add ai image base

ai_service/llm/__init__.py

@@ -0,0 +1,50 @@
+import os
+from enum import Enum
+
+from pydantic import SecretStr
+
+
+class ProviderEnum(str, Enum):
+    """支持的 LLM 服务商"""
+    DEEPSEEK = "deepseek"
+    OPENAI = "openai"
+    TONGYI = "tongyi"
+
+class LLMFactory(object):
+
+    @staticmethod
+    def get_llm(provider: ProviderEnum, model: str = None, **kwargs):
+        if provider == ProviderEnum.DEEPSEEK:
+            from langchain_deepseek import ChatDeepSeek
+            api_key = os.getenv("DEEPSEEK_API_KEY")
+            model = model or "deepseek-chat"
+            return ChatDeepSeek(
+                api_key=SecretStr(api_key),
+                model=model,
+                streaming=True,
+                **kwargs
+            )
+
+        elif provider == ProviderEnum.OPENAI:
+            from langchain_openai import ChatOpenAI
+            api_key = os.getenv("OPENAI_API_KEY")
+            model = model or "gpt-3.5-turbo"
+            return ChatOpenAI(
+                api_key=SecretStr(api_key),
+                model=model,
+                streaming=True,
+                **kwargs
+            )
+
+        elif provider == ProviderEnum.TONGYI:
+            from langchain_community.llms import Tongyi
+            api_key = os.getenv("DASHSCOPE_API_KEY")
+            model = model or "qwen-turbo"
+            return Tongyi(
+                api_key=SecretStr(api_key),
+                model=model,
+                streaming=True,
+                **kwargs
+            )
+        else:
+            raise ValueError(f"不支持的 LLM 服务商: {provider}")

ai_service/llm/adapter/deepseek.py

@@ -0,0 +1,17 @@
+from langchain_deepseek import ChatDeepSeek
+
+from llm.base import MultiModalAICapability
+
+
+class DeepSeekAdapter(MultiModalAICapability):
+    def __init__(self, api_key, model, **kwargs):
+
+        self.llm = ChatDeepSeek(api_key=api_key, model=model, streaming=True)
+
+    async def chat(self, messages, **kwargs):
+        # 兼容 DeepSeek 的调用方式
+        return await self.llm.ainvoke(messages)
+
+    async def stream_chat(self, messages, **kwargs):
+        async for chunk in self.llm.astream(messages):
+            yield chunk

ai_service/llm/adapter/genai.py

@@ -0,0 +1,21 @@
+# 假设有 google genai sdk
+# from google_genai import GenAI
+from llm.base import MultiModalAICapability
+
+
+class GoogleGenAIAdapter(MultiModalAICapability):
+    def __init__(self, api_key, model, **kwargs):
+        self.api_key = api_key
+        self.model = model
+        # self.llm = GenAI(api_key=api_key, model=model)
+
+    async def chat(self, messages, **kwargs):
+        # return await self.llm.chat(messages)
+        raise NotImplementedError("Google GenAI chat未实现")
+
+    async def stream_chat(self, messages, **kwargs):
+        # async for chunk in self.llm.stream_chat(messages):
+        #     yield chunk
+        raise NotImplementedError("Google GenAI stream_chat未实现")
+
+    # 其他能力同理

ai_service/llm/adapter/openai.py

@@ -0,0 +1,25 @@
+from llm.base import MultiModalAICapability
+from langchain_openai import ChatOpenAI
+# from openai import OpenAI # 如需图片/音频/视频等API
+
+class OpenAIAdapter(MultiModalAICapability):
+    def __init__(self, api_key, model, **kwargs):
+        self.llm = ChatOpenAI(api_key=api_key, model=model, streaming=True)
+        self.api_key = api_key
+
+    async def chat(self, messages, **kwargs):
+        return await self.llm.ainvoke(messages)
+
+    async def stream_chat(self, messages, **kwargs):
+        async for chunk in self.llm.astream(messages):
+            yield chunk
+
+    # 如需图片生成（DALL·E），可实现如下
+    def create_image_task(self, prompt, **kwargs):
+        # 伪代码，需用 openai.Image.create
+        # import openai
+        # response = openai.Image.create(api_key=self.api_key, prompt=prompt, ...)
+        # return response
+        raise NotImplementedError("OpenAI 图片生成请用 openai.Image.create 实现")
+
+    # 其他能力同理

