mirror of
https://github.com/snailyp/gemini-balance.git
synced 2026-09-04 23:19:17 +08:00
refactor: 项目结构优化与FastAPI生命周期更新
重构项目目录结构,提高代码组织性和可维护性 将schemas目录重命名为domain,更好地表达领域模型概念 将services目录细分为service/chat、service/image等子目录 将api目录重命名为router,更符合FastAPI惯例 创建utils目录存放通用工具函数 更新FastAPI应用程序生命周期管理 替换已弃用的on_event方法为推荐的lifespan事件处理器 添加应用程序关闭时的日志记录 代码质量改进 抽取常量到constants.py,减少硬编码值 添加helpers.py提供通用工具函数 优化配置管理,使用环境变量和默认值 完善文档字符串,提高代码可读性
This commit is contained in:
@@ -0,0 +1,164 @@
|
||||
# app/services/chat/message_converter.py
|
||||
|
||||
from abc import ABC, abstractmethod
|
||||
import re
|
||||
from typing import Any, Dict, List, Optional
|
||||
import requests
|
||||
import base64
|
||||
|
||||
from app.core.constants import DATA_URL_PATTERN, IMAGE_URL_PATTERN, SUPPORTED_ROLES
|
||||
|
||||
|
||||
class MessageConverter(ABC):
|
||||
"""消息转换器基类"""
|
||||
|
||||
@abstractmethod
|
||||
def convert(self, messages: List[Dict[str, Any]]) -> tuple[List[Dict[str, Any]], Optional[Dict[str, Any]]]:
|
||||
pass
|
||||
|
||||
def _get_mime_type_and_data(base64_string):
|
||||
"""
|
||||
从 base64 字符串中提取 MIME 类型和数据。
|
||||
|
||||
参数:
|
||||
base64_string (str): 可能包含 MIME 类型信息的 base64 字符串
|
||||
|
||||
返回:
|
||||
tuple: (mime_type, encoded_data)
|
||||
"""
|
||||
# 检查字符串是否以 "data:" 格式开始
|
||||
if base64_string.startswith('data:'):
|
||||
# 提取 MIME 类型和数据
|
||||
pattern = DATA_URL_PATTERN
|
||||
match = re.match(pattern, base64_string)
|
||||
if match:
|
||||
mime_type = "image/jpeg" if match.group(1) == "image/jpg" else match.group(1)
|
||||
encoded_data = match.group(2)
|
||||
return mime_type, encoded_data
|
||||
|
||||
# 如果不是预期格式,假定它只是数据部分
|
||||
return None, base64_string
|
||||
|
||||
def _convert_image(image_url: str) -> Dict[str, Any]:
|
||||
if image_url.startswith("data:image"):
|
||||
mime_type, encoded_data = _get_mime_type_and_data(image_url)
|
||||
return {
|
||||
"inline_data": {
|
||||
"mime_type": mime_type,
|
||||
"data": encoded_data
|
||||
}
|
||||
}
|
||||
return {
|
||||
"image_url": {
|
||||
"url": image_url
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
def _convert_image_to_base64(url: str) -> str:
|
||||
"""
|
||||
将图片URL转换为base64编码
|
||||
Args:
|
||||
url: 图片URL
|
||||
Returns:
|
||||
str: base64编码的图片数据
|
||||
"""
|
||||
response = requests.get(url)
|
||||
if response.status_code == 200:
|
||||
# 将图片内容转换为base64
|
||||
img_data = base64.b64encode(response.content).decode('utf-8')
|
||||
return img_data
|
||||
else:
|
||||
raise Exception(f"Failed to fetch image: {response.status_code}")
|
||||
|
||||
|
||||
def _process_text_with_image(text: str) -> List[Dict[str, Any]]:
|
||||
"""
|
||||
处理可能包含图片URL的文本,提取图片并转换为base64
|
||||
|
||||
Args:
|
||||
text: 可能包含图片URL的文本
|
||||
|
||||
Returns:
|
||||
List[Dict[str, Any]]: 包含文本和图片的部分列表
|
||||
"""
|
||||
parts = []
|
||||
img_url_match = re.search(IMAGE_URL_PATTERN, text)
|
||||
if img_url_match:
|
||||
# 提取URL
|
||||
img_url = img_url_match.group(2)
|
||||
# 将URL对应的图片转换为base64
|
||||
try:
|
||||
base64_data = _convert_image_to_base64(img_url)
|
||||
parts.append({
|
||||
"inlineData": {
|
||||
"mimeType": "image/png",
|
||||
"data": base64_data
|
||||
}
|
||||
})
|
||||
except Exception:
|
||||
# 如果转换失败,回退到文本模式
|
||||
parts.append({"text": text})
|
||||
else:
|
||||
# 没有图片URL,作为纯文本处理
|
||||
parts.append({"text": text})
|
||||
return parts
|
||||
|
||||
|
||||
class OpenAIMessageConverter(MessageConverter):
|
||||
"""OpenAI消息格式转换器"""
|
||||
|
||||
def convert(self, messages: List[Dict[str, Any]]) -> tuple[List[Dict[str, Any]], Optional[Dict[str, Any]]]:
|
||||
converted_messages = []
|
||||
system_instruction_parts = []
|
||||
|
||||
for idx, msg in enumerate(messages):
|
||||
role = msg.get("role", "")
|
||||
if role not in SUPPORTED_ROLES:
|
||||
if role == "tool":
|
||||
role = "user"
|
||||
else:
|
||||
# 如果是最后一条消息,则认为是用户消息
|
||||
if idx == len(messages) - 1:
|
||||
role = "user"
|
||||
else:
|
||||
role = "model"
|
||||
|
||||
parts = []
|
||||
# 特别处理最后一个assistant的消息,按\n\n分割
|
||||
if role == "assistant" and idx == len(messages) - 2 and isinstance(msg["content"], str) and msg["content"]:
|
||||
# 按\n\n分割消息
|
||||
content_parts = msg["content"].split("\n\n")
|
||||
for part in content_parts:
|
||||
if not part.strip(): # 跳过空内容
|
||||
continue
|
||||
# 处理可能包含图片的文本
|
||||
parts.extend(_process_text_with_image(part))
|
||||
elif isinstance(msg["content"], str) and msg["content"]:
|
||||
# 请求 gemini 接口时如果包含 content 字段但内容为空时会返回 400 错误,所以需要判断是否为空并移除
|
||||
parts.extend(_process_text_with_image(msg["content"]))
|
||||
elif isinstance(msg["content"], list):
|
||||
for content in msg["content"]:
|
||||
if isinstance(content, str) and content:
|
||||
parts.append({"text": content})
|
||||
elif isinstance(content, dict):
|
||||
if content["type"] == "text" and content["text"]:
|
||||
parts.append({"text": content["text"]})
|
||||
elif content["type"] == "image_url":
|
||||
parts.append(_convert_image(content["image_url"]["url"]))
|
||||
|
||||
if parts:
|
||||
if role == "system":
|
||||
system_instruction_parts.extend(parts)
|
||||
else:
|
||||
converted_messages.append({"role": role, "parts": parts})
|
||||
|
||||
system_instruction = (
|
||||
None
|
||||
if not system_instruction_parts
|
||||
else {
|
||||
"role": "system",
|
||||
"parts": system_instruction_parts,
|
||||
}
|
||||
)
|
||||
return converted_messages, system_instruction
|
||||
Reference in New Issue
Block a user