Spaces:
Runtime error
Runtime error
Update main.py
Browse files
main.py
CHANGED
|
@@ -1,11 +1,19 @@
|
|
| 1 |
import os
|
| 2 |
import io
|
|
|
|
| 3 |
from collections import defaultdict
|
| 4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
from fastapi import FastAPI, Request, Header, BackgroundTasks, HTTPException
|
|
|
|
| 6 |
from fastapi.staticfiles import StaticFiles
|
|
|
|
| 7 |
from google import genai
|
| 8 |
from google.genai import types
|
|
|
|
| 9 |
from linebot import LineBotApi, WebhookHandler
|
| 10 |
from linebot.exceptions import InvalidSignatureError
|
| 11 |
from linebot.models import (
|
|
@@ -15,8 +23,6 @@ from linebot.models import (
|
|
| 15 |
ImageSendMessage,
|
| 16 |
ImageMessage,
|
| 17 |
)
|
| 18 |
-
import PIL.Image
|
| 19 |
-
import uvicorn
|
| 20 |
|
| 21 |
# LangChain 相關匯入
|
| 22 |
from langchain_core.prompts import ChatPromptTemplate
|
|
@@ -26,22 +32,38 @@ from langchain.agents import AgentExecutor, create_tool_calling_agent
|
|
| 26 |
|
| 27 |
|
| 28 |
# ==========================
|
| 29 |
-
#
|
| 30 |
# ==========================
|
| 31 |
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
|
|
|
|
|
|
|
|
|
| 35 |
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
line_handler = WebhookHandler(os.environ["CHANNEL_SECRET"])
|
| 39 |
|
| 40 |
-
#
|
| 41 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
|
| 43 |
# 建立 FastAPI 應用程式
|
| 44 |
app = FastAPI()
|
|
|
|
|
|
|
|
|
|
| 45 |
app.mount("/static", StaticFiles(directory="static"), name="static")
|
| 46 |
|
| 47 |
# 設定 CORS
|
|
@@ -53,12 +75,13 @@ app.add_middleware(
|
|
| 53 |
allow_headers=["*"],
|
| 54 |
)
|
| 55 |
|
| 56 |
-
def
|
| 57 |
"""
|
| 58 |
從 Line 訊息 ID 獲取圖片內容並儲存到暫存檔案。
|
| 59 |
"""
|
| 60 |
try:
|
| 61 |
message_content = line_bot_api.get_message_content(message_id)
|
|
|
|
| 62 |
file_path = f"/tmp/{message_id}.png"
|
| 63 |
with open(file_path, "wb") as f:
|
| 64 |
for chunk in message_content.iter_content():
|
|
@@ -66,25 +89,9 @@ def get_image_url_from_line(message_id):
|
|
| 66 |
print(f"✅ 圖片成功儲存到:{file_path}")
|
| 67 |
return file_path
|
| 68 |
except Exception as e:
|
| 69 |
-
print(f"❌
|
| 70 |
return None
|
| 71 |
|
| 72 |
-
def store_user_message(user_id, message_type, message_content):
|
| 73 |
-
"""
|
| 74 |
-
儲存用戶的訊息。
|
| 75 |
-
"""
|
| 76 |
-
user_message_history[user_id].append(
|
| 77 |
-
{"type": message_type, "content": message_content}
|
| 78 |
-
)
|
| 79 |
-
|
| 80 |
-
def get_previous_message(user_id):
|
| 81 |
-
"""
|
| 82 |
-
獲取用戶的上一則訊息。
|
| 83 |
-
"""
|
| 84 |
-
if user_id in user_message_history and len(user_message_history[user_id]) > 0:
|
| 85 |
-
