LangChain 學習筆記 : 串接 LLM 模型 (二)

晚上繼續測試 LangChain, 本篇旨在測試串接谷歌 Gemini 的方法.

本系列之前的文章參考 :

# LangChain 學習筆記 : 串接 LLM 模型 (一)

關於谷歌原生的 Gemini 串接套件 google-generativeai 用法參考 :

# Google Gemini API 學習筆記 (一) : 文字聊天測試

5. 安裝 langchain-google-genai 套件 :    

LangChain 用來串接 Gemini 的套件為 langchain-google-genai, 與前一篇串接 GPT 模型一樣, langchain-google-genai 套件只是封裝了透過谷歌原生套件 google-generativeai 串接 Gemini 模型的介面而已, 它無法單獨與 Gemini 聊天, 必須搭配 google-generativeai 當底層介面才行. 

不過, 由於 langchain-google-genai 與 google-generativeai 這兩個套件在安裝時會與相依套件 google-ai-generativelanguage 有版本衝突問題 (安裝時會出現 ERROR 訊息), 我向 ChatGPT 尋求解決方案, 經測試發現只要指定特定版本即可順利安裝, 若已安裝此兩套件出現錯誤, 請先用下列指令解除安裝 :

pip uninstall -y google-ai-generativelanguage google-generativeai langchain-google-genai  

D:\python\test>pip uninstall -y google-ai-generativelanguage google-generativeai langchain-google-genai

Found existing installation: google-ai-generativelanguage 0.7.0

Uninstalling google-ai-generativelanguage-0.7.0:

  Successfully uninstalled google-ai-generativelanguage-0.7.0

Found existing installation: google-generativeai 0.8.5

Uninstalling google-generativeai-0.8.5:

  Successfully uninstalled google-generativeai-0.8.5

Found existing installation: langchain-google-genai 2.1.12

Uninstalling langchain-google-genai-2.1.12:

  Successfully uninstalled langchain-google-genai-2.1.12

然後指定安裝 0.6.15 版的 google-ai-generativelanguage : 

pip install google-ai-generativelanguage==0.6.15    

D:\python\test>pip install google-ai-generativelanguage==0.6.15   

Collecting google-ai-generativelanguage==0.6.15

  Using cached google_ai_generativelanguage-0.6.15-py3-none-any.whl.metadata (5.7 kB)

Requirement already satisfied: google-api-core!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (2.19.2)

Requirement already satisfied: google-auth!=2.24.0,!=2.25.0,<3.0.0dev,>=2.14.1 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from google-ai-generativelanguage==0.6.15) (2.22.0)

Requirement already satisfied: proto-plus<2.0.0dev,>=1.22.3 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-ai-generativelanguage==0.6.15) (1.24.0)

Requirement already satisfied: protobuf!=4.21.0,!=4.21.1,!=4.21.2,!=4.21.3,!=4.21.4,!=4.21.5,<6.0.0dev,>=3.20.2 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-ai-generativelanguage==0.6.15) (5.28.1)

Requirement already satisfied: googleapis-common-protos<2.0.dev0,>=1.56.2 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-api-core!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (1.65.0)

Requirement already satisfied: requests<3.0.0.dev0,>=2.18.0 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from google-api-core!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (2.31.0)

Requirement already satisfied: grpcio<2.0dev,>=1.33.2 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (1.66.1)

Requirement already satisfied: grpcio-status<2.0.dev0,>=1.33.2 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (1.66.1)

Requirement already satisfied: cachetools<6.0,>=2.0.0 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from google-auth!=2.24.0,!=2.25.0,<3.0.0dev,>=2.14.1->google-ai-generativelanguage==0.6.15) (5.3.1)

Requirement already satisfied: pyasn1-modules>=0.2.1 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from google-auth!=2.24.0,!=2.25.0,<3.0.0dev,>=2.14.1->google-ai-generativelanguage==0.6.15) (0.3.0)

Requirement already satisfied: rsa<5,>=3.1.4 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from google-auth!=2.24.0,!=2.25.0,<3.0.0dev,>=2.14.1->google-ai-generativelanguage==0.6.15) (4.9)

Requirement already satisfied: six>=1.9.0 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-auth!=2.24.0,!=2.25.0,<3.0.0dev,>=2.14.1->google-ai-generativelanguage==0.6.15) (1.17.0)

Requirement already satisfied: urllib3<2.0 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from google-auth!=2.24.0,!=2.25.0,<3.0.0dev,>=2.14.1->google-ai-generativelanguage==0.6.15) (1.26.19)

Requirement already satisfied: charset-normalizer<4,>=2 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from requests<3.0.0.dev0,>=2.18.0->google-api-core!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (3.2.0)

Requirement already satisfied: idna<4,>=2.5 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from requests<3.0.0.dev0,>=2.18.0->google-api-core!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (3.4)

Requirement already satisfied: certifi>=2017.4.17 in c:\users\tony1\appdata\local\programs\thonny\lib\site-packages (from requests<3.0.0.dev0,>=2.18.0->google-api-core!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-api-core[grpc]!=2.0.*,!=2.1.*,!=2.10.*,!=2.2.*,!=2.3.*,!=2.4.*,!=2.5.*,!=2.6.*,!=2.7.*,!=2.8.*,!=2.9.*,<3.0.0dev,>=1.34.1->google-ai-generativelanguage==0.6.15) (2025.6.15)

Requirement already satisfied: pyasn1>=0.1.3 in c:\users\tony1\appdata\roaming\python\python310\site-packages (from rsa<5,>=3.1.4->google-auth!=2.24.0,!=2.25.0,<3.0.0dev,>=2.14.1->google-ai-generativelanguage==0.6.15) (0.5.0)

