ChatReka

此筆記本提供 Reka 聊天模型 入門的快速概觀。

Reka 有多種聊天模型。您可以在 Reka 文件中找到關於其最新模型及其成本、上下文視窗和支援的輸入類型等資訊。

概觀

整合詳細資訊

類別	套件	本地	可序列化	JS 支援	套件下載次數	套件最新版本
[ChatReka]	langchain_community	✅	❌	❌

模型功能

工具呼叫	結構化輸出	JSON 模式	影像輸入	音訊輸入	視訊輸入	Token 層級串流	原生非同步	Token 使用量	Logprobs
✅	❌	❌	✅	✅	✅	✅	✅	❌	❌

設定

若要存取 Reka 模型，您需要建立一個 Reka 開發人員帳戶、取得 API 金鑰，並透過 'pip install reka-api' 安裝 langchain_community 整合套件和 reka python 套件。

憑證

前往 https://platform.reka.ai/ 註冊 Reka 並產生 API 金鑰。完成後，設定 REKA_API_KEY 環境變數

安裝

LangChain ModuleName 整合位於 langchain_community 套件中

%pip install -qU langchain_community reka-api

Note: you may need to restart the kernel to use updated packages.

實例化

import getpass
import os

os.environ["REKA_API_KEY"] = getpass.getpass("Enter your Reka API key: ")

選用：使用 Langsmith 追蹤模型的執行

import getpass
import os

os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_API_KEY"] = getpass.getpass("Enter your Langsmith API key: ")

from langchain_community.chat_models import ChatReka

model = ChatReka()

API 參考：ChatReka

調用

model.invoke("hi")

AIMessage(content=' Hello! How can I help you today? If you have a question, need assistance, or just want to chat, feel free to let me know. Have a great day!\n\n', additional_kwargs={}, response_metadata={}, id='run-61522ec2-0587-4fd5-a492-5b205fd8860c-0')

影像輸入

from langchain_core.messages import HumanMessage

image_url = "https://v0.docs.reka.ai/_images/000000245576.jpg"

message = HumanMessage(
    content=[
        {"type": "text", "text": "describe the weather in this image"},
        {
            "type": "image_url",
            "image_url": {"url": image_url},
        },
    ],
)
response = model.invoke([message])
print(response.content)

API 參考：HumanMessage

 The image shows an indoor setting with no visible windows or natural light, and there are no indicators of weather conditions. The focus is on a cat sitting on a computer keyboard, and the background includes a computer monitor and various office supplies.

多個影像作為輸入

message = HumanMessage(
    content=[
        {"type": "text", "text": "What are the difference between the two images? "},
        {
            "type": "image_url",
            "image_url": {
                "url": "https://cdn.pixabay.com/photo/2019/07/23/13/51/shepherd-dog-4357790_1280.jpg"
            },
        },
        {
            "type": "image_url",
            "image_url": {
                "url": "https://cdn.pixabay.com/photo/2024/02/17/00/18/cat-8578562_1280.jpg"
            },
        },
    ],
)
response = model.invoke([message])
print(response.content)

 The first image features two German Shepherds, one adult and one puppy, in a vibrant, lush green setting. The adult dog is carrying a large stick in its mouth, running through what appears to be a grassy field, with the puppy following close behind. Both dogs exhibit striking physical characteristics typical of the breed, such as pointed ears and dense fur.

The second image shows a close-up of a single cat with striking blue eyes, likely a breed like the Siberian or Maine Coon, in a natural outdoor setting. The cat's fur is lighter, possibly a mix of white and gray, and it has a more subdued expression compared to the dogs. The background is blurred, suggesting a focus on the cat's face.

Overall, the differences lie in the subjects (two dogs vs. one cat), the setting (lush, vibrant grassy field vs. a more muted outdoor background), and the overall mood and activity depicted (playful and active vs. serene and focused).

