| 1 | # -*- coding: utf-8 -*- |
| 2 | """ |
| 3 | Qwen3.5-VL 多模态大模型 API 客户端（DashScope 多模态接口专用） |
| 4 | 只支持 Qwen3.5-VL 及兼容 DashScope 多模态对话接口 |
| 5 | 参考官方文档：https://help.aliyun.com/zh/model-studio/qwen-api-reference |
| 6 | """ |
| 7 | |
| 8 | import os |
| 9 | |
| 10 | try: |
| 11 | import dashscope |
| 12 | from dashscope import MultiModalConversation |
| 13 | except ImportError: |
| 14 | dashscope = None |
| 15 | MultiModalConversation = None |
| 16 | import logging |
| 17 | |
| 18 | logger = logging.getLogger(__name__) |
| 19 | from typing import Any, Dict, List, Optional |
| 20 | |
| 21 | class QwenVLClient: |
| 22 | def __init__(self, |
| 23 | api_key: Optional[str] = None, |
| 24 | base_url: Optional[str] = None): |
| 25 | """ |
| 26 | Qwen3.5-VL 多模态客户端 |
| 27 | :param api_key: DashScope/Qwen3.5 API Key |
| 28 | :param model: 模型名（如 qwen3.5-plus/qwen3.5-max 等） |
| 29 | """ |
| 30 | self.api_key = api_key or os.getenv("DASHSCOPE_API_KEY") |
| 31 | |
| 32 | def chat( |
| 33 | self, |
| 34 | text: str, |
| 35 | images: List[str], |
| 36 | model: str, |
| 37 | stream: bool = False, |
| 38 | parameters: Optional[Dict] = None, |
| 39 | videos: Optional[List[str]] = None, |
| 40 | **kwargs |
| 41 | ) -> Any: |
| 42 | """ |
| 43 | 使用阿里云 dashscope SDK 进行多模态对话（文本+图片/视频），风格与 image_dashscope.py 一致。 |
| 44 | :param text: 文本内容 |
| 45 | :param images: 图片路径列表（支持本地路径或URL，内部会转换为file://绝对路径） |
| 46 | :param videos: 视频路径列表（支持本地路径或URL，内部会转换为file://绝对路径） |
| 47 | :param model: 模型名（支持qwen3.5-plus, qwen3-vl-plus） |
| 48 | :param stream: 是否流式输出（暂不支持流式） |
| 49 | :param parameters: 其他API参数 |
| 50 | :return: API响应内容 dict |
| 51 | """ |
| 52 | if dashscope is None or MultiModalConversation is None: |
| 53 | raise RuntimeError("dashscope package not installed. Run: pip install dashscope") |
| 54 | |
| 55 | dashscope.api_key = self.api_key |
| 56 | # 只支持非流式 |
| 57 | try: |
| 58 | content = [ |
| 59 | {"text": text}, |
| 60 | *({"image": p} for p in images), |
| 61 | *({"video": p} for p in videos or []), |
| 62 | ] |
| 63 | messages = [{"role": "user", "content": content}] |
| 64 | response = MultiModalConversation.call( |
| 65 | model=model, |
| 66 | messages=messages, |
| 67 | api_key=self.api_key, |
| 68 | enable_thinking=False, |
| 69 | **(parameters or {}) |
| 70 | ) |
| 71 | if hasattr(response, 'status_code') and response.status_code == 200: |
| 72 | # qwen3.5-plus 的返回格式为 { choices: [ { message: { content: [...] } } ] } |
| 73 | resp = response.output.choices[0].message.content[0] |
| 74 | if resp.get('text'): |
| 75 | return resp['text'] |
| 76 | return resp |
| 77 | else: |
| 78 | raise RuntimeError(f"DashScope QwenVLClient failed: {getattr(response, 'message', response)}") |
| 79 | except Exception as e: |
| 80 | raise RuntimeError(f"DashScope QwenVLClient error: {e}") |
| 81 | |
| 82 | |
| 83 | if __name__ == "__main__": |
| 84 | import sys |
| 85 | import time |
| 86 | import json |
| 87 | sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) |
| 88 | from config import Config |
| 89 | |
| 90 | # 支持的 VLM 模型列表 |
| 91 | MODELS = ["qwen3.6-plus", "qwen3.6-flash", "kimi-k2.6"] |
| 92 | |
| 93 | print("=== Qwen VL (DashScope) 多模态可用性测试 ===") |
| 94 | api_key = getattr(Config, "DASHSCOPE_API_KEY", None) or os.getenv("DASHSCOPE_API_KEY", "") |
| 95 | if not api_key: |
| 96 | print("✗ DASHSCOPE_API_KEY 未设置，跳过") |
| 97 | sys.exit(1) |
| 98 | print(f" API Key: {api_key[:6]}***{api_key[-4:]}") |
| 99 | client = QwenVLClient(api_key=api_key) |
| 100 | |
| 101 | # 测试图片 |
| 102 | img_path = '' |
| 103 | abs_img_path = os.path.abspath(img_path) |
| 104 | if not os.path.exists(img_path): |
| 105 | img_path = "code/result/image/test_avail/test_input.png" |
| 106 | abs_img_path = os.path.abspath(img_path) |
| 107 | if not os.path.exists(img_path): |
| 108 | print("✗ 测试图片不存在，跳过") |
| 109 | sys.exit(0) |
| 110 | |
| 111 | text = "请描述这张图片的内容" |
| 112 | print(f"\n[多模态] Prompt: {text}") |
| 113 | print(f" 图片: {img_path}") |
| 114 | |
| 115 | for model in MODELS: |
| 116 | print(f"\n--- 测试模型: {model} ---") |
| 117 | t0 = time.time() |
| 118 | try: |
| 119 | result = client.chat(text=text, images=[img_path], model=model, stream=False) |
| 120 | elapsed = time.time() - t0 |
| 121 | if result: |
| 122 | print(f"✓ 返回结果 ({elapsed:.1f}s): {str(result)[:200]}") |
| 123 | else: |
| 124 | print(f"✗ 返回空结果 ({elapsed:.1f}s)") |
| 125 | except Exception as e: |
| 126 | print(f"✗ 失败: {e}") |
| 127 |