| 1 | # -*- coding: utf-8 -*- |
| 2 | """ |
| 3 | Google Gemini 多模态大模型 API 客户端 (OpenAI 兼容格式) |
| 4 | 支持 gemini-2.5-flash-image, gemini-2.5-pro 等视觉模型 |
| 5 | |
| 6 | 可用模型: |
| 7 | - gemini-2.5-flash-image (性价比最高) |
| 8 | - gemini-2.5-pro (效果最好) |
| 9 | - gemini-3-pro-preview |
| 10 | - gemini-3-pro-image-preview |
| 11 | """ |
| 12 | |
| 13 | import os |
| 14 | import sys |
| 15 | |
| 16 | models_dir = os.path.dirname(os.path.abspath(__file__)) |
| 17 | backend_dir = os.path.dirname(models_dir) |
| 18 | if backend_dir not in sys.path: |
| 19 | sys.path.insert(0, backend_dir) |
| 20 | |
| 21 | import time |
| 22 | import base64 |
| 23 | import logging |
| 24 | import httpx |
| 25 | from openai import OpenAI |
| 26 | from typing import Dict, List, Optional |
| 27 | from config import Config |
| 28 | |
| 29 | logger = logging.getLogger(__name__) |
| 30 | |
| 31 | |
| 32 | class GeminiVLClient: |
| 33 | """ |
| 34 | Gemini VLM 客户端，使用 OpenAI 兼容格式调用 |
| 35 | """ |
| 36 | def __init__(self, api_key: Optional[str] = None, base_url: Optional[str] = None): |
| 37 | """ |
| 38 | Gemini 多模态客户端 |
| 39 | :param api_key: Gemini API Key |
| 40 | :param base_url: 自定义 Base URL（可选，用于代理） |
| 41 | """ |
| 42 | self.api_key = api_key or Config.GEMINI_API_KEY |
| 43 | default_url = "https://generativelanguage.googleapis.com/v1beta" |
| 44 | self.base_url = base_url or Config.GOOGLE_GEMINI_BASE_URL or default_url |
| 45 | if self.base_url and not self.base_url.endswith("/v1"): |
| 46 | self.base_url = self.base_url.rstrip("/") + "/v1" |
| 47 | kwargs = {"api_key": self.api_key, "base_url": self.base_url} |
| 48 | proxy = Config.provider_proxy("gemini") |
| 49 | if proxy: |
| 50 | kwargs["http_client"] = httpx.Client(proxy=proxy) |
| 51 | self.client = OpenAI(**kwargs) |
| 52 | self.max_attempts = 10 |
| 53 | |
| 54 | def _encode_image(self, image_path: str) -> str: |
| 55 | """将本地图片编码为 base64""" |
| 56 | abs_path = os.path.abspath(image_path) |
| 57 | with open(abs_path, "rb") as f: |
| 58 | return base64.b64encode(f.read()).decode("utf-8") |
| 59 | |
| 60 | def _get_mime_type(self, image_path: str) -> str: |
| 61 | """根据文件扩展名获取 MIME 类型""" |
| 62 | ext = os.path.splitext(image_path)[1].lower() |
| 63 | mime_types = { |
| 64 | ".jpg": "image/jpeg", |
| 65 | ".jpeg": "image/jpeg", |
| 66 | ".png": "image/png", |
| 67 | ".webp": "image/webp", |
| 68 | ".gif": "image/gif" |
| 69 | } |
| 70 | return mime_types.get(ext, "image/jpeg") |
| 71 | |
| 72 | def chat(self, text: str, images: List[str], model: str = "gemini-2.5-flash-image", |
| 73 | parameters: Optional[Dict] = None) -> str: |
| 74 | """ |
| 75 | 使用 Gemini 进行多模态对话（文本+图片） |
| 76 | :param text: 文本内容 |
| 77 | :param images: 图片路径列表（支持本地路径或URL） |
| 78 | :param model: 模型名（如 gemini-2.5-flash-image, gemini-2.5-pro） |
| 79 | :param parameters: 其他API参数 |
| 80 | :return: API响应内容 |
| 81 | """ |
| 82 | # 构建消息格式 |
| 83 | content: list = [{"type": "text", "text": text}] |
| 84 | |
| 85 | # 处理图片 |