return user_message_history[user_id][-1]
|
| 86 |
-
return {"type": "text", "content": "No message!"}
|
| 87 |
-
|
| 88 |
|
| 89 |
# ==========================
|
| 90 |
# LangChain 工具定義
|
|
@@ -99,33 +106,37 @@ def generate_and_upload_image(prompt: str) -> str:
|
|
| 99 |
prompt: 用於生成圖片的文字提示。
|
| 100 |
|
| 101 |
Returns:
|
| 102 |
-
|
| 103 |
"""
|
| 104 |
try:
|
| 105 |
-
|
| 106 |
-
|
|
|
|
| 107 |
contents=prompt,
|
| 108 |
-
|
| 109 |
)
|
| 110 |
|
| 111 |
image_binary = None
|
| 112 |
for part in response.candidates[0].content.parts:
|
| 113 |
-
if part.inline_data
|
| 114 |
image_binary = part.inline_data.data
|
| 115 |
break
|
| 116 |
|
| 117 |
if image_binary:
|
| 118 |
image = PIL.Image.open(io.BytesIO(image_binary))
|
| 119 |
# 隨機生成一個檔案名以避免衝突
|
| 120 |
-
file_name = f"
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
image_url = os.path.join(os.getenv("HF_SPACE"), file_name)
|
| 124 |
-
return image_url
|
| 125 |
|
| 126 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
except Exception as e:
|
| 128 |
-
return f"
|
| 129 |
|
| 130 |
@tool
|
| 131 |
def analyze_image_with_text(image_path: str, user_text: str) -> str:
|
|
@@ -144,19 +155,15 @@ def analyze_image_with_text(image_path: str, user_text: str) -> str:
|
|
| 144 |
return "圖片路徑無效,無法進行分析。"
|
| 145 |
|
| 146 |
img_user = PIL.Image.open(image_path)
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
out = response.text
|
| 153 |
else:
|
| 154 |
-
|
| 155 |
except Exception as e:
|
| 156 |
-
|
| 157 |
-
out = f"Gemini執行出錯: {e}"
|
| 158 |
-
|
| 159 |
-
return out
|
| 160 |
|
| 161 |
|
| 162 |
# ==========================
|
|
@@ -167,18 +174,22 @@ def analyze_image_with_text(image_path: str, user_text: str) -> str:
|
|
| 167 |
tools = [generate_and_upload_image, analyze_image_with_text]
|
| 168 |
|
| 169 |
# 建立 LLM 模型實例
|
| 170 |
-
llm = ChatGoogleGenerativeAI(
|
| 171 |
|
| 172 |
# 建立提示模板
|
| 173 |
-
|
| 174 |
-
("system", "
|
| 175 |
-
|
|
|
|
|
|
|
|
|
|
| 176 |
("user", "{input}"),
|
| 177 |
("placeholder", "{agent_scratchpad}"),
|
| 178 |
])
|
| 179 |
|
|
|
|
| 180 |
# 建立代理人
|
| 181 |
-
agent = create_tool_calling_agent(llm, tools,
|
| 182 |
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
|
| 183 |
|
| 184 |
# ==========================
|
|
@@ -187,16 +198,17 @@ agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
|
|
| 187 |
|
| 188 |
@app.get("/")
|
| 189 |
def root():
|
| 190 |
-
return {"
|
| 191 |
|
| 192 |
@app.post("/webhook")
|
| 193 |
async def webhook(
|
| 194 |
request: Request,
|
| 195 |
background_tasks: BackgroundTasks,
|
| 196 |
-
x_line_signature=Header(None),
|
| 197 |
):
|
| 198 |
body = await request.body()
|
| 199 |
