Azure添加向量嵌入到索引时出现技能集错误求助

问题根源 #

报错提示Cannot iterate over non-array '/document/pages'和输入类型不匹配，核心原因是AzureOpenAIEmbeddingSkill的上下文配置错误：

• SplitSkill已将文档拆分为数组类型的/document/pages，但EmbeddingSkill的context设为/document（文档根级别），此时使用/document/pages/*会尝试在根级别迭代数组，不符合技能对单个字符串输入的要求。

修复方案 #

调整EmbeddingSkill的上下文到数组项级别，让技能逐个处理拆分后的每个page：

1. 将EmbeddingSkill的context改为/document/pages/*，表示遍历/document/pages数组中的每一项
2. 输入source简化为/（当前上下文即为单个page的内容），或保持/document/pages/*（配合新上下文也可正常工作）

修改后的技能集配置 #

{
  "@odata.context": "https://redacted/$metadata#skillsets/$entity",
  "@odata.etag": "\"something\"",
  "name": "something-skillset",
  "description": "",
  "skills": [
    {
      "@odata.type": "#Microsoft.Skills.Text.SplitSkill",
      "name": "Text split skill",
      "description": "Splits text into pages small enough to vectorize",
      "context": "/document",
      "defaultLanguageCode": "en",
      "textSplitMode": "pages",
      "maximumPageLength": 2000,
      "pageOverlapLength": 500,
      "maximumPagesToTake": 0,
      "inputs": [
        {
          "name": "text",
          "source": "/document/content"
        }
      ],
      "outputs": [
        {
          "name": "textItems",
          "targetName": "/document/pages"
        }
      ]
    },
    {
      "@odata.type": "#Microsoft.Skills.Text.AzureOpenAIEmbeddingSkill",
      "name": "Create vector embedding for pages",
      "description": "",
      "context": "/document/pages/*",
      "resourceUri": "https://something.openai.azure.com",
      "apiKey": "<redacted>",
      "deploymentId": "text-embedding-ada-002",
      "inputs": [
        {
          "name": "text",
          "source": "/"
        }
      ],
      "outputs": [
        {
          "name": "embedding",
          "targetName": "contentVector"
        }
      ],
      "authIdentity": null
    }
  ],
  "cognitiveServices": {
    "@odata.type": "#Microsoft.Azure.Search.DefaultCognitiveServices",
    "description": null
  },
  "knowledgeStore": null,
  "indexProjections": null,
  "encryptionKey": null
}

额外说明 #

• 调整后，EmbeddingSkill会自动遍历/document/pages中的每个字符串元素，为每个page生成对应的向量嵌入
• 若要将向量映射到索引字段，需确保索引中配置了类型为Collection(Edm.Single)的向量字段，并在索引器的字段映射中关联/document/pages/*/contentVector

内容的提问来源于stack exchange，提问作者Toodleey