| 86 | if images: |
| 87 | for img_path in images: |
| 88 | if img_path.startswith("data:"): |
| 89 | # Base64 数据 URL，直接使用 |
| 90 | content.append({ |
| 91 | "type": "image_url", |
| 92 | "image_url": {"url": img_path} |
| 93 | }) |
| 94 | elif img_path.startswith("http"): |
| 95 | # URL 图片 |
| 96 | content.append({ |
| 97 | "type": "image_url", |
| 98 | "image_url": {"url": img_path} |
| 99 | }) |
| 100 | else: |
| 101 | # 本地图片 - 转为 base64 |
| 102 | mime_type = self._get_mime_type(img_path) |
| 103 | base64_data = self._encode_image(img_path) |
| 104 | data_url = f"data:{mime_type};base64,{base64_data}" |
| 105 | content.append({ |
| 106 | "type": "image_url", |
| 107 | "image_url": {"url": data_url} |
| 108 | }) |
| 109 | |
| 110 | messages = [{"role": "user", "content": content}] |
| 111 | |
| 112 | attempts = 0 |
| 113 | while attempts < self.max_attempts: |
| 114 | try: |
| 115 | # 直接使用模型名 |
| 116 | response = self.client.chat.completions.create( |
| 117 | model=model, |
| 118 | messages=messages, |
| 119 | temperature=parameters.get("temperature", 0.7) if parameters else 0.7 |
| 120 | ) |
| 121 | |
| 122 | if response.choices and len(response.choices) > 0: |
| 123 | return response.choices[0].message.content |
| 124 | |
| 125 | except Exception as e: |
| 126 | logger.warning("Gemini VLM request failed: %s", e) |
| 127 | attempts += 1 |
| 128 | if attempts < self.max_attempts: |
| 129 | time.sleep(10) |
| 130 | |
| 131 | raise Exception("GeminiVL: 达到最大重试次数，仍未获得有效响应。") |
| 132 | |
| 133 | |
| 134 | if __name__ == "__main__": |
| 135 | import sys |
| 136 | sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) |
| 137 | from config import Config |
| 138 | |
| 139 | # 支持的 VLM 模型列表 |
| 140 | MODELS = ["gemini-2.5-flash-image", "gemini-2.0-flash"] |
| 141 | |
| 142 | print("=== Gemini VL 多模态可用性测试 ===") |
| 143 | api_key = Config.GEMINI_API_KEY |
| 144 | if not api_key: |
| 145 | print("✗ GEMINI_API_KEY 未设置，跳过") |
| 146 | sys.exit(1) |
| 147 | print(f" API Key: {api_key[:6]}***") |
| 148 | client = GeminiVLClient(api_key=api_key) |
| 149 | |
| 150 | # 测试图片（使用示例图片） |
| 151 | img_path = "" |
| 152 | if not os.path.exists(img_path): |
| 153 | print(f"✗ 测试图片不存在: {img_path}") |
| 154 | img_path = "backend/code/result/image/test_avail/test_input.png" |
| 155 | if not os.path.exists(img_path): |
| 156 | print("✗ 跳过图片测试（无测试图片）") |
| 157 | sys.exit(0) |
| 158 | |
| 159 | text = "请描述这张图片的内容" |
| 160 | print(f"\n[多模态] Prompt: {text}") |
| 161 | print(f" 图片: {img_path}") |
| 162 | |
| 163 | for model in MODELS: |
| 164 | print(f"\n--- 测试模型: {model} ---") |
| 165 | t0 = time.time() |
| 166 | try: |
| 167 | result = client.chat(text=text, images=[img_path], model=model) |
| 168 | elapsed = time.time() - t0 |
| 169 | if result: |
| 170 | print(f"✓ 返回结果 ({elapsed:.1f}s): {result[:200]}") |
| 171 | else: |
| 172 | print(f"✗ 返回空结果 ({elapsed:.1f}s)") |
| 173 | except Exception as e: |
| 174 | print(f"✗ 失败: {e}") |
| 175 |