try:
|
|
|
|
| 200 |
background_tasks.add_task(
|
| 201 |
line_handler.handle, body.decode("utf-8"), x_line_signature
|
| 202 |
)
|
|
@@ -207,56 +219,64 @@ async def webhook(
|
|
| 207 |
@line_handler.add(MessageEvent, message=(ImageMessage, TextMessage))
|
| 208 |
def handle_message(event):
|
| 209 |
user_id = event.source.user_id
|
| 210 |
-
|
| 211 |
-
#
|
| 212 |
-
if event.message
|
| 213 |
-
image_path =
|
| 214 |
if image_path:
|
| 215 |
-
|
| 216 |
-
|
| 217 |
-
|
| 218 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 219 |
else:
|
| 220 |
line_bot_api.reply_message(
|
| 221 |
-
event.reply_token, TextSendMessage(text="
|
| 222 |
)
|
| 223 |
-
|
| 224 |
-
#
|
| 225 |
-
elif event.message
|
| 226 |
user_text = event.message.text
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
|
| 230 |
-
# 根據上一則訊息類型,動態傳遞給代理人
|
| 231 |
-
if previous_message["type"] == "image":
|
| 232 |
-
image_path = previous_message["content"]
|
| 233 |
-
agent_input = {
|
| 234 |
-
"input": f"請根據這張圖片回答問題。圖片的路徑是 {image_path},我的問題是:{user_text}"
|
| 235 |
-
}
|
| 236 |
-
# 清除上一則圖片訊息,避免重複觸發
|
| 237 |
-
user_message_history[user_id].pop()
|
| 238 |
-
else:
|
| 239 |
-
agent_input = {"input": user_text}
|
| 240 |
try:
|
| 241 |
# 運行代理人
|
| 242 |
response = agent_executor.invoke(agent_input)
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 247 |
line_bot_api.push_message(
|
| 248 |
event.source.user_id,
|
| 249 |
[
|
| 250 |
-
TextSendMessage(text="✨
|
| 251 |
ImageSendMessage(original_content_url=image_url, preview_image_url=image_url)
|
| 252 |
]
|
| 253 |
)
|
| 254 |
else:
|
| 255 |
-
|
|
|
|
| 256 |
except Exception as e:
|
| 257 |
print(f"代理人執行出錯: {e}")
|
| 258 |
-
|
| 259 |
-
line_bot_api.reply_message(event.reply_token, TextSendMessage(text=
|
| 260 |
|
| 261 |
if __name__ == "__main__":
|
| 262 |
-
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
import io
|
| 3 |
+
import re
|
| 4 |
from collections import defaultdict
|
| 5 |
+
import PIL.Image
|
| 6 |
+
import uvicorn
|
| 7 |
+
import requests
|
| 8 |
+
from pydantic_settings import BaseSettings
|
| 9 |
+
|
| 10 |
from fastapi import FastAPI, Request, Header, BackgroundTasks, HTTPException
|
| 11 |
+
from fastapi.middleware.cors import CORSMiddleware
|
| 12 |
from fastapi.staticfiles import StaticFiles
|
| 13 |
+
|
| 14 |
from google import genai
|
| 15 |
from google.genai import types
|
| 16 |
+
|
| 17 |
from linebot import LineBotApi, WebhookHandler
|
| 18 |
from linebot.exceptions import InvalidSignatureError
|
| 19 |
from linebot.models import (
|
|
|
|
| 23 |
ImageSendMessage,
|
| 24 |
ImageMessage,
|
| 25 |
)
|
|
|
|
|
|
|
| 26 |
|
| 27 |
# LangChain 相關匯入
|
