mirror of
https://github.com/Zippland/Snap-Solver.git
synced 2026-01-19 09:41:15 +08:00
397 lines
18 KiB
Python
397 lines
18 KiB
Python
import json
|
||
import requests
|
||
import os
|
||
from typing import Generator
|
||
from openai import OpenAI
|
||
from .base import BaseModel
|
||
|
||
class DeepSeekModel(BaseModel):
|
||
def __init__(self, api_key: str, temperature: float = 0.7, system_prompt: str = None, language: str = None, model_name: str = "deepseek-reasoner"):
|
||
super().__init__(api_key, temperature, system_prompt, language)
|
||
self.model_name = model_name
|
||
|
||
def get_default_system_prompt(self) -> str:
|
||
return """You are an expert at analyzing questions and providing detailed solutions. When presented with an image of a question:
|
||
1. First read and understand the question carefully
|
||
2. Break down the key components of the question
|
||
3. Provide a clear, step-by-step solution
|
||
4. If relevant, explain any concepts or theories involved
|
||
5. If there are multiple approaches, explain the most efficient one first"""
|
||
|
||
def get_model_identifier(self) -> str:
|
||
"""根据模型名称返回正确的API标识符"""
|
||
# 通过模型名称来确定实际的API调用标识符
|
||
if self.model_name == "deepseek-chat":
|
||
return "deepseek-chat"
|
||
# deepseek-reasoner是默认的推理模型名称
|
||
return "deepseek-reasoner"
|
||
|
||
def analyze_text(self, text: str, proxies: dict = None) -> Generator[dict, None, None]:
|
||
"""Stream DeepSeek's response for text analysis"""
|
||
try:
|
||
# Initial status
|
||
yield {"status": "started", "content": ""}
|
||
|
||
# 保存原始环境变量
|
||
original_env = {
|
||
'http_proxy': os.environ.get('http_proxy'),
|
||
'https_proxy': os.environ.get('https_proxy')
|
||
}
|
||
|
||
try:
|
||
# 如果提供了代理设置，通过环境变量设置
|
||
if proxies:
|
||
if 'http' in proxies:
|
||
os.environ['http_proxy'] = proxies['http']
|
||
if 'https' in proxies:
|
||
os.environ['https_proxy'] = proxies['https']
|
||
|
||
# 初始化DeepSeek客户端，不再使用session对象
|
||
client = OpenAI(
|
||
api_key=self.api_key,
|
||
base_url="https://api.deepseek.com"
|
||
)
|
||
|
||
# 添加系统语言指令
|
||
system_prompt = self.system_prompt
|
||
language = self.language or '中文'
|
||
if not any(phrase in system_prompt for phrase in ['Please respond in', '请用', '使用', '回答']):
|
||
system_prompt = f"{system_prompt}\n\n请务必使用{language}回答。"
|
||
|
||
# 构建请求参数
|
||
params = {
|
||
"model": self.get_model_identifier(),
|
||
"messages": [
|
||
{
|
||
'role': 'system',
|
||
'content': system_prompt
|
||
},
|
||
{
|
||
'role': 'user',
|
||
'content': text
|
||
}
|
||
],
|
||
"stream": True
|
||
}
|
||
|
||
# 只有非推理模型才设置temperature参数
|
||
if not self.model_name.endswith('reasoner') and self.temperature is not None:
|
||
params["temperature"] = self.temperature
|
||
|
||
print(f"调用DeepSeek API: {self.get_model_identifier()}, 是否设置温度: {not self.model_name.endswith('reasoner')}")
|
||
|
||
response = client.chat.completions.create(**params)
|
||
|
||
# 使用两个缓冲区，分别用于常规内容和思考内容
|
||
response_buffer = ""
|
||
thinking_buffer = ""
|
||
|
||
for chunk in response:
|
||
# 打印chunk以调试
|
||
try:
|
||
print(f"DeepSeek API返回chunk: {chunk}")
|
||
except:
|
||
print("无法打印chunk")
|
||
|
||
try:
|
||
# 处理推理模型的思考内容
|
||
if hasattr(chunk.choices[0].delta, 'reasoning_content'):
|
||
content = chunk.choices[0].delta.reasoning_content
|
||
if content:
|
||
# 累积思考内容
|
||
thinking_buffer += content
|
||
|
||
# 只在积累一定数量的字符或遇到句子结束标记时才发送
|
||
if len(content) >= 20 or content.endswith(('.', '!', '?', '。', '！', '？', '\n')):
|
||
yield {
|
||
"status": "thinking",
|
||
"content": thinking_buffer
|
||
}
|
||
|
||
# 处理常规内容
|
||
elif hasattr(chunk.choices[0].delta, 'content'):
|
||
content = chunk.choices[0].delta.content
|
||
if content:
|
||
# 累积响应内容
|
||
response_buffer += content
|
||
