mirror of
https://github.com/snailyp/gemini-balance.git
synced 2026-05-16 13:27:35 +08:00
ef4a528611
引入了思考模型 (THINKING_MODELS) 和相应的预算映射 (THINKING_BUDGET_MAP) 的概念，允许在配置中指定用于特定内部处理流程（如“思考过程”）的模型及其 token 预算。 主要变更包括： 后端 (Python): - 在 `Settings` 中添加了 `THINKING_MODELS` (List[str]) 和 `THINKING_BUDGET_MAP` (Dict[str, float]) 配置项。 - 增强了 `config._parse_db_value` 函数，以正确解析来自数据库或环境变量的列表和字典字符串（包括处理单引号和提供更详细的日志）。 - 更新了相关服务（如 `GeminiChatService`, `ModelService`, `ConfigService`）以识别和利用这些新配置。 - 调整了中间件和路由以适应可能的逻辑变更。 前端 (HTML/JavaScript): - 在配置编辑器 (`config_editor.html`, `config_editor.js`) 中添加了新的 UI 部分来管理思考模型列表和预算映射。 - 实现了动态添加/删除思考模型的功能，并自动关联/解除关联对应的预算映射条目。 - 预算映射中的模型名称（键）是只读的，自动从思考模型列表同步；预算值（值）是可编辑的数字输入。 - 更新了表单数据的加载 (`populateForm`) 和收集 (`collectFormData`) 逻辑，以正确处理新的列表和映射类型。 - 移除了手动添加预算映射的按钮，改为自动关联。 - 改进了数组和映射项的 DOM 操作逻辑，包括使用 UUID 来关联模型和预算项。
65 lines
2.6 KiB
Python
65 lines
2.6 KiB
Python
# app/services/chat/api_client.py
|
|
|
|
from typing import Dict, Any, AsyncGenerator
|
|
import httpx
|
|
from abc import ABC, abstractmethod
|
|
|
|
from app.core.constants import DEFAULT_TIMEOUT
|
|
|
|
|
|
class ApiClient(ABC):
|
|
"""API客户端基类"""
|
|
|
|
@abstractmethod
|
|
async def generate_content(self, payload: Dict[str, Any], model: str, api_key: str) -> Dict[str, Any]:
|
|
pass
|
|
|
|
@abstractmethod
|
|
async def stream_generate_content(self, payload: Dict[str, Any], model: str, api_key: str) -> AsyncGenerator[str, None]:
|
|
pass
|
|
|
|
|
|
class GeminiApiClient(ApiClient):
|
|
"""Gemini API客户端"""
|
|
|
|
def __init__(self, base_url: str, timeout: int = DEFAULT_TIMEOUT):
|
|
self.base_url = base_url
|
|
self.timeout = timeout
|
|
|
|
def _get_real_model(self, model: str) -> str:
|
|
if model.endswith("-search"):
|
|
model = model[:-7]
|
|
if model.endswith("-image"):
|
|
model = model[:-6]
|
|
if model.endswith("-non-thinking"):
|
|
model = model[:-13]
|
|
if "-search" in model and "-non-thinking" in model:
|
|
model = model[:-20]
|
|
return model
|
|
|
|
async def generate_content(self, payload: Dict[str, Any], model: str, api_key: str) -> Dict[str, Any]:
|
|
timeout = httpx.Timeout(self.timeout, read=self.timeout)
|
|
model = self._get_real_model(model)
|
|
|
|
async with httpx.AsyncClient(timeout=timeout) as client:
|
|
url = f"{self.base_url}/models/{model}:generateContent?key={api_key}"
|
|
response = await client.post(url, json=payload)
|
|
if response.status_code != 200:
|
|
error_content = response.text
|
|
raise Exception(f"API call failed with status code {response.status_code}, {error_content}")
|
|
return response.json()
|
|
|
|
async def stream_generate_content(self, payload: Dict[str, Any], model: str, api_key: str) -> AsyncGenerator[str, None]:
|
|
timeout = httpx.Timeout(self.timeout, read=self.timeout)
|
|
model = self._get_real_model(model)
|
|
|
|
async with httpx.AsyncClient(timeout=timeout) as client:
|
|
url = f"{self.base_url}/models/{model}:streamGenerateContent?alt=sse&key={api_key}"
|
|
async with client.stream(method="POST", url=url, json=payload) as response:
|
|
if response.status_code != 200:
|
|
error_content = await response.aread()
|
|
error_msg = error_content.decode("utf-8")
|
|
raise Exception(f"API call failed with status code {response.status_code}, {error_msg}")
|
|
async for line in response.aiter_lines():
|
|
yield line
|