| 28 |
from langchain_core.prompts import ChatPromptTemplate
|
|
|
|
| 32 |
|
| 33 |
|
| 34 |
# ==========================
|
| 35 |
+
# 環境變數與設定管理
|
| 36 |
# ==========================
|
| 37 |
|
| 38 |
+
class Settings(BaseSettings):
|
| 39 |
+
"""使用 Pydantic 管理環境變數"""
|
| 40 |
+
google_api_key: str
|
| 41 |
+
channel_access_token: str
|
| 42 |
+
channel_secret: str
|
| 43 |
+
base_url: str # 應用程式的公開網址,例如 ngrok 或 Hugging Face Space 的 URL
|
| 44 |
|
| 45 |
+
class Config:
|
| 46 |
+
env_file = ".env"
|
|
|
|
| 47 |
|
| 48 |
+
# 載入設定
|
| 49 |
+
settings = Settings()
|
| 50 |
+
|
| 51 |
+
# ==========================
|
| 52 |
+
# API 客戶端與工具函式初始化
|
| 53 |
+
# ==========================
|
| 54 |
+
|
| 55 |
+
# 設置 Google AI API 金鑰
|
| 56 |
+
genai.configure(api_key=settings.google_api_key)
|
| 57 |
+
|
| 58 |
+
# 設置 Line Bot API
|
| 59 |
+
line_bot_api = LineBotApi(settings.channel_access_token)
|
| 60 |
+
line_handler = WebhookHandler(settings.channel_secret)
|
| 61 |
|
| 62 |
# 建立 FastAPI 應用程式
|
| 63 |
app = FastAPI()
|
| 64 |
+
|
| 65 |
+
# 確保靜態檔案目錄存在
|
| 66 |
+
os.makedirs("static", exist_ok=True)
|
| 67 |
app.mount("/static", StaticFiles(directory="static"), name="static")
|
| 68 |
|
| 69 |
# 設定 CORS
|
|
|
|
| 75 |
allow_headers=["*"],
|
| 76 |
)
|
| 77 |
|
| 78 |
+
def get_image_from_line(message_id: str) -> str | None:
|
| 79 |
"""
|
| 80 |
從 Line 訊息 ID 獲取圖片內容並儲存到暫存檔案。
|
| 81 |
"""
|
| 82 |
try:
|
| 83 |
message_content = line_bot_api.get_message_content(message_id)
|
| 84 |
+
# 使用 /tmp 目錄儲存暫存檔案,適合多數雲端環境
|
| 85 |
file_path = f"/tmp/{message_id}.png"
|
| 86 |
with open(file_path, "wb") as f:
|
| 87 |
for chunk in message_content.iter_content():
|
|
|
|
| 89 |
print(f"✅ 圖片成功儲存到:{file_path}")
|
| 90 |
return file_path
|
| 91 |
except Exception as e:
|
| 92 |
+
print(f"❌ 從 Line 取得圖片失敗:{e}")
|
| 93 |
return None
|
| 94 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
# ==========================
|
| 97 |
# LangChain 工具定義
|
|
|
|
| 106 |
prompt: 用於生成圖片的文字提示。
|
| 107 |
|
| 108 |
Returns:
|
| 109 |
+
回傳生成圖片的公開 URL。
|
| 110 |
"""
|
| 111 |
try:
|
| 112 |
+
# 使用 gemini-2.5-flash-image-preview 模型進行圖片生成
|
| 113 |
+
model = genai.GenerativeModel('gemini-2.5-flash-image-preview')
|
| 114 |
+
response = model.generate_content(
|
| 115 |
contents=prompt,
|
| 116 |
+
generation_config=types.GenerationConfig(response_modalities=['IMAGE'])
|
| 117 |
)
|
| 118 |
|
| 119 |
image_binary = None
|
| 120 |
for part in response.candidates[0].content.parts:
|
| 121 |
+
if part.inline_data and part.inline_data.data:
|
| 122 |
image_binary = part.inline_data.data
|
| 123 |
break
|
| 124 |
|
| 125 |
if image_binary:
|
| 126 |