鏈結

from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate(
    [
        (
            "system",
            "You are a helpful assistant that translates {input_language} to {output_language}.",
        ),
        ("human", "{input}"),
    ]
)

chain = prompt | model
chain.invoke(
    {
        "input_language": "English",
        "output_language": "German",
        "input": "I love programming.",
    }
)

API 參考：ChatPromptTemplate

AIMessage(content=' Ich liebe Programmieren.\n\n', additional_kwargs={}, response_metadata={}, id='run-ffc4ace1-b73a-4fb3-ad0f-57e60a0f9b8d-0')

搭配 tavtly api 搜尋使用

工具使用和代理程式建立

定義工具

我們首先需要建立想要使用的工具。我們主要選擇的工具將是 Tavily - 一個搜尋引擎。我們在 LangChain 中有一個內建工具，可以輕鬆地將 Tavily 搜尋引擎用作工具。

import getpass
import os

os.environ["TAVILY_API_KEY"] = getpass.getpass("Enter your Tavily API key: ")

from langchain_community.tools.tavily_search import TavilySearchResults

search = TavilySearchResults(max_results=2)
search_results = search.invoke("what is the weather in SF")
print(search_results)
# If we want, we can create other tools.
# Once we have all the tools we want, we can put them in a list that we will reference later.
tools = [search]

API 參考：TavilySearchResults

[{'url': 'https://www.weatherapi.com/', 'content': "{'location': {'name': 'San Francisco', 'region': 'California', 'country': 'United States of America', 'lat': 37.775, 'lon': -122.4183, 'tz_id': 'America/Los_Angeles', 'localtime_epoch': 1730484342, 'localtime': '2024-11-01 11:05'}, 'current': {'last_updated_epoch': 1730484000, 'last_updated': '2024-11-01 11:00', 'temp_c': 11.1, 'temp_f': 52.0, 'is_day': 1, 'condition': {'text': 'Mist', 'icon': '//cdn.weatherapi.com/weather/64x64/day/143.png', 'code': 1030}, 'wind_mph': 2.9, 'wind_kph': 4.7, 'wind_degree': 247, 'wind_dir': 'WSW', 'pressure_mb': 1019.0, 'pressure_in': 30.08, 'precip_mm': 0.0, 'precip_in': 0.0, 'humidity': 100, 'cloud': 100, 'feelslike_c': 11.1, 'feelslike_f': 52.0, 'windchill_c': 10.3, 'windchill_f': 50.5, 'heatindex_c': 10.8, 'heatindex_f': 51.5, 'dewpoint_c': 10.4, 'dewpoint_f': 50.6, 'vis_km': 2.8, 'vis_miles': 1.0, 'uv': 3.0, 'gust_mph': 3.8, 'gust_kph': 6.1}}"}, {'url': 'https://weatherspark.com/h/m/557/2024/1/Historical-Weather-in-January-2024-in-San-Francisco-California-United-States', 'content': 'San Francisco Temperature History January 2024\nHourly Temperature in January 2024 in San Francisco\nCompare San Francisco to another city:\nCloud Cover in January 2024 in San Francisco\nDaily Precipitation in January 2024 in San Francisco\nObserved Weather in January 2024 in San Francisco\nHours of Daylight and Twilight in January 2024 in San Francisco\nSunrise & Sunset with Twilight in January 2024 in San Francisco\nSolar Elevation and Azimuth in January 2024 in San Francisco\nMoon Rise, Set & Phases in January 2024 in San Francisco\nHumidity Comfort Levels in January 2024 in San Francisco\nWind Speed in January 2024 in San Francisco\nHourly Wind Speed in January 2024 in San Francisco\nHourly Wind Direction in 2024 in San Francisco\nAtmospheric Pressure in January 2024 in San Francisco\nData Sources\n See all nearby weather stations\nLatest Report — 1:56 PM\nFri, Jan 12, 2024\xa0\xa0\xa0\xa04 min ago\xa0\xa0\xa0\xa0UTC 21:56\nCall Sign KSFO\nTemp.\n54.0°F\nPrecipitation\nNo Report\nWind\n8.1 mph\nCloud Cover\nMostly Cloudy\n14,000 ft\nRaw: KSFO 122156Z 08007KT 10SM FEW030 SCT050 BKN140 12/07 A3022 While having the tremendous advantages of temporal and spatial completeness, these reconstructions: (1) are based on computer models that may have model-based errors, (2) are coarsely sampled on a 50 km grid and are therefore unable to reconstruct the local variations of many microclimates, and (3) have particular difficulty with the weather in some coastal areas, especially small islands.\n We further caution that our travel scores are only as good as the data that underpin them, that weather conditions at any given location and time are unpredictable and variable, and that the definition of the scores reflects a particular set of preferences that may not agree with those of any particular reader.\n January 2024 Weather History in San Francisco California, United States\nThe data for this report comes from the San Francisco International Airport.'}]

