ]
          }
        }
      }
    }
  ],
  "toolChoice": {
    "tool": {
      "name": "ProductAnalysis"
    }
  }
}
 

當您稍後使用模型呼叫工具時，您將會收到以結構描述格式回應的輸出。例如，以下是使用 Python 呼叫模型的範例：

  import boto3

client = boto3.client("bedrock-runtime")
model_id = "amazon.nova-lite-1-5-v1:0"

user_query = """The Amazon Kindle Scribe is a state-of-the-art e-reader designed for both reading and writing, featuring a 10.2-inch paper-like display and a premium stylus. This versatile device allows users to enjoy books, take notes, annotate PDFs, and even sketch, making it ideal for readers, students, and professionals. Priced at $339.99, it falls under the electronics category and boasts features like a front light, adjustable warm light settings, and up to 12 weeks of battery life on a single charge. Customer ratings for the Kindle Scribe average around 4.5 stars, reflecting its high user satisfaction."""

messages = [{
   "role": "user",
   "content": [{
        "text": user_query
   }]
}]

system = [{"text": "Leverage the ProductAnalysis tool to extract product information"}]
inference_params = {"temperature": 0}

response = client.converse(modelId=model_id, system=system, messages=messages, toolConfig=tool_config, inferenceConfig=inference_params)
print(next(
    block["toolUse"]
    for block in response["output"]["message"]["content"]
    if "toolUse" in block
))
 

然後，輸出會顯示如下：

 {
  "toolUseId": "tooluse_hke1FUeuRbKXK8DPqIptVg",
  "name": "ProductAnalysis",
  "input": {
    "name": "Amazon Kindle Scribe",
    "rating": 4.5,
    "features": [
      "10.2-inch paper-like display",
      "premium stylus",
      "front light",
      "adjustable warm light settings",
      "up to 12 weeks of battery life"
    ],
    "category": "electronics",
    "price": 339.99
  }
}