image = PIL.Image.open(io.BytesIO(image_binary))
|
| 127 |
# 隨機生成一個檔案名以避免衝突
|
| 128 |
+
file_name = f"{os.urandom(16).hex()}.png"
|
| 129 |
+
file_path = os.path.join("static", file_name)
|
| 130 |
+
image.save(file_path, format="PNG")
|
|
|
|
|
|
|
| 131 |
|
| 132 |
+
# 使用環境變數中的 BASE_URL 來建立完整的公開網址
|
| 133 |
+
image_url = f"{settings.base_url}/{file_path}"
|
| 134 |
+
print(f"✅ 圖片生成成功,URL: {image_url}")
|
| 135 |
+
return f"圖片生成成功,請查看此 URL: {image_url}"
|
| 136 |
+
|
| 137 |
+
return "圖片生成失敗,未收到有效的圖片資料。"
|
| 138 |
except Exception as e:
|
| 139 |
+
return f"圖片生成與上傳過程中發生錯誤: {e}"
|
| 140 |
|
| 141 |
@tool
|
| 142 |
def analyze_image_with_text(image_path: str, user_text: str) -> str:
|
|
|
|
| 155 |
return "圖片路徑無效,無法進行分析。"
|
| 156 |
|
| 157 |
img_user = PIL.Image.open(image_path)
|
| 158 |
+
model = genai.GenerativeModel("gemini-1.5-flash") # 使用 gemini 1.5 flash 模型
|
| 159 |
+
response = model.generate_content([img_user, user_text])
|
| 160 |
+
|
| 161 |
+
if response.text:
|
| 162 |
+
return response.text
|
|
|
|
| 163 |
else:
|
| 164 |
+
return "Gemini 沒有給出答案,請嘗試換個方式提問!"
|
| 165 |
except Exception as e:
|
| 166 |
+
return f"圖片分析過程中發生錯誤: {e}"
|
|
|
|
|
|
|
|
|
|
| 167 |
|
| 168 |
|
| 169 |
# ==========================
|
|
|
|
| 174 |
tools = [generate_and_upload_image, analyze_image_with_text]
|
| 175 |
|
| 176 |
# 建立 LLM 模型實例
|
| 177 |
+
llm = ChatGoogleGenerativeAI(model="gemini-1.5-flash", temperature=0.3)
|
| 178 |
|
| 179 |
# 建立提示模板
|
| 180 |
+
prompt = ChatPromptTemplate.from_messages([
|
| 181 |
+
("system", """你是一個強大的圖像生成與問答助理。
|
| 182 |
+
- 當用戶的指令明顯是要生成圖片時 (例如:'畫一張...'、'生成...'、'幫我做一張圖...'),請使用 `generate_and_upload_image` 工具。
|
| 183 |
+
- 當用戶的指令包含圖片路徑 (image_path) 和問題時,請使用 `analyze_image_with_text` 工具。
|
| 184 |
+
- 成功執行 `generate_and_upload_image` 工具後,你會獲得一個 URL,你的最終回答必須包含這個 URL。
|
| 185 |
+
- 如果工具執行過程中產生任何錯誤訊息,請以友善的方式解讀並回應給用戶。"""),
|
| 186 |
("user", "{input}"),
|
| 187 |
("placeholder", "{agent_scratchpad}"),
|
| 188 |
])
|
| 189 |
|
| 190 |
+
|
| 191 |
# 建立代理人
|
| 192 |
+
agent = create_tool_calling_agent(llm, tools, prompt)
|
| 193 |
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
|
| 194 |
|
| 195 |
# ==========================
|
|
|
|
| 198 |
|
| 199 |
@app.get("/")
|
| 200 |
def root():
|
| 201 |
+
return {"message": "Line Bot is running!"}
|
| 202 |
|
| 203 |
@app.post("/webhook")
|
| 204 |
async def webhook(
|
| 205 |
request: Request,
|
| 206 |
background_tasks: BackgroundTasks,
|
| 207 |
+
x_line_signature: str = Header(None),
|
| 208 |
):
|
| 209 |
body = await request.body()
|
| 210 |
try:
|
| 211 |
+
# 使用背景任務處理 Webhook,避免 Line Server 超時
|
| 212 |
background_tasks.add_task(
|
