After thinking about the closure of the Qwen model, the schema verification failed.

[heavenzhen999]

What happened

使用了 qwen3.7-flash 模型，并关闭思考模式。

The exact command

Paste the full command, redacting only real API keys.

npx @liustack/modlens analyze --input ~/Desktop/1.jpg

The full error

我再尝试解析一个示例图片时报错了，应该是关闭思考出现的问题。

Error: OpenAI-compatible API returned JSON that does not match the vision schema (missing: visual.notes). Retry, or switch to gemini-api / anthropic for enforced schemas. Got: {
"summary": "The image is a vertical, dark-themed flowchart titled 'Cyberpunk 2077 Main Quest Flow' (赛博朋克2077主线任务流程), outlining the narrative paths of the video game. It begins with three character origin choices (Rogue, Corpo, Street Kid) that merge into initial quests, then split into three prima...

Harness and environment

• Harness (Claude Code / Codex / Pi / OpenCode / plain terminal): Deepseek Harness
• Provider (-p):
• modlens --version: 3.16.4
• node --version: 22.20.0
• OS: MacOS Tahoe 26.5.2

Expected

这个已经找到一个解决方案，会在下边评论中，贴上我的 config.json 内容。若采用这个 config.json，则希望在文档中增加 Qwen 关闭思考模式的配置说明，或者开发者可以采用其他的解决方案。

[heavenzhen999]

在 extraBody 中，增加了 response_format ，强制声明返回的数据结构。当然我这边仅测试了 qwen3.7-flash ，费用比较优惠。

{
  "providers": {
    "openai": {
      "baseUrl": "https://{workspace}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1",
      "apiKey": "token",
      "model": "qwen3.7-flash",
      "extraBody": {
        "enable_thinking": false,
        "preserve_thinking": false,
        "response_format": {
          "type": "json_schema",
          "json_schema": {
            "name": "vision_result",
            "strict": true,
            "schema": {
              "$schema": "http://json-schema.org/draft-07/schema#",
              "type": "object",
              "properties": {
                "summary": {
                  "type": "string"
                },
                "ocr": {
                  "type": "object",
                  "properties": {
                    "full_text": {
                      "type": "string"
                    },
                    "lines": {
                      "type": "array",
                      "items": {
                        "type": "object",
                        "properties": {
                          "text": {
                            "type": "string"
                          },
                          "language": {
                            "anyOf": [
                              {
                                "type": "string"
                              },
                              {
                                "type": "null"
                              }
                            ]
                          }
                        },
                        "required": [
                          "text",
                          "language"
                        ],
                        "additionalProperties": false
                      }
                    }
                  },
                  "required": [
                    "full_text",
                    "lines"
                  ],
                  "additionalProperties": false
                },
                "layout": {
                  "type": "object",
                  "properties": {
                    "regions": {
                      "type": "array",
                      "items": {
                        "type": "object",
                        "properties": {
                          "type": {
                            "type": "string",
                            "description": "A short kind for this region. Prefer a common one where it fits: title, heading, paragraph, list, table, chart, form, code, image, icon, link, nav, button, search. Any other short label is fine when none of those describe it."
                          },
                          "reading_order": {
                            "type": "number"
                          },
                          "text": {
                            "type": "string"
                          }
                        },
                        "required": [
                          "type",
                          "reading_order",
                          "text"
                        ],
                        "additionalProperties": false
                      }
                    }
                  },
                  "required": [
                    "regions"
                  ],
                  "additionalProperties": false
                },
                "semantics": {
                  "type": "object",
                  "properties": {
                    "scene": {
                      "type": "string"
                    },
                    "intent": {
                      "anyOf": [
                        {
                          "type": "string"
                        },
                        {
                          "type": "null"
                        }
                      ]
                    },
                    "entities": {
                      "type": "array",
                      "items": {
                        "type": "object",
                        "properties": {
                          "name": {
                            "type": "string"
                          },
                          "type": {
                            "type": "string"
                          },
                          "evidence": {
                            "anyOf": [
                              {
                                "type": "string"
                              },
                              {
                                "type": "null"
                              }
                            ]
                          }
                        },
                        "required": [
                          "name",
                          "type",
                          "evidence"
                        ],
                        "additionalProperties": false
                      }
                    },
                    "relations": {
                      "anyOf": [
                        {
                          "type": "array",
                          "items": {
                            "type": "object",
                            "properties": {
                              "subject": {
                                "type": "string"
                              },
                              "predicate": {
                                "type": "string"
                              },
                              "object": {
                                "type": "string"
                              }
                            },
                            "required": [
                              "subject",
                              "predicate",
                              "object"
                            ],
                            "additionalProperties": false
                          }
                        },
                        {
                          "type": "null"
                        }
                      ]
                    }
                  },
                  "required": [
                    "scene",
                    "intent",
                    "entities",
                    "relations"
                  ],
                  "additionalProperties": false
                },
                "visual": {
                  "type": "object",
                  "properties": {
                    "dominant_colors": {
                      "anyOf": [
                        {
                          "type": "array",
                          "items": {
                            "type": "string"
                          }
                        },
                        {
                          "type": "null"
                        }
                      ]
                    },
                    "style": {
                      "anyOf": [
                        {
                          "type": "string"
                        },
                        {
                          "type": "null"
                        }
                      ]
                    },
                    "notes": {
                      "anyOf": [
                        {
                          "type": "array",
                          "items": {
                            "type": "string"
                          }
                        },
                        {
                          "type": "null"
                        }
                      ]
                    }
                  },
                  "additionalProperties": false,
                  "required": [
                    "dominant_colors",
                    "style",
                    "notes"
                  ]
                },
                "uncertainty": {
                  "type": "array",
                  "items": {
                    "type": "string"
                  }
                }
              },
              "required": [
                "summary",
                "ocr",
                "layout",
                "semantics",
                "visual",
                "uncertainty"
              ],
              "additionalProperties": false
            }
          }
        }
      }
    }
  },
  "provider": "openai"
}