我們現在可以看到啟用此模型進行工具呼叫會是什麼樣子。為了啟用該功能，我們使用 .bind_tools 為語言模型提供這些工具的知識

model_with_tools = model.bind_tools(tools)

我們現在可以呼叫模型。我們先用一般訊息呼叫它，看看它如何回應。我們可以查看 content 欄位以及 tool_calls 欄位。

from langchain_core.messages import HumanMessage

response = model_with_tools.invoke([HumanMessage(content="Hi!")])

print(f"ContentString: {response.content}")
print(f"ToolCalls: {response.tool_calls}")

API 參考：HumanMessage

ContentString:  Hello! How can I help you today? If you have a question or need information on a specific topic, feel free to ask. Just type your search query and I'll do my best to assist using the available function.


ToolCalls: []

現在，讓我們嘗試使用一些預期會呼叫工具的輸入來呼叫它。

response = model_with_tools.invoke([HumanMessage(content="What's the weather in SF?")])

print(f"ContentString: {response.content}")
print(f"ToolCalls: {response.tool_calls}")

ContentString: 
ToolCalls: [{'name': 'tavily_search_results_json', 'args': {'query': 'weather in SF'}, 'id': '2548c622-3553-42df-8220-39fde0632bdb', 'type': 'tool_call'}]

我們可以看到現在沒有文字內容，但是有一個工具呼叫！它希望我們呼叫 Tavily 搜尋工具。

這還沒有呼叫該工具 - 它只是告訴我們要呼叫。為了實際呼叫它，我們會想要建立我們的代理程式。

建立代理程式

現在我們已經定義了工具和 LLM，我們可以建立代理程式。我們將使用 LangGraph 來建構代理程式。目前，我們正在使用高階介面來建構代理程式，但 LangGraph 的優點在於，這個高階介面是由低階、高度可控的 API 支援，以防您想要修改代理程式邏輯。

現在，我們可以使用 LLM 和工具初始化代理程式。

請注意，我們傳入的是 model，而不是 model_with_tools。那是因為 create_react_agent 會在底層為我們呼叫 .bind_tools。

from langgraph.prebuilt import create_react_agent

agent_executor = create_react_agent(model, tools)

API 參考：create_react_agent

現在讓我們在一個應該調用工具的範例中試試看

response = agent_executor.invoke({"messages": [HumanMessage(content="hi!")]})

response["messages"]

[HumanMessage(content='hi!', additional_kwargs={}, response_metadata={}, id='0ab1f3c7-9079-42d4-8a8a-13af5f6c226b'),
 AIMessage(content=' Hello! How can I help you today? If you have a question or need information on a specific topic, feel free to ask. For example, you can start with a search query like "latest news on climate change" or "biography of Albert Einstein".\n\n', additional_kwargs={}, response_metadata={}, id='run-276d9dcd-13f3-481d-b562-8fe3962d9ba1-0')]

為了確切了解底層發生的情況（並確保它沒有呼叫工具），我們可以查看 LangSmith 追蹤：https://smith.langchain.com/public/2372d9c5-855a-45ee-80f2-94b63493563d/r

response = agent_executor.invoke(
    {"messages": [HumanMessage(content="whats the weather in sf?")]}
)
response["messages"]