| 213 |
line_handler.handle, body.decode("utf-8"), x_line_signature
|
| 214 |
)
|
|
|
|
| 219 |
@line_handler.add(MessageEvent, message=(ImageMessage, TextMessage))
|
| 220 |
def handle_message(event):
|
| 221 |
user_id = event.source.user_id
|
| 222 |
+
|
| 223 |
+
# 處理圖片上傳:使用者上傳圖片後直接進行分析
|
| 224 |
+
if isinstance(event.message, ImageMessage):
|
| 225 |
+
image_path = get_image_from_line(event.message.id)
|
| 226 |
if image_path:
|
| 227 |
+
try:
|
| 228 |
+
# 組合給代理人的輸入,使用一個通用的問題來分析圖片
|
| 229 |
+
agent_input = {
|
| 230 |
+
"input": f"這是一張使用者上傳的圖片,請詳細描述你看到了什麼。圖片的路徑是 '{image_path}'。"
|
| 231 |
+
}
|
| 232 |
+
# 運行代理人
|
| 233 |
+
response = agent_executor.invoke(agent_input)
|
| 234 |
+
output_text = response["output"]
|
| 235 |
+
|
| 236 |
+
# 回覆分析結果
|
| 237 |
+
line_bot_api.reply_message(event.reply_token, TextSendMessage(text=output_text))
|
| 238 |
+
|
| 239 |
+
except Exception as e:
|
| 240 |
+
print(f"代理人執行出錯: {e}")
|
| 241 |
+
error_message = f"圖片分析時發生錯誤,請稍後再試。\n錯誤訊息:{e}"
|
| 242 |
+
line_bot_api.reply_message(event.reply_token, TextSendMessage(text=error_message))
|
| 243 |
else:
|
| 244 |
line_bot_api.reply_message(
|
| 245 |
+
event.reply_token, TextSendMessage(text="❌ 圖片接收失敗,請再試一次。")
|
| 246 |
)
|
| 247 |
+
|
| 248 |
+
# 處理文字訊息:主要用於生成圖片或一般問答
|
| 249 |
+
elif isinstance(event.message, TextMessage):
|
| 250 |
user_text = event.message.text
|
| 251 |
+
agent_input = {"input": user_text}
|
| 252 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 253 |
try:
|
| 254 |
# 運行代理人
|
| 255 |
response = agent_executor.invoke(agent_input)
|
| 256 |
+
output_text = response["output"]
|
| 257 |
+
|
| 258 |
+
# 使用正規表示法尋找 URL,更穩定
|
| 259 |
+
image_url_match = re.search(r'https?://\S+\.(?:png|jpg|jpeg|gif)', output_text, re.IGNORECASE)
|
| 260 |
+
|
| 261 |
+
if image_url_match:
|
| 262 |
+
image_url = image_url_match.group(0)
|
| 263 |
+
# 推送訊息,包含生成的圖片
|
| 264 |
line_bot_api.push_message(
|
| 265 |
event.source.user_id,
|
| 266 |
[
|
| 267 |
+
TextSendMessage(text="✨ 這是我為您生成的圖片喔~"),
|
| 268 |
ImageSendMessage(original_content_url=image_url, preview_image_url=image_url)
|
| 269 |
]
|
| 270 |
)
|
| 271 |
else:
|
| 272 |
+
# 若無圖片 URL,則直接回覆文字
|
| 273 |
+
line_bot_api.reply_message(event.reply_token, TextSendMessage(text=output_text))
|
| 274 |
except Exception as e:
|
| 275 |
print(f"代理人執行出錯: {e}")
|
| 276 |
+
error_message = f"代理人執行時發生錯誤,請稍後再試。\n錯誤訊息:{e}"
|
| 277 |
+
line_bot_api.reply_message(event.reply_token, TextSendMessage(text=error_message))
|
| 278 |
|
| 279 |
if __name__ == "__main__":
|
| 280 |
+
# 使用 settings 物件中的設定
|
| 281 |
+
uvicorn.run("main:app", host="0.0.0.0", port=7860, reload=True)
|
| 282 |
+
|