mirror of
https://github.com/linyqh/NarratoAI.git
synced 2026-09-16 17:06:32 +00:00
e6e39d2dcd
- 新增短剧解说全流程四类提示词模板：解说文案生成、片段规划、文案画面匹配、脚本修复 - 重构原有脚本生成提示词至v2.1，改为基于上游规划片段生成合规解说脚本 - 为LLM基础服务层新增流式文本生成接口，完善OpenAI兼容提供商的流式实现，支持流式回调与推理内容提取 - 重构OpenAI兼容文本提供商的生成逻辑，提取公共参数构建方法 - 新增多语言国际化文案，覆盖解说语言、短剧类型、原片占比等配置项与交互提示 - 新增多套单元测试，覆盖脚本校验、适配器流程、工具函数等模块 - 封装SubtitleAnalyzerAdapter，统一短剧解说脚本生成的整套业务接口 - 新增前端交互所需的解说文案审核相关提示文案
303 lines
11 KiB
Python
303 lines
11 KiB
Python
"""
|
||
统一的大模型服务接口
|
||
|
||
提供简化的API接口，方便现有代码迁移到新的架构
|
||
"""
|
||
|
||
from typing import List, Dict, Any, Optional, Union
|
||
from pathlib import Path
|
||
import PIL.Image
|
||
from loguru import logger
|
||
|
||
from .manager import LLMServiceManager
|
||
from .validators import OutputValidator
|
||
from .exceptions import LLMServiceError
|
||
from app.services.prompts import PromptManager
|
||
|
||
# 提供商注册由 webui.py:main() 显式调用（见 LLM 提供商注册机制重构）
|
||
# 这样更可靠，错误也更容易调试
|
||
|
||
|
||
class UnifiedLLMService:
|
||
"""统一的大模型服务接口"""
|
||
|
||
@staticmethod
|
||
async def analyze_images(images: List[Union[str, Path, PIL.Image.Image]],
|
||
prompt: str,
|
||
provider: Optional[str] = None,
|
||
batch_size: int = 10,
|
||
**kwargs) -> List[str]:
|
||
"""
|
||
分析图片内容
|
||
|
||
Args:
|
||
images: 图片路径列表或PIL图片对象列表
|
||
prompt: 分析提示词
|
||
provider: 视觉模型提供商名称，如果不指定则使用配置中的默认值
|
||
batch_size: 批处理大小
|
||
**kwargs: 其他参数
|
||
|
||
Returns:
|
||
分析结果列表
|
||
|
||
Raises:
|
||
LLMServiceError: 服务调用失败时抛出
|
||
"""
|
||
try:
|
||
# 获取视觉模型提供商
|
||
vision_provider = LLMServiceManager.get_vision_provider(provider)
|
||
|
||
# 执行图片分析
|
||
results = await vision_provider.analyze_images(
|
||
images=images,
|
||
prompt=prompt,
|
||
batch_size=batch_size,
|
||
**kwargs
|
||
)
|
||
|
||
logger.info(f"图片分析完成，共处理 {len(images)} 张图片，生成 {len(results)} 个结果")
|
||
return results
|
||
|
||
except Exception as e:
|
||
logger.error(f"图片分析失败: {str(e)}")
|
||
raise LLMServiceError(f"图片分析失败: {str(e)}")
|
||
|
||
@staticmethod
|
||
async def generate_text(prompt: str,
|
||
system_prompt: Optional[str] = None,
|
||
provider: Optional[str] = None,
|
||
temperature: float = 1.0,
|
||
max_tokens: Optional[int] = None,
|
||
response_format: Optional[str] = None,
|
||
**kwargs) -> str:
|
||
"""
|
||
生成文本内容
|
||
|
||
Args:
|
||
prompt: 用户提示词
|
||
system_prompt: 系统提示词
|
||
provider: 文本模型提供商名称，如果不指定则使用配置中的默认值
|
||
temperature: 生成温度
|
||
max_tokens: 最大token数
|
||
response_format: 响应格式 ('json' 或 None)
|
||
**kwargs: 其他参数
|
||
|
||
Returns:
|
||
生成的文本内容
|
||
|
||
Raises:
|
||
LLMServiceError: 服务调用失败时抛出
|
||
"""
|
||
try:
|
||
# 获取文本模型提供商
|
||
text_provider = LLMServiceManager.get_text_provider(provider)
