已经可以在单一环境中运行tts与cli

2025-12-21 04:39:58 +08:00
parent 09140e4000
commit f8c2a36a2f
4 changed files with 107 additions and 56 deletions
--- a/cli.py
+++ b/cli.py
@@ -6,22 +6,22 @@ from llama_index.llms.ollama import Ollama
 from llama_index.core.chat_engine import SimpleChatEngine
 from llama_index.core.chat_engine.types import StreamingAgentChatResponse
 from llama_index.core.llms import ChatMessage
-try:
-    from llama_index.core.llms.types import ImageBlock, TextBlock
-except ImportError:
-    try:
-        # 尝试其他可能的导入路径
-        from llama_index.core import ImageBlock, TextBlock
-    except ImportError:
-        # 如果都失败，定义简单的占位类
-        class ImageBlock:
-            def __init__(self, base64_str=None, path=None):
-                self.base64_str = base64_str
-                self.path = path
-        
-        class TextBlock:
-            def __init__(self, text=""):
-                self.text = text
+# try:
+#     from llama_index.core.llms.types import ImageBlock, TextBlock
+# except ImportError:
+#     try:
+#         # 尝试其他可能的导入路径
+#         from llama_index.core import ImageBlock, TextBlock
+#     except ImportError:
+#         # 如果都失败，定义简单的占位类
+#         class ImageBlock:
+#             def __init__(self, base64_str=None, path=None):
+#                 self.base64_str = base64_str
+#                 self.path = path
+#        
+#         class TextBlock:
+#             def __init__(self, text=""):
+#                 self.text = text
 from llama_index.core import Settings
 from Convention.Runtime.File import ToolFile
 import requests
@@ -358,7 +358,7 @@ def image_file_to_base64(image_path: str) -> str:
            PrintColorful(ConsoleFrontColor.LIGHTRED_EX, f"读取图片文件失败 {image_path}: {e}")
        return None

-async def _ollama_stream_chat_with_image(llm: Ollama, messages: list, image_base64: str, end_symbol: list) -> None:
+async def _ollama_stream_chat_with_image(llm: Ollama, messages: list, image_base64: str, end_symbol: list) -> str:
    """
    直接调用 Ollama API 进行带图片的流式聊天
    
@@ -382,11 +382,15 @@ async def _ollama_stream_chat_with_image(llm: Ollama, messages: list, image_base
            "role": msg.role if hasattr(msg, 'role') else "user",
            "content": msg.content if hasattr(msg, 'content') else str(msg)
        }
-        # 如果是第一条用户消息且有图片，添加图片
-        if (hasattr(msg, 'role') and msg.role == "user") and image_base64 and len(api_messages) == 0:
-            api_msg["images"] = [image_base64]
        api_messages.append(api_msg)
    
+    # 如果有图片，添加到最后一条用户消息（当前用户消息）
+    if image_base64:
+        for i in range(len(api_messages) - 1, -1, -1):
+            if api_messages[i].get("role") == "user":
+                api_messages[i]["images"] = [image_base64]
+                break
+    
    payload = {
        "model": model,
        "messages": api_messages,
@@ -441,6 +445,8 @@ async def _ollama_stream_chat_with_image(llm: Ollama, messages: list, image_base
        if len(buffer_response) > 0:
            if TTS_ENABLE:
                await play_vocal(buffer_response)
+        
+        return buffer_response
    except Exception as e:
        PrintColorful(ConsoleFrontColor.LIGHTRED_EX, f"Ollama API 调用错误: {e}")
        if VERBOSE:
@@ -467,7 +473,7 @@ def capture_screenshot() -> str:
            PrintColorful(ConsoleFrontColor.LIGHTRED_EX, f"截图失败: {e}")
        return None

-async def achat(llm: Ollama, message: str, image_base64: Optional[str] = None, auto_screenshot: bool = False) -> None:
+async def achat(llm: Ollama, message: str, image_base64: Optional[str] = None, auto_screenshot: bool = False, conversation_history: Optional[list] = None) -> list:
    """
    使用 Ollama LLM 进行多模态聊天
    
@@ -501,16 +507,22 @@ async def achat(llm: Ollama, message: str, image_base64: Optional[str] = None, a
    # 图片将通过 Ollama 的底层 API 传递
    chat_message = ChatMessage(role="user", content=user_message)
    
-    # 构建消息列表
-    messages = [chat_message]
+    # 构建消息列表 - 使用对话历史（如果提供）
+    if conversation_history is None:
+        conversation_history = []
    
