Perplexity 的 Sonar API 提供了一种将实时、有依据的网络搜索与高级推理和深度研究能力相结合的解决方案。

何时使用：

• 当你的应用需要直接来自网络的及时、相关数据时，例如动态内容更新或当前事件跟踪。
• 对于需要支持带有集成推理和深度研究的复杂用户查询的产品，例如数字助手或高级搜索引擎。

在开始之前，请确保安装 llama_index

%pip install llama-index-llms-perplexity

!pip install llama-index

初始设置

截至 2025 年 4 月 12 日，LLaMa Index 中的 Perplexity LLM 类支持以下模型：

模型	上下文长度	模型类型
sonar-deep-research	128k	聊天补全
sonar-reasoning-pro	128k	聊天补全
sonar-reasoning	128k	聊天补全
sonar-pro	200k	聊天补全
sonar	128k	聊天补全
r1-1776	128k	聊天补全

• sonar-pro 的最大输出 token 限制为 8k。
• 推理模型会输出思维链响应。
• r1-1776 是一个离线聊天模型，不使用 Perplexity 搜索子系统。

你可以在此处找到最新支持的模型。
速率限制见此处。
定价见此处。

import getpass
import os

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

from llama_index.llms.perplexity import Perplexity

PPLX_API_KEY = __import__("os").environ.get("PPLX_API_KEY")

llm = Perplexity(api_key=PPLX_API_KEY, model="sonar-pro", temperature=0.2)

# Import the ChatMessage class from the llama_index library.
from llama_index.core.llms import ChatMessage

# Create a list of dictionaries where each dictionary represents a chat message.
# Each dictionary contains a 'role' key (e.g., system or user) and a 'content' key with the corresponding message.
messages_dict = [
    {"role": "system", "content": "Be precise and concise."},
    {
        "role": "user",
        "content": "Tell me the latest news about the US Stock Market.",
    },
]

# Convert each dictionary in the list to a ChatMessage object using unpacking (**msg) in a list comprehension.
messages = [ChatMessage(**msg) for msg in messages_dict]

# Print the list of ChatMessage objects to verify the conversion.
print(messages)

[ChatMessage(role=<MessageRole.SYSTEM: 'system'>, additional_kwargs={}, blocks=[TextBlock(block_type='text', text='Be precise and concise.')]), ChatMessage(role=<MessageRole.USER: 'user'>, additional_kwargs={}, blocks=[TextBlock(block_type='text', text='Tell me the latest news about the US Stock Market.')])]

聊天

response = llm.chat(messages)
print(response)

assistant: The latest update on the U.S. stock market indicates a strong performance recently. A significant 10% rally occurred on Wednesday, which contributed substantially to market gains. Additionally, the market closed strongly on Friday, with a 2% increase, ending near the intraday high. This reflects robust momentum, particularly in mega and large-cap growth stocks[1].

异步聊天

对于异步对话处理，使用 achat 方法发送消息并等待响应：

# Asynchronously send the list of chat messages to the LLM using the 'achat' method.
# This method returns a ChatResponse object containing the model's answer.
response = await llm.achat(messages)

print(response)

assistant: The U.S. stock market has recently experienced significant gains. A major rally on Wednesday resulted in a 10% surge, contributing substantially to the market's overall upside. Additionally, the market closed strongly on Friday, with a 2% increase, ending near the intraday high. This performance highlights robust momentum, particularly in mega-cap and large-cap growth stocks[1].

流式聊天

对于需要实时逐 token 接收响应的场景，使用 stream_chat 方法：

# Call the stream_chat method on the LLM instance, which returns a generator or iterable
# for streaming the chat response one delta (token or chunk) at a time.
response = llm.stream_chat(messages)

# Iterate over each streaming response chunk.
for r in response:
    # Print the delta (the new chunk of generated text) without adding a newline.
    print(r.delta, end="")

The latest news about the U.S. stock market indicates a strong performance recently. The New York Stock Exchange (NYSE) experienced a significant rally, with a 10% surge on Wednesday, followed by a 2% gain on Friday. This upward momentum brought the market near its intraday high, driven by strength in mega-cap and large-cap growth stocks[1].

异步流式聊天

类似地，对于异步流式输出，astream_chat 方法提供了一种异步处理响应增量的方式：

# Asynchronously call the astream_chat method on the LLM instance,
# which returns an asynchronous generator that yields response chunks.
resp = await llm.astream_chat(messages)

# Asynchronously iterate over each response chunk from the generator.
# For each chunk (delta), print the chunk's text content.
async for delta in resp:
    print(delta.delta, end="")

The latest updates on the U.S. stock market indicate significant positive momentum. The New York Stock Exchange (NYSE) experienced a strong rally, with a notable 10% surge on Wednesday. This was followed by a 2% gain on Friday, closing near the intraday high. The market's performance has been driven by mega and large-cap growth stocks, contributing to the overall upside[1].

工具调用

Perplexity 模型可以轻松封装为 llamaindex 工具，以便在数据处理或对话工作流中调用。该工具使用由 Perplexity 提供支持的实时生成式搜索，并配置了更新后的默认模型（“sonar-pro”）并启用了 enable_search_classifier 参数。

以下是定义并注册该工具的示例：

from llama_index.core.tools import FunctionTool
from llama_index.llms.perplexity import Perplexity
from llama_index.core.llms import ChatMessage

def query_perplexity(query: str) -> str:
    """
    Queries the Perplexity API via the LlamaIndex integration.

    This function instantiates a Perplexity LLM with updated default settings
    (using model "sonar-pro" and enabling search classifier so that the API can
    intelligently decide if a search is needed), wraps the query into a ChatMessage,
    and returns the generated response content.
    """
    pplx_api_key = (
        "your-perplexity-api-key"  # Replace with your actual API key
    )

    llm = Perplexity(
        api_key=pplx_api_key,
        model="sonar-pro",
        temperature=0.7,
        enable_search_classifier=True,  # This will determine if the search component is necessary in this particular context
    )

    messages = [ChatMessage(role="user", content=query)]
    response = llm.chat(messages)
    return response.message.content

# Create the tool from the query_perplexity function
query_perplexity_tool = FunctionTool.from_defaults(fn=query_perplexity)