[HumanMessage(content='whats the weather in sf?', additional_kwargs={}, response_metadata={}, id='af276c61-3df7-4241-8cb0-81d1f1477bb3'),
 AIMessage(content='', additional_kwargs={'tool_calls': [{'id': '86da84b8-0d44-444f-8448-7f134f9afa41', 'type': 'function', 'function': {'name': 'tavily_search_results_json', 'arguments': '{"query": "weather in SF"}'}}]}, response_metadata={}, id='run-abe1b8e2-98a6-4f69-8f95-278ac8c141ff-0', tool_calls=[{'name': 'tavily_search_results_json', 'args': {'query': 'weather in SF'}, 'id': '86da84b8-0d44-444f-8448-7f134f9afa41', 'type': 'tool_call'}]),
 ToolMessage(content='[{"url": "https://www.weatherapi.com/", "content": "{\'location\': {\'name\': \'San Francisco\', \'region\': \'California\', \'country\': \'United States of America\', \'lat\': 37.775, \'lon\': -122.4183, \'tz_id\': \'America/Los_Angeles\', \'localtime_epoch\': 1730483436, \'localtime\': \'2024-11-01 10:50\'}, \'current\': {\'last_updated_epoch\': 1730483100, \'last_updated\': \'2024-11-01 10:45\', \'temp_c\': 11.4, \'temp_f\': 52.5, \'is_day\': 1, \'condition\': {\'text\': \'Mist\', \'icon\': \'//cdn.weatherapi.com/weather/64x64/day/143.png\', \'code\': 1030}, \'wind_mph\': 2.2, \'wind_kph\': 3.6, \'wind_degree\': 237, \'wind_dir\': \'WSW\', \'pressure_mb\': 1019.0, \'pressure_in\': 30.08, \'precip_mm\': 0.0, \'precip_in\': 0.0, \'humidity\': 100, \'cloud\': 100, \'feelslike_c\': 11.8, \'feelslike_f\': 53.2, \'windchill_c\': 11.2, \'windchill_f\': 52.1, \'heatindex_c\': 11.7, \'heatindex_f\': 53.0, \'dewpoint_c\': 10.1, \'dewpoint_f\': 50.1, \'vis_km\': 2.8, \'vis_miles\': 1.0, \'uv\': 3.0, \'gust_mph\': 3.0, \'gust_kph\': 4.9}}"}, {"url": "https://www.timeanddate.com/weather/@z-us-94134/ext", "content": "Forecasted weather conditions the coming 2 weeks for San Francisco. Sign in. News. News Home; Astronomy News; Time Zone News ... 01 pm: Mon Nov 11: 60 / 53 °F: Tstorms early. Broken clouds. 54 °F: 19 mph: ↑: 70%: 58%: 0.20\\" 0 (Low) 6:46 am: 5:00 pm * Updated Monday, October 28, 2024 2:24:10 pm San Francisco time - Weather by CustomWeather"}]', name='tavily_search_results_json', id='de8c8d78-ae24-4a8a-9c73-795c1e4fdd41', tool_call_id='86da84b8-0d44-444f-8448-7f134f9afa41', artifact={'query': 'weather in SF', 'follow_up_questions': None, 'answer': None, 'images': [], 'results': [{'title': 'Weather in San Francisco', 'url': 'https://www.weatherapi.com/', 'content': "{'location': {'name': 'San Francisco', 'region': 'California', 'country': 'United States of America', 'lat': 37.775, 'lon': -122.4183, 'tz_id': 'America/Los_Angeles', 'localtime_epoch': 1730483436, 'localtime': '2024-11-01 10:50'}, 'current': {'last_updated_epoch': 1730483100, 'last_updated': '2024-11-01 10:45', 'temp_c': 11.4, 'temp_f': 52.5, 'is_day': 1, 'condition': {'text': 'Mist', 'icon': '//cdn.weatherapi.com/weather/64x64/day/143.png', 'code': 1030}, 'wind_mph': 2.2, 'wind_kph': 3.6, 'wind_degree': 237, 'wind_dir': 'WSW', 'pressure_mb': 1019.0, 'pressure_in': 30.08, 'precip_mm': 0.0, 'precip_in': 0.0, 'humidity': 100, 'cloud': 100, 'feelslike_c': 11.8, 'feelslike_f': 53.2, 'windchill_c': 11.2, 'windchill_f': 52.1, 'heatindex_c': 11.7, 'heatindex_f': 53.0, 'dewpoint_c': 10.1, 'dewpoint_f': 50.1, 'vis_km': 2.8, 'vis_miles': 1.0, 'uv': 3.0, 'gust_mph': 3.0, 'gust_kph': 4.9}}", 'score': 0.9989501, 'raw_content': None}, {'title': 'San Francisco, USA 14 day weather forecast - timeanddate.com', 'url': 'https://www.timeanddate.com/weather/@z-us-94134/ext', 'content': 'Forecasted weather conditions the coming 2 weeks for San Francisco. Sign in. News. News Home; Astronomy News; Time Zone News ... 01 pm: Mon Nov 11: 60 / 53 °F: Tstorms early. Broken clouds. 54 °F: 19 mph: ↑: 70%: 58%: 0.20" 0 (Low) 6:46 am: 5:00 pm * Updated Monday, October 28, 2024 2:24:10 pm San Francisco time - Weather by CustomWeather', 'score': 0.9938309, 'raw_content': None}], 'response_time': 3.56}),
 AIMessage(content=' The current weather in San Francisco is mist with a temperature of 11.4°C (52.5°F). There is a 100% humidity and the wind is blowing at 2.2 mph from the WSW direction. The forecast for the coming weeks shows a mix of cloudy and partly cloudy days with some chances of thunderstorms. Temperatures are expected to range between 53°F and 60°F.\n\n', additional_kwargs={}, response_metadata={}, id='run-de4207d6-e8e8-4382-ad16-4de0dcf0812a-0')]