-    # 如果有系统提示，添加到消息列表开头
-    if SYSTEM_PROMPT_PATH is not None:
+    # 构建完整的消息列表（包含历史记录）
+    messages = conversation_history.copy()
+    
+    # 如果对话历史为空，且需要系统提示，添加 system 消息（只添加一次）
+    if len(messages) == 0 and SYSTEM_PROMPT_PATH is not None:
        system_prompt = ToolFile(SYSTEM_PROMPT_PATH).LoadAsText()
        if system_prompt:
-            # 将系统提示添加到用户消息中，因为 Ollama 可能不支持 system role
-            user_message = f"{system_prompt}\n\n{user_message}"
-            messages[0] = ChatMessage(role="user", content=user_message)
+            system_msg = ChatMessage(role="system", content=system_prompt)
+            messages.append(system_msg)
+    
+    # 添加当前用户消息
+    messages.append(chat_message)
    
    buffer_response = ""
    end_symbol = ['。', '？', '！']
@@ -519,33 +531,41 @@ async def achat(llm: Ollama, message: str, image_base64: Optional[str] = None, a
        # 如果有图片，需要直接调用 Ollama API，因为 llama-index 的封装可能不支持图片
        if image_base64:
            # 直接调用 Ollama 的流式 API
-            await _ollama_stream_chat_with_image(llm, messages, image_base64, end_symbol)
-            return
+            assistant_response = await _ollama_stream_chat_with_image(llm, messages, image_base64, end_symbol)
+        else:
+            # 使用流式聊天（无图片时）
+            streaming_response = await llm.astream_chat(messages)
+            
+            # 实时输出流式文本
+            async for chunk in streaming_response.async_response_gen():
+                await asyncio.sleep(0.01)
+                print(chunk, end='', flush=True)
+                for ch in chunk:
+                    buffer_response += ch
+                    if len(buffer_response) > 20:
+                        if ch in end_symbol:
+                            if TTS_ENABLE:
+                                await play_vocal(buffer_response.strip())
+                            buffer_response = ""
+            
+            assistant_response = buffer_response.strip()
+            if len(assistant_response) > 0:
+                if TTS_ENABLE:
+                    await play_vocal(assistant_response)
        
-        # 使用流式聊天（无图片时）
-        streaming_response = await llm.astream_chat(messages)
+        # 更新对话历史：添加用户消息和助手响应
+        updated_history = messages.copy()
+        if assistant_response:
+            assistant_msg = ChatMessage(role="assistant", content=assistant_response)
+            updated_history.append(assistant_msg)
        
-        # 实时输出流式文本
-        async for chunk in streaming_response.async_response_gen():
-            await asyncio.sleep(0.01)
-            print(chunk, end='', flush=True)
-            for ch in chunk:
-                buffer_response += ch
-                if len(buffer_response) > 20:
-                    if ch in end_symbol:
-                        if TTS_ENABLE:
-                            await play_vocal(buffer_response.strip())
-                        buffer_response = ""
-        
-        buffer_response = buffer_response.strip()
-        if len(buffer_response) > 0:
-            if TTS_ENABLE:
-                await play_vocal(buffer_response)
+        return updated_history
    except Exception as e:
        PrintColorful(ConsoleFrontColor.LIGHTRED_EX, f"聊天错误: {e}")
        if VERBOSE:
            import traceback
            traceback.print_exc()
+        return conversation_history if conversation_history else []


 def add_speaker() -> None:
@@ -588,6 +608,9 @@ async def event_loop(llm: Ollama) -> None:
    PrintColorful(ConsoleFrontColor.LIGHTYELLOW_EX, "支持的图片格式: .png, .jpg, .jpeg")
    PrintColorful(ConsoleFrontColor.LIGHTYELLOW_EX, "输入 'quit' 或 'exit' 退出\n")
    
+    # 维护对话历史，避免重复发送系统提示词
+    conversation_history = []
+    
    message = input("请开始对话: ")
    wait_second = AUTO_SPEAK_WAIT_SECOND
    try:
@@ -614,19 +637,19 @@ async def event_loop(llm: Ollama) -> None:
                    PrintColorful(ConsoleFrontColor.LIGHTGREEN_EX, "截图成功")
            
            PrintColorful(ConsoleFrontColor.GREEN, "AI: ", is_reset=False, end='')
-            await achat(llm, message, image_base64)
+            conversation_history = await achat(llm, message, image_base64, False, conversation_history)
            PrintColorful(ConsoleFrontColor.RESET,"")
            
            # 等待用户输入
            message = await ainput(wait_second)
            if not message:
                # 用户没有输入，触发 AI 自主发言（会自动截图）
-                wait_second = max(wait_second*1.5, 3600)
+                wait_second = min(wait_second*1.5, 3600)
                if VERBOSE:
                    PrintColorful(ConsoleFrontColor.LIGHTYELLOW_EX, f"用户无输入，等待 {wait_second} 秒后 AI 自主发言...")
                # 触发 AI 自主发言（会自动截图）
                PrintColorful(ConsoleFrontColor.GREEN, "AI: ", is_reset=False, end='')
-                await achat(llm, "", None, auto_screenshot=True)
+                conversation_history = await achat(llm, "", None, auto_screenshot=True, conversation_history=conversation_history)
                PrintColorful(ConsoleFrontColor.RESET,"")
            else:
                wait_second = AUTO_SPEAK_WAIT_SECOND