print(f"累积响应内容: '{content}', 当前buffer: '{response_buffer}'")
|
||
|
||
# 只在积累一定数量的字符或遇到句子结束标记时才发送
|
||
if len(content) >= 10 or content.endswith(('.', '!', '?', '。', '！', '？', '\n')):
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
# 尝试直接从message内容获取
|
||
elif hasattr(chunk.choices[0], 'message') and hasattr(chunk.choices[0].message, 'content'):
|
||
content = chunk.choices[0].message.content
|
||
if content:
|
||
response_buffer += content
|
||
print(f"从message获取内容: '{content}'")
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
# 检查是否有finish_reason，表示生成结束
|
||
elif hasattr(chunk.choices[0], 'finish_reason') and chunk.choices[0].finish_reason:
|
||
print(f"生成结束，原因: {chunk.choices[0].finish_reason}")
|
||
|
||
# 如果没有内容但有思考内容，把思考内容作为正文显示
|
||
if not response_buffer and thinking_buffer:
|
||
response_buffer = thinking_buffer
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
except Exception as e:
|
||
print(f"解析响应chunk时出错: {str(e)}")
|
||
continue
|
||
|
||
# 确保发送最终的缓冲内容
|
||
if thinking_buffer:
|
||
yield {
|
||
"status": "thinking_complete",
|
||
"content": thinking_buffer
|
||
}
|
||
|
||
# 如果推理完成后没有正文内容，则使用思考内容作为最终响应
|
||
if not response_buffer and thinking_buffer:
|
||
response_buffer = thinking_buffer
|
||
|
||
# 发送最终响应内容
|
||
if response_buffer:
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
|
||
# 发送完成状态
|
||
yield {
|
||
"status": "completed",
|
||
"content": response_buffer
|
||
}
|
||
|
||
except Exception as e:
|
||
error_msg = str(e)
|
||
print(f"DeepSeek API调用出错: {error_msg}")
|
||
|
||
# 提供具体的错误信息
|
||
if "invalid_api_key" in error_msg.lower():
|
||
error_msg = "DeepSeek API密钥无效，请检查您的API密钥"
|
||
elif "rate_limit" in error_msg.lower():
|
||
error_msg = "DeepSeek API请求频率超限，请稍后再试"
|
||
elif "quota_exceeded" in error_msg.lower():
|
||
error_msg = "DeepSeek API配额已用完，请续费或等待下个计费周期"
|
||
|
||
yield {
|
||
"status": "error",
|
||
"error": f"DeepSeek API错误: {error_msg}"
|
||
}
|
||
finally:
|
||
# 恢复原始环境变量
|
||
for key, value in original_env.items():
|
||
if value is None:
|
||
if key in os.environ:
|
||
del os.environ[key]
|
||
else:
|
||
os.environ[key] = value
|
||
|
||
except Exception as e:
|
||
error_msg = str(e)
|
||
print(f"调用DeepSeek模型时发生错误: {error_msg}")
|
||
|
||
if "invalid_api_key" in error_msg.lower():
|
||
error_msg = "API密钥无效，请检查设置"
|
||
elif "rate_limit" in error_msg.lower():
|
||
error_msg = "API请求频率超限，请稍后再试"
|
||
|
||
yield {
|
||
"status": "error",
|
||
"error": f"DeepSeek API错误: {error_msg}"
|
||
}
|
||
|
||
def analyze_image(self, image_data: str, proxies: dict = None) -> Generator[dict, None, None]:
|
||
"""Stream DeepSeek's response for image analysis"""
|
||
try:
|
||
# 检查我们是否有支持图像的模型
|
||
if self.model_name == "deepseek-chat" or self.model_name == "deepseek-reasoner":
|
||
yield {
|
||
"status": "error",
|
||
"error": "当前DeepSeek模型不支持图像分析，请使用Anthropic或OpenAI的多模态模型"
|
||
}
|
||
return
|
||
|
||
# Initial status
|
||
yield {"status": "started", "content": ""}
|
||
|
||
# 保存原始环境变量
|
||
original_env = {
|
||
'http_proxy': os.environ.get('http_proxy'),
|
||
'https_proxy': os.environ.get('https_proxy')
|
||
}
|
||
|
||
try:
|
||
# 如果提供了代理设置，通过环境变量设置
|
||
if proxies:
|
||
if 'http' in proxies:
|
||
os.environ['http_proxy'] = proxies['http']
|
||
if 'https' in proxies:
|
||
os.environ['https_proxy'] = proxies['https']
|
||
|
||
# 初始化DeepSeek客户端，不再使用session对象
|
||
client = OpenAI(
|
||
api_key=self.api_key,
|
||
base_url="https://api.deepseek.com"
|
||
)
|
||
|
||
# 检查系统提示词是否已包含语言设置指令
|
||
system_prompt = self.system_prompt
|
||
language = self.language or '中文'
|
||
if not any(phrase in system_prompt for phrase in ['Please respond in', '请用', '使用', '回答']):
|
||
system_prompt = f"{system_prompt}\n\n请务必使用{language}回答，无论问题是什么语言。即使在分析图像时也请使用{language}回答。"
|
||
|
||