我們可以查看 LangSmith 追蹤，以確保它有效地呼叫了搜尋工具。

https://smith.langchain.com/public/013ef704-654b-4447-8428-637b343d646e/r

我們已經了解如何使用 .invoke 呼叫代理程式以取得最終回應。如果代理程式執行多個步驟，這可能需要一段時間。為了顯示中間進度，我們可以將訊息在發生時串流回傳。

for chunk in agent_executor.stream(
    {"messages": [HumanMessage(content="whats the weather in sf?")]}
):
    print(chunk)
    print("----")

{'agent': {'messages': [AIMessage(content='', additional_kwargs={'tool_calls': [{'id': '2457d3ea-f001-4b8c-a1ed-3dc3d1381639', 'type': 'function', 'function': {'name': 'tavily_search_results_json', 'arguments': '{"query": "weather in San Francisco"}'}}]}, response_metadata={}, id='run-0363deab-84d2-4319-bb1e-b55b47fe2274-0', tool_calls=[{'name': 'tavily_search_results_json', 'args': {'query': 'weather in San Francisco'}, 'id': '2457d3ea-f001-4b8c-a1ed-3dc3d1381639', 'type': 'tool_call'}])]}}
----
{'tools': {'messages': [ToolMessage(content='[{"url": "https://www.weatherapi.com/", "content": "{\'location\': {\'name\': \'San Francisco\', \'region\': \'California\', \'country\': \'United States of America\', \'lat\': 37.775, \'lon\': -122.4183, \'tz_id\': \'America/Los_Angeles\', \'localtime_epoch\': 1730483636, \'localtime\': \'2024-11-01 10:53\'}, \'current\': {\'last_updated_epoch\': 1730483100, \'last_updated\': \'2024-11-01 10:45\', \'temp_c\': 11.4, \'temp_f\': 52.5, \'is_day\': 1, \'condition\': {\'text\': \'Mist\', \'icon\': \'//cdn.weatherapi.com/weather/64x64/day/143.png\', \'code\': 1030}, \'wind_mph\': 2.2, \'wind_kph\': 3.6, \'wind_degree\': 237, \'wind_dir\': \'WSW\', \'pressure_mb\': 1019.0, \'pressure_in\': 30.08, \'precip_mm\': 0.0, \'precip_in\': 0.0, \'humidity\': 100, \'cloud\': 100, \'feelslike_c\': 11.8, \'feelslike_f\': 53.2, \'windchill_c\': 11.2, \'windchill_f\': 52.1, \'heatindex_c\': 11.7, \'heatindex_f\': 53.0, \'dewpoint_c\': 10.1, \'dewpoint_f\': 50.1, \'vis_km\': 2.8, \'vis_miles\': 1.0, \'uv\': 3.0, \'gust_mph\': 3.0, \'gust_kph\': 4.9}}"}, {"url": "https://weather.com/weather/monthly/l/69bedc6a5b6e977993fb3e5344e3c06d8bc36a1fb6754c3ddfb5310a3c6d6c87", "content": "Weather.com brings you the most accurate monthly weather forecast for San Francisco, CA with average/record and high/low temperatures, precipitation and more. ... 11. 66 ° 55 ° 12. 