Using cached google_ai_generativelanguage-0.6.15-py3-none-any.whl (1.3 MB)

Installing collected packages: google-ai-generativelanguage

Successfully installed google-ai-generativelanguage-0.6.15

其次安裝 0.8.5 版的 google-generativeai : 

pip install google-generativeai==0.8.5 --no-deps

D:\python\test>pip install google-generativeai==0.8.5 --no-deps  

Collecting google-generativeai==0.8.5

  Using cached google_generativeai-0.8.5-py3-none-any.whl.metadata (3.9 kB)

Using cached google_generativeai-0.8.5-py3-none-any.whl (155 kB)

Installing collected packages: google-generativeai

Successfully installed google-generativeai-0.8.5

最後安裝 2.1.12 版的 langchain-google-genai : 

pip install langchain-google-genai==2.1.12 --no-deps

D:\python\test>pip install langchain-google-genai==2.1.12 --no-deps   

Collecting langchain-google-genai==2.1.12

  Using cached langchain_google_genai-2.1.12-py3-none-any.whl.metadata (7.1 kB)

Using cached langchain_google_genai-2.1.12-py3-none-any.whl (50 kB)

Installing collected packages: langchain-google-genai

Successfully installed langchain-google-genai-2.1.12

6. 檢視 langchain-google-genai 套件內容 :    

匯入 langchain_google_genai 後用 dir() 檢視其內容 : 

>>> import langchain_google_genai     

>>> dir(langchain_google_genai)   

['AqaInput', 'AqaOutput', 'ChatGoogleGenerativeAI', 'DoesNotExistsException', 'GenAIAqa', 'GoogleGenerativeAI', 'GoogleGenerativeAIEmbeddings', 'GoogleVectorStore', 'HarmBlockThreshold', 'HarmCategory', 'Modality', '__all__', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '_common', '_enums', '_function_utils', '_genai_extension', '_image_utils', 'chat_models', 'embeddings', 'genai_aqa', 'google_vector_store', 'llms']

其中的 ChatGoogleGenerativeAI 類別便是用來串接 Gemini 模型用的, 其建構式 ChatGoogleGenerativeAI() 的參數如下表所示 : 

參數名稱	說明
model	Gemini 模型名稱，例如 "gemini-1.5-pro"，對應 API 的 model 參數
temperature	生成隨機性，範圍 0~1，對應 API 的 temperature
max_output_tokens	最大生成 token 數量，對應 API 的 max_output_tokens
top_p	nucleus sampling 控制生成多樣性，對應 API 的 top_p
top_k	top-k sampling 控制生成多樣性，對應 API 的 top_k
api_key	Google API Key，可直接傳入或使用環境變數 GOOGLE_API_KEY
request_timeout	請求超時秒數，對應 SDK 的 timeout / request_timeout
model_kwargs	可傳遞額外參數給 Gemini SDK，例如 stop_sequences、candidate_count 等
verbose	布林值，是否印出 debug 訊息
client	可傳自訂 Gemini SDK client，一般不用

在互動環境測試如下 :

我已將 Gemini API Key 儲存在環境變數檔案 .env 裡面 (名稱為 'GEMINI_API_KEY'), 先用 dotenv 套件讀取出來放在 gemini_api_key 變數中 :

>>> from dotenv import dotenv_values    

config=dotenv_values('.env')     

gemini_api_key=config.get('GEMINI_API_KEY')    

從 langchain_google_genai 套件中匯入 ChatGoogleGenerativeAI 類別 :  

>>> from langchain_google_genai import ChatGoogleGenerativeAI   

然後呼叫 ChatGoogleGenerativeAI() 建構函式並傳入 api_key 與 model 參數來建立 ChatGoogleGenerativeAI 物件 : 

>>> chat_model=ChatGoogleGenerativeAI(api_key=gemini_api_key, model='gemini-2.5-flash')    

>>> type(chat_model)         

<class 'langchain_google_genai.chat_models.ChatGoogleGenerativeAI'>   

注意, 目前可用的 Gemini 模型如下 (之前測試使用的 1.5 版已下架, 無法串接了, 如果 model 指定舊版 1.5 版會得到 404 錯誤) : 

模型名稱	描述	適合場景	穩定性
gemini-2.5-flash	快速、低延遲的輕量模型	聊天、即時回應	穩定
gemini-2.5-pro	先進推理模型，支援長上下文	複雜問題、程式碼生成	穩定
gemini-2.5-flash-lite	更輕量的 Flash 變體	行動裝置、成本敏感應用	穩定
gemini-2.0-flash	舊 2.0 版本，快速但較舊	基本任務	穩定

這樣就可以呼叫 ChatGoogleGenerativeAI 物件的 invoke() 方法並傳入提示詞來起始一個對話 : 

>>> response=chat_model.invoke('你是誰?')     

>>> type(response)     

<class 'langchain_core.messages.ai.AIMessage'>     

檢視 invoke() 傳回的 AIMessage 物件內容 : 

>>> response   

AIMessage(content='我是一个大型语言模型，由 Google 训练。', additional_kwargs={}, response_metadata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name': 'gemini-2.5-flash', 'safety_ratings': []}, id='run--9c115a61-fc2d-4c74-a611-a7da5465b65b-0')

可見模型的回應是放在其 content 屬性內 : 

>>> response.content     

'我是一个大型语言模型，由 Google 训练。'

由這兩篇測試可知 LangChain 的好處是對於不同的模型都可以用相同的 API 與模型交談.