|
||
|
||
# 执行文本生成
|
||
result = await text_provider.generate_text(
|
||
prompt=prompt,
|
||
system_prompt=system_prompt,
|
||
temperature=temperature,
|
||
max_tokens=max_tokens,
|
||
response_format=response_format,
|
||
**kwargs
|
||
)
|
||
|
||
logger.info(f"文本生成完成，生成内容长度: {len(result)} 字符")
|
||
return result
|
||
|
||
except Exception as e:
|
||
logger.error(f"文本生成失败: {str(e)}")
|
||
raise LLMServiceError(f"文本生成失败: {str(e)}")
|
||
|
||
@staticmethod
|
||
async def generate_text_stream(prompt: str,
|
||
system_prompt: Optional[str] = None,
|
||
provider: Optional[str] = None,
|
||
temperature: float = 1.0,
|
||
max_tokens: Optional[int] = None,
|
||
response_format: Optional[str] = None,
|
||
on_chunk=None,
|
||
**kwargs) -> str:
|
||
"""
|
||
流式生成文本内容；不支持流式的 provider 会退化为一次性返回。
|
||
"""
|
||
try:
|
||
text_provider = LLMServiceManager.get_text_provider(provider)
|
||
result = await text_provider.generate_text_stream(
|
||
prompt=prompt,
|
||
system_prompt=system_prompt,
|
||
temperature=temperature,
|
||
max_tokens=max_tokens,
|
||
response_format=response_format,
|
||
on_chunk=on_chunk,
|
||
**kwargs
|
||
)
|
||
|
||
logger.info(f"流式文本生成完成，生成内容长度: {len(result)} 字符")
|
||
return result
|
||
|
||
except Exception as e:
|
||
logger.error(f"流式文本生成失败: {str(e)}")
|
||
raise LLMServiceError(f"流式文本生成失败: {str(e)}")
|
||
|
||
@staticmethod
|
||
async def generate_narration_script(prompt: str,
|
||
provider: Optional[str] = None,
|
||
temperature: float = 1.0,
|
||
validate_output: bool = True,
|
||
**kwargs) -> List[Dict[str, Any]]:
|
||
"""
|
||
生成解说文案
|
||
|
||
Args:
|
||
prompt: 提示词
|
||
provider: 文本模型提供商名称
|
||
temperature: 生成温度
|
||
validate_output: 是否验证输出格式
|
||
**kwargs: 其他参数
|
||
|
||
Returns:
|
||
解说文案列表
|
||
|
||
Raises:
|
||
LLMServiceError: 服务调用失败时抛出
|
||
"""
|
||
try:
|
||
# 生成文本
|
||
result = await UnifiedLLMService.generate_text(
|
||
prompt=prompt,
|
||
provider=provider,
|
||
temperature=temperature,
|
||
response_format="json",
|
||
**kwargs
|
||
)
|
||
|
||
# 验证输出格式
|
||
if validate_output:
|
||
narration_items = OutputValidator.validate_narration_script(result)
|
||
logger.info(f"解说文案生成并验证完成，共 {len(narration_items)} 个片段")
|
||
return narration_items
|
||
else:
|
||
# 简单的JSON解析
|
||
import json
|
||
parsed_result = json.loads(result)
|
||
if "items" in parsed_result:
|
||
return parsed_result["items"]
|
||
else:
|
||
return parsed_result
|
||
|
||
except Exception as e:
|
||
logger.error(f"解说文案生成失败: {str(e)}")
|
||
raise LLMServiceError(f"解说文案生成失败: {str(e)}")
|
||
|
||
@staticmethod
|
||
async def analyze_subtitle(subtitle_content: str,
|
||
provider: Optional[str] = None,
|
||
