ChatCerebras

本筆記本提供 Cerebras 聊天模型的快速入門概述。如需 ChatCerebras 所有功能和配置的詳細文件，請參閱 API 參考文件。

在 Cerebras，我們開發了世界最大、最快的 AI 處理器 Wafer-Scale Engine-3 (WSE-3)。Cerebras CS-3 系統由 WSE-3 驅動，代表了新一代 AI 超級電腦，它以前所未有的效能和可擴展性，為生成式 AI 訓練和推論樹立了標準。

選擇 Cerebras 作為您的推論供應商，您可以：

• 針對 AI 推論工作負載實現前所未有的速度
• 以高吞吐量進行商業建構
• 透過我們無縫的叢集技術輕鬆擴展您的 AI 工作負載

我們的 CS-3 系統可以快速輕鬆地叢集，以創建世界上最大的 AI 超級電腦，從而簡化最大模型的部署和運行。領先的企業、研究機構和政府已經在使用 Cerebras 解決方案來開發專有模型並訓練流行的開源模型。

想體驗 Cerebras 的強大功能嗎？請查看我們的網站以獲取更多資源，並探索透過 Cerebras Cloud 或本地部署來存取我們技術的選項！

有關 Cerebras Cloud 的更多資訊，請訪問 cloud.cerebras.ai。我們的 API 參考文件可在 inference-docs.cerebras.ai 查閱。

概述

整合細節

類別	套件	本地	可序列化	JS 支援	套件下載次數	最新套件
ChatCerebras	langchain-cerebras	❌	測試版	❌

模型功能

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

設定

pip install langchain-cerebras

憑證

從 cloud.cerebras.ai 取得 API 金鑰，並將其新增至您的環境變數中

export CEREBRAS_API_KEY="your-api-key-here"

import getpass
import os

if "CEREBRAS_API_KEY" not in os.environ:
    os.environ["CEREBRAS_API_KEY"] = getpass.getpass("Enter your Cerebras API key: ")

Enter your Cerebras API key:  ········

如果您想要取得模型呼叫的自動追蹤，您也可以透過取消註解下方內容來設定您的 LangSmith API 金鑰

# os.environ["LANGSMITH_API_KEY"] = getpass.getpass("Enter your LangSmith API key: ")
# os.environ["LANGSMITH_TRACING"] = "true"

安裝

LangChain Cerebras 整合位於 langchain-cerebras 套件中

%pip install -qU langchain-cerebras

實例化

現在我們可以實例化我們的模型物件並產生聊天完成

from langchain_cerebras import ChatCerebras

llm = ChatCerebras(
    model="llama-3.3-70b",
    # other params...
)

API 參考文件：ChatCerebras

調用

messages = [
    (
        "system",
        "You are a helpful assistant that translates English to French. Translate the user sentence.",
    ),
    ("human", "I love programming."),
]
ai_msg = llm.invoke(messages)
ai_msg

AIMessage(content='Je adore le programmation.', response_metadata={'token_usage': {'completion_tokens': 7, 'prompt_tokens': 35, 'total_tokens': 42}, 'model_name': 'llama3-8b-8192', 'system_fingerprint': 'fp_be27ec77ff', 'finish_reason': 'stop'}, id='run-e5d66faf-019c-4ac6-9265-71093b13202d-0', usage_metadata={'input_tokens': 35, 'output_tokens': 7, 'total_tokens': 42})

串鏈

我們可以像這樣使用提示範本串鏈我們的模型

from langchain_cerebras import ChatCerebras
from langchain_core.prompts import ChatPromptTemplate

llm = ChatCerebras(
    model="llama-3.3-70b",
    # other params...
)

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

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

API 參考文件：ChatCerebras | ChatPromptTemplate

AIMessage(content='Ich liebe Programmieren!\n\n(Literally: I love programming!)', response_metadata={'token_usage': {'completion_tokens': 14, 'prompt_tokens': 30, 'total_tokens': 44}, 'model_name': 'llama3-8b-8192', 'system_fingerprint': 'fp_be27ec77ff', 'finish_reason': 'stop'}, id='run-e1d2ebb8-76d1-471b-9368-3b68d431f16a-0', usage_metadata={'input_tokens': 30, 'output_tokens': 14, 'total_tokens': 44})

串流

from langchain_cerebras import ChatCerebras
from langchain_core.prompts import ChatPromptTemplate

llm = ChatCerebras(
    model="llama-3.3-70b",
    # other params...
)

system = "You are an expert on animals who must answer questions in a manner that a 5 year old can understand."
human = "I want to learn more about this animal: {animal}"
prompt = ChatPromptTemplate.from_messages([("system", system), ("human", human)])

chain = prompt | llm

for chunk in chain.stream({"animal": "Lion"}):
    print(chunk.content, end="", flush=True)

API 參考文件：ChatCerebras | ChatPromptTemplate

OH BOY! Let me tell you all about LIONS!

