| 1 | # -*- coding: utf-8 -*- |
| 2 | """ |
| 3 | OpenAI GPT 多模态大模型 API 客户端 |
| 4 | 支持 gpt-4o, gpt-5.4 (未来版本) 等视觉模型 |
| 5 | |
| 6 | 代理由 provider 的 enable_proxy 决定，通过 httpx.Client 注入到 OpenAI SDK。 |
| 7 | """ |
| 8 | |
| 9 | import os |
| 10 | import sys |
| 11 | |
| 12 | models_dir = os.path.dirname(os.path.abspath(__file__)) |
| 13 | backend_dir = os.path.dirname(models_dir) |
| 14 | if backend_dir not in sys.path: |
| 15 | sys.path.insert(0, backend_dir) |
| 16 | |
| 17 | import time |
| 18 | import base64 |
| 19 | import httpx |
| 20 | import logging |
| 21 | from openai import OpenAI |
| 22 | from typing import Dict, List, Optional |
| 23 | from config import Config |
| 24 | |
| 25 | logger = logging.getLogger(__name__) |
| 26 | |
| 27 | |
| 28 | class GPTVLClient: |
| 29 | """ |
| 30 | GPT VLM 客户端，支持多模态对话 |
| 31 | """ |
| 32 | def __init__(self, |
| 33 | api_key: Optional[str] = None, |
| 34 | base_url: Optional[str] = None, |
| 35 | proxy: Optional[str] = None, |
| 36 | timeout: float = 60.0): |
| 37 | """ |
| 38 | GPT 多模态客户端 |
| 39 | :param api_key: OpenAI API Key |
| 40 | :param base_url: 自定义 Base URL（如果传入，则不使用本地代理） |
| 41 | :param timeout: 超时时间 |
| 42 | """ |
| 43 | # 优先使用传入的 api_key，否则从 config.yaml 读取 |
| 44 | self.api_key = api_key or Config.OPENAI_API_KEY |
| 45 | self.timeout = timeout |
| 46 | |
| 47 | kwargs = {"api_key": self.api_key, "timeout": self.timeout} |
| 48 | |
| 49 | self.base_url = base_url |
| 50 | if proxy is None: |
| 51 | proxy = Config.provider_proxy("openai") |
| 52 | if proxy: |
| 53 | kwargs["http_client"] = httpx.Client( |
| 54 | proxy=proxy, |
| 55 | timeout=self.timeout, |
| 56 | ) |
| 57 | if self.base_url: |
| 58 | kwargs["base_url"] = self.base_url |
| 59 | self.client = OpenAI(**kwargs) |
| 60 | self.max_attempts = 5 |
| 61 | |
| 62 | def _encode_image(self, image_path: str) -> str: |
| 63 | """将本地图片编码为 base64""" |
| 64 | abs_path = os.path.abspath(image_path) |
| 65 | with open(abs_path, "rb") as f: |
| 66 | return base64.b64encode(f.read()).decode("utf-8") |
| 67 | |
| 68 | def _get_mime_type(self, image_path: str) -> str: |
| 69 | """根据文件扩展名获取 MIME 类型""" |
| 70 | ext = os.path.splitext(image_path)[1].lower() |
| 71 | mime_types = { |
| 72 | ".jpg": "image/jpeg", |
| 73 | ".jpeg": "image/jpeg", |
| 74 | ".png": "image/png", |
| 75 | ".webp": "image/webp", |
| 76 | ".gif": "image/gif" |
| 77 | } |
| 78 | return mime_types.get(ext, "image/jpeg") |
| 79 | |
| 80 | def chat(self, text: str, images: List[str], model: str = "gpt-5.4", |
| 81 | parameters: Optional[Dict] = None) -> str: |
| 82 | """ |
| 83 | 使用 GPT 进行多模态对话（文本+图片） |
| 84 | :param text: 文本内容 |
| 85 | :param images: 图片路径列表（支持本地路径或URL） |
| 86 | :param model: 模型名（如 gpt-4o, gpt-5.4） |
| 87 | :param parameters: 其他API参数 |
| 88 | :return: API响应内容 |
| 89 | """ |
| 90 | if not self.api_key: |
| 91 | raise ValueError("OPENAI_API_KEY 未设置") |