ai_service/llm/adapter/tongyi.py

@@ -0,0 +1,51 @@
+from langchain_deepseek import ChatDeepSeek
+from http import HTTPStatus
+from urllib.parse import urlparse, unquote
+from pathlib import PurePosixPath
+import requests
+from dashscope import ImageSynthesis
+import os
+
+from llm.base import MultiModalAICapability
+
+
+class TongYiAdapter(MultiModalAICapability):
+    def __init__(self, api_key, model, **kwargs):
+        self.api_key = api_key
+        self.model = model
+        self.llm = ChatDeepSeek(api_key=api_key, model=model, streaming=True)
+
+    async def chat(self, messages, **kwargs):
+        # 兼容 DeepSeek 的调用方式
+        return await self.llm.ainvoke(messages)
+
+    async def stream_chat(self, messages, **kwargs):
+        async for chunk in self.llm.astream(messages):
+            yield chunk
+
+    @staticmethod
+    def create_image_task(api_key, model, prompt: str, style='<watercolor>', size='1024*1024', n=1):
+        """创建异步图片生成任务"""
+        rsp = ImageSynthesis.async_call(
+            api_key=api_key,
+            model=model,
+            prompt=prompt,
+            n=n,
+            style=style,
+            size=size
+        )
+        if rsp.status_code == HTTPStatus.OK:
+            return rsp
+        else:
+            raise Exception(f"Failed, status_code: {rsp.status_code}, code: {rsp.code}, message: {rsp.message}")
+
+    @staticmethod
+    def fetch_image_task_status(task):
+        """获取异步图片任务状态"""
+        status = ImageSynthesis.fetch(task)
+        if status.status_code == HTTPStatus.OK:
+            return status.output.task_status
+        else:
+            raise Exception(f"Failed, status_code: {status.status_code}, code: {status.code}, message: {status.message}")
+
+    

ai_service/llm/base.py

@@ -0,0 +1,37 @@
+from abc import ABC
+
+class MultiModalAICapability(ABC):
+    # 对话能力
+    async def chat(self, messages, **kwargs):
+        raise NotImplementedError("chat not supported by this provider")
+
+    async def stream_chat(self, messages, **kwargs):
+        raise NotImplementedError("stream_chat not supported by this provider")
+
+    # 图片生成能力
+    def create_image_task(self, prompt, **kwargs):
+        raise NotImplementedError("image generation not supported by this provider")
+
+    def fetch_image_task_status(self, task):
+        raise NotImplementedError("image task status not supported by this provider")
+
+    def fetch_image_result(self, task):
+        raise NotImplementedError("image result not supported by this provider")
+
+    # 视频生成能力
+    def create_video_task(self, prompt, **kwargs):
+        raise NotImplementedError("video generation not supported by this provider")
+
+    def fetch_video_task_status(self, task):
+        raise NotImplementedError("video task status not supported by this provider")
+
+    def fetch_video_result(self, task):
+        raise NotImplementedError("video result not supported by this provider")
+
+    # 知识库能力
+    def query_knowledge(self, query, **kwargs):
+        raise NotImplementedError("knowledge query not supported by this provider")
+
+    # 语音合成能力
+    def synthesize_speech(self, text, **kwargs):
+        raise NotImplementedError("speech synthesis not supported by this provider")

ai_service/llm/factory.py

@@ -0,0 +1,34 @@
+from .adapter.deepseek import DeepSeekAdapter
+from .adapter.genai import GoogleGenAIAdapter
+from .adapter.openai import OpenAIAdapter
+from .adapter.tongyi import TongYiAdapter
+
+
+def get_adapter(provider, api_key, model, **kwargs):
+    if provider == 'deepseek':
+        return DeepSeekAdapter(api_key, model, **kwargs)
+    elif provider == 'tongyi':
+        return TongYiAdapter(api_key, model, **kwargs)
+    elif provider == 'openai':
+        return OpenAIAdapter(api_key, model, **kwargs)
+    elif provider == 'google-genai':
+        return GoogleGenAIAdapter(api_key, model, **kwargs)
+    else:
+        raise ValueError('不支持的服务商')
+
+#  使用示例
+# adapter = get_adapter('tongyi', api_key='xxx', model='wanx_v1')
+
+# 对话
+# try:
+#     result = await adapter.chat(messages)
+# except NotImplementedError:
+#     print("该服务商不支持对话能力")
+
+# # 图片生成
+# try:
+#     task = adapter.create_image_task(prompt="一只猫")
+#     status = adapter.fetch_image_task_status(task)
+#     result = adapter.fetch_image_result(task)
+# except NotImplementedError:
+#     print("该服务商不支持图片生成")