Lions are the kings of the jungle! They're really big and have beautiful, fluffy manes around their necks. The mane is like a big, golden crown!

Lions live in groups called prides. A pride is like a big family, and the lionesses (that's what we call the female lions) take care of the babies. The lionesses are like the mommies, and they teach the babies how to hunt and play.

Lions are very good at hunting. They work together to catch their food, like zebras and antelopes. They're super fast and can run really, really fast!

But lions are also very sleepy. They like to take long naps in the sun, and they can sleep for up to 20 hours a day! Can you imagine sleeping that much?

Lions are also very loud. They roar really loudly to talk to each other. It's like they're saying, "ROAR! I'm the king of the jungle!"

And guess what? Lions are very social. They like to play and cuddle with each other. They're like big, furry teddy bears!

So, that's lions! Aren't they just the coolest?

非同步

from langchain_cerebras import ChatCerebras
from langchain_core.prompts import ChatPromptTemplate

llm = ChatCerebras(
    model="llama-3.3-70b",
    # other params...
)

prompt = ChatPromptTemplate.from_messages(
    [
        (
            "human",
            "Let's play a game of opposites. What's the opposite of {topic}? Just give me the answer with no extra input.",
        )
    ]
)
chain = prompt | llm
await chain.ainvoke({"topic": "fire"})

API 參考文件：ChatCerebras | ChatPromptTemplate

AIMessage(content='Ice', response_metadata={'token_usage': {'completion_tokens': 2, 'prompt_tokens': 36, 'total_tokens': 38}, 'model_name': 'llama3-8b-8192', 'system_fingerprint': 'fp_be27ec77ff', 'finish_reason': 'stop'}, id='run-7434bdde-1bec-44cf-827b-8d978071dfe8-0', usage_metadata={'input_tokens': 36, 'output_tokens': 2, 'total_tokens': 38})

非同步串流

from langchain_cerebras import ChatCerebras
from langchain_core.prompts import ChatPromptTemplate

llm = ChatCerebras(
    model="llama-3.3-70b",
    # other params...
)

prompt = ChatPromptTemplate.from_messages(
    [
        (
            "human",
            "Write a long convoluted story about {subject}. I want {num_paragraphs} paragraphs.",
        )
    ]
)
chain = prompt | llm

async for chunk in chain.astream({"num_paragraphs": 3, "subject": "blackholes"}):
    print(chunk.content, end="", flush=True)

API 參考文件：ChatCerebras | ChatPromptTemplate

In the distant reaches of the cosmos, there existed a peculiar phenomenon known as the "Eclipse of Eternity," a swirling vortex of darkness that had been shrouded in mystery for eons. It was said that this blackhole, born from the cataclysmic collision of two ancient stars, had been slowly devouring the fabric of space-time itself, warping the very essence of reality. As the celestial bodies of the galaxy danced around it, they began to notice a strange, almost imperceptible distortion in the fabric of space, as if the blackhole's gravitational pull was exerting an influence on the very course of events itself.

As the centuries passed, astronomers from across the galaxy became increasingly fascinated by the Eclipse of Eternity, pouring over ancient texts and scouring the cosmos for any hint of its secrets. One such scholar, a brilliant and reclusive astrophysicist named Dr. Elara Vex, became obsessed with unraveling the mysteries of the blackhole. She spent years pouring over ancient texts, deciphering cryptic messages and hidden codes that hinted at the existence of a long-lost civilization that had once thrived in the heart of the blackhole itself. According to legend, this ancient civilization had possessed knowledge of the cosmos that was beyond human comprehension, and had used their mastery of the universe to create the Eclipse of Eternity as a gateway to other dimensions.

As Dr. Vex delved deeper into her research, she began to experience strange and vivid dreams, visions that seemed to transport her to the very heart of the blackhole itself. In these dreams, she saw ancient beings, their faces twisted in agony as they were consumed by the void. She saw stars and galaxies, their light warped and distorted by the blackhole's gravitational pull. And she saw the Eclipse of Eternity itself, its swirling vortex of darkness pulsing with an otherworldly energy that seemed to be calling to her. As the dreams grew more vivid and more frequent, Dr. Vex became convinced that she was being drawn into the heart of the blackhole, and that the secrets of the universe lay waiting for her on the other side.

API 參考文件

如需 ChatCerebras 所有功能和配置的詳細文件，請參閱 API 參考文件：https://langchain-python.dev.org.tw/api_reference/cerebras/chat_models/langchain_cerebras.chat_models.ChatCerebras.html#

相關內容

• 聊天模型概念指南
• 聊天模型操作指南