[liustack]

两条都做了，3.16.6 已发布。你不用再手写那份 schema 了。

先说诊断，和你想的不完全一样

visual 在契约里没有必填列表，所以 visual.notes 缺失本来就是被接受的。它能被报出来，说明模型返回了这个字段但形状不对。我按这个思路验了一遍：缺失 → 通过，null → 拒绝，字符串 → 拒绝。

null 几乎肯定就是实际形状，而且你自己的绕行 schema 正是把每个可选字段都写成 anyOf [type, null]——你不得不合法化的，就是它。关掉思考的模型在没话可说的位置写 null 是很自然的行为。

所以这是我们的 bug：没什么可说时写 null，和干脆不写，是同一个意思，我们却只接受后者，为一个空列表废掉整次成功识别。

现在 null 会在校验前被丢弃，键直接删掉。评审提醒了一个我一开始没想到的点：光「接受 null」不够，那等于把 null 放行给下游，而文档承诺这些字段要么不存在要么是声明的类型。所以是归一化而不是放行，你拿到的结果里不会再有 null。契约里全部七个可选位置都有测试盯着。

顺带修了报错措辞：原来一律说 missing，而它其实也包括「存在但形状不对」，读到「missing: visual.notes」再去找一个明明在那儿的字段，方向就错了。现在是 wrong or missing。

你要的那条也做了

不用再手写两百行：

modlens config set openai.structuredOutput true

契约会以 response_format: json_schema 的严格形式发出去（全字段必填、additionalProperties: false、可选的改成 nullable），也就是你手写的那个形状。区别是它由 modlens 校验用的同一份 schema 推导而来，没有第二份副本需要你跟着同步——契约以后变了，发出去的自动跟着变。

默认关闭，因为不支持这个字段的网关会直接 400。你在 extraBody 里设的 response_format 优先级更高，所以你现有的配置照常工作，两个都开时用你的。

配置手册和故障排查（中英四份）都写进去了，也点明了关掉思考会让形状错误更容易出现，这两项常常一起用。

npx -y @liustack/modlens@3.16.6 doctor

谢谢，报告带着可用方案来的，还标明只在 qwen3.7-flash 上验过，省了很多来回。