temperature: float = 1.0,
|
||
validate_output: bool = True,
|
||
**kwargs) -> str:
|
||
"""
|
||
分析字幕内容
|
||
|
||
Args:
|
||
subtitle_content: 字幕内容
|
||
provider: 文本模型提供商名称
|
||
temperature: 生成温度
|
||
validate_output: 是否验证输出格式
|
||
**kwargs: 其他参数
|
||
|
||
Returns:
|
||
分析结果
|
||
|
||
Raises:
|
||
LLMServiceError: 服务调用失败时抛出
|
||
"""
|
||
try:
|
||
prompt = PromptManager.get_prompt(
|
||
category="short_drama_narration",
|
||
name="plot_analysis",
|
||
parameters={"subtitle_content": subtitle_content},
|
||
)
|
||
prompt_object = PromptManager.get_prompt_object(
|
||
category="short_drama_narration",
|
||
name="plot_analysis",
|
||
)
|
||
system_prompt = prompt_object.get_system_prompt()
|
||
|
||
# 生成分析结果
|
||
result = await UnifiedLLMService.generate_text(
|
||
prompt=prompt,
|
||
system_prompt=system_prompt,
|
||
provider=provider,
|
||
temperature=temperature,
|
||
**kwargs
|
||
)
|
||
|
||
# 验证输出格式
|
||
if validate_output:
|
||
validated_result = OutputValidator.validate_subtitle_analysis(result)
|
||
logger.info("字幕分析完成并验证通过")
|
||
return validated_result
|
||
else:
|
||
return result
|
||
|
||
except Exception as e:
|
||
logger.error(f"字幕分析失败: {str(e)}")
|
||
raise LLMServiceError(f"字幕分析失败: {str(e)}")
|
||
|
||
@staticmethod
|
||
def get_provider_info() -> Dict[str, Any]:
|
||
"""
|
||
获取所有提供商信息
|
||
|
||
Returns:
|
||
提供商信息字典
|
||
"""
|
||
return LLMServiceManager.get_provider_info()
|
||
|
||
@staticmethod
|
||
def list_vision_providers() -> List[str]:
|
||
"""
|
||
列出所有视觉模型提供商
|
||
|
||
Returns:
|
||
提供商名称列表
|
||
"""
|
||
return LLMServiceManager.list_vision_providers()
|
||
|
||
@staticmethod
|
||
def list_text_providers() -> List[str]:
|
||
"""
|
||
列出所有文本模型提供商
|
||
|
||
Returns:
|
||
提供商名称列表
|
||
"""
|
||
return LLMServiceManager.list_text_providers()
|
||
|
||
@staticmethod
|
||
def clear_cache():
|
||
"""清空提供商实例缓存"""
|
||
LLMServiceManager.clear_cache()
|
||
logger.info("已清空大模型服务缓存")
|
||
|
||
|
||
# 为了向后兼容，提供一些便捷函数
|
||
async def analyze_images_unified(images: List[Union[str, Path, PIL.Image.Image]],
|
||
prompt: str,
|
||
provider: Optional[str] = None,
|
||
batch_size: int = 10) -> List[str]:
|
||
"""便捷的图片分析函数"""
|
||
return await UnifiedLLMService.analyze_images(images, prompt, provider, batch_size)
|
||
|
||
|
||
async def generate_text_unified(prompt: str,
|
||
system_prompt: Optional[str] = None,
|
||
provider: Optional[str] = None,
|
||
temperature: float = 1.0,
|
||
response_format: Optional[str] = None) -> str:
|
||
"""便捷的文本生成函数"""
|
||
return await UnifiedLLMService.generate_text(
|
||
prompt, system_prompt, provider, temperature, response_format=response_format
|
||
)
|