# 构建请求参数
|
||
params = {
|
||
"model": self.get_model_identifier(),
|
||
"messages": [
|
||
{
|
||
'role': 'system',
|
||
'content': system_prompt
|
||
},
|
||
{
|
||
'role': 'user',
|
||
'content': f"Here's an image of a question to analyze: data:image/png;base64,{image_data}"
|
||
}
|
||
],
|
||
"stream": True
|
||
}
|
||
|
||
# 只有非推理模型才设置temperature参数
|
||
if not self.model_name.endswith('reasoner') and self.temperature is not None:
|
||
params["temperature"] = self.temperature
|
||
|
||
response = client.chat.completions.create(**params)
|
||
|
||
# 使用两个缓冲区，分别用于常规内容和思考内容
|
||
response_buffer = ""
|
||
thinking_buffer = ""
|
||
|
||
for chunk in response:
|
||
# 打印chunk以调试
|
||
try:
|
||
print(f"DeepSeek图像API返回chunk: {chunk}")
|
||
except:
|
||
print("无法打印chunk")
|
||
|
||
try:
|
||
# 处理推理模型的思考内容
|
||
if hasattr(chunk.choices[0].delta, 'reasoning_content'):
|
||
content = chunk.choices[0].delta.reasoning_content
|
||
if content:
|
||
# 累积思考内容
|
||
thinking_buffer += content
|
||
|
||
# 只在积累一定数量的字符或遇到句子结束标记时才发送
|
||
if len(content) >= 20 or content.endswith(('.', '!', '?', '。', '！', '？', '\n')):
|
||
yield {
|
||
"status": "thinking",
|
||
"content": thinking_buffer
|
||
}
|
||
# 处理常规内容
|
||
elif hasattr(chunk.choices[0].delta, 'content'):
|
||
content = chunk.choices[0].delta.content
|
||
if content:
|
||
# 累积响应内容
|
||
response_buffer += content
|
||
print(f"累积图像响应内容: '{content}', 当前buffer: '{response_buffer}'")
|
||
|
||
# 只在积累一定数量的字符或遇到句子结束标记时才发送
|
||
if len(content) >= 10 or content.endswith(('.', '!', '?', '。', '！', '？', '\n')):
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
# 尝试直接从message内容获取
|
||
elif hasattr(chunk.choices[0], 'message') and hasattr(chunk.choices[0].message, 'content'):
|
||
content = chunk.choices[0].message.content
|
||
if content:
|
||
response_buffer += content
|
||
print(f"从message获取图像内容: '{content}'")
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
# 检查是否有finish_reason，表示生成结束
|
||
elif hasattr(chunk.choices[0], 'finish_reason') and chunk.choices[0].finish_reason:
|
||
print(f"图像生成结束，原因: {chunk.choices[0].finish_reason}")
|
||
|
||
# 如果没有内容但有思考内容，把思考内容作为正文显示
|
||
if not response_buffer and thinking_buffer:
|
||
response_buffer = thinking_buffer
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
except Exception as e:
|
||
print(f"解析图像响应chunk时出错: {str(e)}")
|
||
continue
|
||
|
||
# 确保发送最终的缓冲内容
|
||
if thinking_buffer:
|
||
yield {
|
||
"status": "thinking_complete",
|
||
"content": thinking_buffer
|
||
}
|
||
|
||
# 如果推理完成后没有正文内容，则使用思考内容作为最终响应
|
||
if not response_buffer and thinking_buffer:
|
||
response_buffer = thinking_buffer
|
||
|
||
# 发送最终响应内容
|
||
if response_buffer:
|
||
yield {
|
||
"status": "streaming",
|
||
"content": response_buffer
|
||
}
|
||
|
||
# 发送完成状态
|
||
yield {
|
||
"status": "completed",
|
||
"content": response_buffer
|
||
}
|
||
|
||
except Exception as e:
|
||
error_msg = str(e)
|
||
print(f"DeepSeek API调用出错: {error_msg}")
|
||
|
||
# 提供具体的错误信息
|
||
if "invalid_api_key" in error_msg.lower():
|
||
error_msg = "DeepSeek API密钥无效，请检查您的API密钥"
|
||
elif "rate_limit" in error_msg.lower():
|
||
error_msg = "DeepSeek API请求频率超限，请稍后再试"
|
||
|
||
yield {
|
||
"status": "error",
|
||
"error": f"DeepSeek API错误: {error_msg}"
|
||
}
|
||
finally:
|
||
# 恢复原始环境变量
|
||
for key, value in original_env.items():
|
||
if value is None:
|
||
if key in os.environ:
|
||
del os.environ[key]
|
||
else:
|
||
os.environ[key] = value
|
||
|
||
except Exception as e:
|
||
error_msg = str(e)
|
||
if "invalid_api_key" in error_msg.lower():
|
||
error_msg = "API密钥无效，请检查设置"
|
||
elif "rate_limit" in error_msg.lower():
|
||
error_msg = "API请求频率超限，请稍后再试"
|
||
|
||
yield {
|
||
"status": "error",
|
||
"error": f"DeepSeek API错误: {error_msg}"
|
||
}
|