69 ° 60"}]', name='tavily_search_results_json', id='e675f99b-130f-4e98-8477-badd45938d9d', tool_call_id='2457d3ea-f001-4b8c-a1ed-3dc3d1381639', artifact={'query': 'weather in San Francisco', 'follow_up_questions': None, 'answer': None, 'images': [], 'results': [{'title': 'Weather in San Francisco', 'url': 'https://www.weatherapi.com/', 'content': "{'location': {'name': 'San Francisco', 'region': 'California', 'country': 'United States of America', 'lat': 37.775, 'lon': -122.4183, 'tz_id': 'America/Los_Angeles', 'localtime_epoch': 1730483636, 'localtime': '2024-11-01 10:53'}, 'current': {'last_updated_epoch': 1730483100, 'last_updated': '2024-11-01 10:45', 'temp_c': 11.4, 'temp_f': 52.5, 'is_day': 1, 'condition': {'text': 'Mist', 'icon': '//cdn.weatherapi.com/weather/64x64/day/143.png', 'code': 1030}, 'wind_mph': 2.2, 'wind_kph': 3.6, 'wind_degree': 237, 'wind_dir': 'WSW', 'pressure_mb': 1019.0, 'pressure_in': 30.08, 'precip_mm': 0.0, 'precip_in': 0.0, 'humidity': 100, 'cloud': 100, 'feelslike_c': 11.8, 'feelslike_f': 53.2, 'windchill_c': 11.2, 'windchill_f': 52.1, 'heatindex_c': 11.7, 'heatindex_f': 53.0, 'dewpoint_c': 10.1, 'dewpoint_f': 50.1, 'vis_km': 2.8, 'vis_miles': 1.0, 'uv': 3.0, 'gust_mph': 3.0, 'gust_kph': 4.9}}", 'score': 0.9968992, 'raw_content': None}, {'title': 'Monthly Weather Forecast for San Francisco, CA - weather.com', 'url': 'https://weather.com/weather/monthly/l/69bedc6a5b6e977993fb3e5344e3c06d8bc36a1fb6754c3ddfb5310a3c6d6c87', 'content': 'Weather.com brings you the most accurate monthly weather forecast for San Francisco, CA with average/record and high/low temperatures, precipitation and more. ... 11. 66 ° 55 ° 12. 69 ° 60', 'score': 0.97644573, 'raw_content': None}], 'response_time': 3.16})]}}
----
{'agent': {'messages': [AIMessage(content=' The current weather in San Francisco is misty with a temperature of 11.4°C (52.5°F). The wind is blowing at 2.2 mph (3.6 kph) from the WSW direction. The humidity is at 100%, and the visibility is 2.8 km (1.0 miles). The monthly forecast shows average temperatures ranging from 55°F to 66°F (13°C to 19°C) with some precipitation expected.\n\n', additional_kwargs={}, response_metadata={}, id='run-99ccf444-d286-4244-a5a5-7b1b511153a6-0')]}}
----

API 參考

https://docs.reka.ai/quick-start

相關連結

• 聊天模型 概念指南
• 聊天模型 操作指南