| 92 | |
| 93 | # 构建消息格式 |
| 94 | content: list = [{"type": "text", "text": text}] |
| 95 | |
| 96 | # 处理图片 |
| 97 | if images: |
| 98 | for img_path in images: |
| 99 | if img_path.startswith("data:"): |
| 100 | # Base64 数据 URL |
| 101 | content.append({ |
| 102 | "type": "image_url", |
| 103 | "image_url": {"url": img_path} |
| 104 | }) |
| 105 | elif img_path.startswith("http"): |
| 106 | # URL 图片 |
| 107 | content.append({ |
| 108 | "type": "image_url", |
| 109 | "image_url": {"url": img_path} |
| 110 | }) |
| 111 | else: |
| 112 | # 本地图片 - 转为 base64 |
| 113 | mime_type = self._get_mime_type(img_path) |
| 114 | base64_data = self._encode_image(img_path) |
| 115 | data_url = f"data:{mime_type};base64,{base64_data}" |
| 116 | content.append({ |
| 117 | "type": "image_url", |
| 118 | "image_url": {"url": data_url} |
| 119 | }) |
| 120 | |
| 121 | messages = [{"role": "user", "content": content}] |
| 122 | |
| 123 | attempts = 0 |
| 124 | while attempts < self.max_attempts: |
| 125 | try: |
| 126 | response = self.client.chat.completions.create( |
| 127 | model=model, |
| 128 | messages=messages, |
| 129 | temperature=parameters.get("temperature", 0.7) if parameters else 0.7 |
| 130 | ) |
| 131 | |
| 132 | if response.choices and len(response.choices) > 0: |
| 133 | return response.choices[0].message.content |
| 134 | |
| 135 | except Exception as e: |
| 136 | logger.warning("GPT VLM request failed: %s", e) |
| 137 | attempts += 1 |
| 138 | if attempts < self.max_attempts: |
| 139 | time.sleep(2 * attempts) # 指数退避 |
| 140 | |
| 141 | raise Exception("GPTVL: 达到最大重试次数，仍未获得有效响应。") |
| 142 | |
| 143 | |
| 144 | if __name__ == "__main__": |
| 145 | import sys |
| 146 | sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) |
| 147 | from config import Config |
| 148 | |
| 149 | print("=== GPT VL 多模态可用性测试 ===") |
| 150 | if not Config.OPENAI_API_KEY: |
| 151 | print("✗ 未设置 OPENAI_API_KEY，跳过") |
| 152 | sys.exit(1) |
| 153 | client = GPTVLClient(api_key=Config.OPENAI_API_KEY, base_url=Config.OPENAI_BASE_URL, proxy=Config.provider_proxy("openai")) |
| 154 | |
| 155 | # 支持的 VLM 模型列表，包含用户要求的 gpt-5.4 |
| 156 | MODELS = ["gpt-5.4"] |
| 157 | |
| 158 | img_path = "code/result/image/test_avail/test_input.png" |
| 159 | if not os.path.exists(img_path): |
| 160 | print(f"✗ 测试图片不存在: {img_path}") |
| 161 | else: |
| 162 | text = "请描述这张图片的内容" |
| 163 | for model in MODELS: |
| 164 | print(f"\n--- 测试模型: {model} ---") |
| 165 | print(f"输入文本: {text}") |
| 166 | print(f"输入图片: {img_path}") |
| 167 | t0 = time.time() |
| 168 | try: |
| 169 | result = client.chat(text=text, images=[img_path], model=model) |
| 170 | elapsed = time.time() - t0 |
| 171 | if result: |
| 172 | print(f"✓ 返回结果 ({elapsed:.1f}s): {result[:200]}") |
| 173 | else: |
| 174 | print(f"✗ 返回空结果 ({elapsed:.1f}s)") |
| 175 | except Exception as e: |
| 176 | print(f"✗ 失败: {e}") |
| 177 |