Vespa DB近邻搜索报错：query(q_embedding)应为张量而非字符串

问题场景 #

使用Vespa DB对embeddings执行nearestNeighbor近邻搜索时，按官方文档构造查询请求，尝试从products文档集检索结果，返回类型不匹配错误，且Vespa容器未生成错误日志：

Expected "query(q_embedding)" to be a tensor, but it is the string

原始请求、响应及Schema #

原始请求 #

{
    "yql": "SELECT * FROM products WHERE ([{\"targetHits\":10, \"approximate\": true}]nearestNeighbor(text_embedding, q_embedding))",
    "ranking": {
        "profile": "default",
        "listFeatures": true,
        "features": {
            "query(q_embedding)": [
                0.06836861371994019,
                0,
                0.005350692197680473,
                // 省略剩余数值
                0
            ]
        }
    }
}

错误响应 #

{
    "root": {
        "id": "toplevel",
        "relevance": 1.0,
        "fields": {
            "totalCount": 0
        },
        "errors": [
            {
                "code": 4,
                "summary": "Invalid query parameter",
                "source": "products",
                "message": "Expected 'query(q_embedding)' to be a tensor, but it is the string '[\n                0.06836861371994019,\n                0,\n                ...（省略部分数值）\n            ]'"
            }
        ]
    }
}

Schema定义 #

schema products {
    document products {
        field id type int {
        }
        field gender type string {
            indexing: summary | attribute
        }
        field masterCategory type string {
            indexing: summary | attribute
        }
        field subCategory type string {
            indexing: summary | attribute
        }
        field articleType type string {
            indexing: summary | attribute
        }
        field baseColour type string {
            indexing: summary | attribute
        }
        field season type string {
            indexing: summary | attribute
        }
        field year type int {
            indexing: summary | attribute
        }
        field usage type string {
            indexing: summary | attribute
        }
        field productDisplayName type string {
            indexing: summary | attribute
        }
        field image type string {
            indexing: summary | attribute
        }
        field image_embedding type tensor<float>(x[128]) {
            indexing: attribute
            attribute {
                distance-metric: euclidean
            }
        }
        field text_embedding type tensor<float>(x[128]) {
            indexing: attribute
            attribute {
                distance-metric: euclidean
            }
        }
    }

    fieldset default {
        fields: id, gender, masterCategory, subCategory, articleType, baseColour, season, year, usage, productDisplayName, image, image_embedding, text_embedding
    }

    rank-profile default {
        first-phase {
            expression: closeness(field, image_embedding) + closeness(field, text_embedding)
        }
    }
}

错误原因 #

Vespa要求查询中的张量特征必须显式声明张量类型，与Schema中定义的tensor<float>(x[128])结构完全匹配。原始请求直接传递数值数组，被Vespa解析为字符串类型，导致类型不匹配报错。

修复方案 #

将query(q_embedding)的数值数组包装为张量对象，明确指定type为tensor<float>(x[128])，values字段保留原始数值数组。

修改后的请求 #

{
    "yql": "SELECT * FROM products WHERE ([{\"targetHits\":10, \"approximate\": true}]nearestNeighbor(text_embedding, q_embedding))",
    "ranking": {
        "profile": "default",
        "listFeatures": true,
        "features": {
            "query(q_embedding)": {
                "type": "tensor<float>(x[128])",
                "values": [
                    0.06836861371994019,
                    0,
                    0.005350692197680473,
                    // 省略剩余数值
                    0
                ]
            }
        }
    }
}

验证 #

修改请求后重新发送，Vespa会正确识别q_embedding为张量类型，执行近邻搜索并返回匹配的文档列表，错误提示消失。

内容的提问来源于stack exchange，提问作者Rogue