Maritalk¶

简介¶

MariTalk 是由巴西公司 Maritaca AI 开发的智能助手。该工具基于经过专门训练的语言模型，能够出色理解葡萄牙语。

本笔记本通过两个示例演示如何在 Llama Index 中使用 MariTalk：

1. 通过聊天方法获取宠物名字建议；
2. 通过完整方法结合少量示例对电影评论进行负面/正面分类。

安装¶

如果您在 Colab 上打开此 Notebook，很可能需要安装 LlamaIndex。

In [ ]:

!pip install llama-index
!pip install llama-index-llms-maritalk
!pip install asyncio

!pip install llama-index !pip install llama-index-llms-maritalk !pip install asyncio

API 密钥¶

您需要从 chat.maritaca.ai 获取 API 密钥（在"Chaves da API" 部分）。

示例 1 - 使用聊天功能获取宠物名字建议¶

In [ ]:

from llama_index.core.llms import ChatMessage
from llama_index.llms.maritalk import Maritalk

import asyncio

# To customize your API key, do this
# otherwise it will lookup MARITALK_API_KEY from your env variable
llm = Maritalk(api_key="<your_maritalk_api_key>", model="sabia-2-medium")

# Call chat with a list of messages
messages = [
    ChatMessage(
        role="system",
        content="You are an assistant specialized in suggesting pet names. Given the animal, you must suggest 4 names.",
    ),
    ChatMessage(role="user", content="I have a dog."),
]

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


# Async chat
async def get_dog_name(llm, messages):
    response = await llm.achat(messages)
    print(response)


asyncio.run(get_dog_name(llm, messages))

from llama_index.core.llms import ChatMessage from llama_index.llms.maritalk import Maritalk import asyncio # To customize your API key, do this # otherwise it will lookup MARITALK_API_KEY from your env variable llm = Maritalk(api_key="", model="sabia-2-medium") # Call chat with a list of messages messages = [ ChatMessage( role="system", content="You are an assistant specialized in suggesting pet names. Given the animal, you must suggest 4 names.", ), ChatMessage(role="user", content="I have a dog."), ] # Sync chat response = llm.chat(messages) print(response) # Async chat async def get_dog_name(llm, messages): response = await llm.achat(messages) print(response) asyncio.run(get_dog_name(llm, messages))

流式生成¶

对于涉及生成长文本的任务（例如撰写长篇论文或翻译大型文档），采用逐部分接收响应的方式（即在文本生成过程中逐步获取）比等待完整文本生成更具优势。这种方式能显著提升应用程序的响应速度与运行效率，特别是在处理大规模文本生成时。我们提供两种实现方案：同步式与异步式。

In [ ]:

# Sync streaming chat
response = llm.stream_chat(messages)
for chunk in response:
    print(chunk.delta, end="", flush=True)


# Async streaming chat
async def get_dog_name_streaming(llm, messages):
    async for chunk in await llm.astream_chat(messages):
        print(chunk.delta, end="", flush=True)


asyncio.run(get_dog_name_streaming(llm, messages))

# Sync streaming chat response = llm.stream_chat(messages) for chunk in response: print(chunk.delta, end="", flush=True) # Async streaming chat async def get_dog_name_streaming(llm, messages): async for chunk in await llm.astream_chat(messages): print(chunk.delta, end="", flush=True) asyncio.run(get_dog_name_streaming(llm, messages))

示例 2 - 使用 Complete 方法的少样本示例¶

我们建议在使用少样本示例时采用 llm.complete() 方法

In [ ]:

prompt = """Classifique a resenha de filme como "positiva" ou "negativa".

Resenha: Gostei muito do filme, é o melhor do ano!
Classe: positiva

Resenha: O filme deixa muito a desejar.
Classe: negativa

Resenha: Apesar de longo, valeu o ingresso..
Classe:"""

# Sync complete
response = llm.complete(prompt)
print(response)


# Async complete
async def classify_review(llm, prompt):
    response = await llm.acomplete(prompt)
    print(response)


asyncio.run(classify_review(llm, prompt))

prompt = """Classifique a resenha de filme como "positiva" ou "negativa". Resenha: Gostei muito do filme, é o melhor do ano! Classe: positiva Resenha: O filme deixa muito a desejar. Classe: negativa Resenha: Apesar de longo, valeu o ingresso.. Classe:""" # Sync complete response = llm.complete(prompt) print(response) # Async complete async def classify_review(llm, prompt): response = await llm.acomplete(prompt) print(response) asyncio.run(classify_review(llm, prompt))

In [ ]:

# Sync streaming complete
response = llm.stream_complete(prompt)
for chunk in response:
    print(chunk.delta, end="", flush=True)


# Async streaming complete
async def classify_review_streaming(llm, prompt):
    async for chunk in await llm.astream_complete(prompt):
        print(chunk.delta, end="", flush=True)


asyncio.run(classify_review_streaming(llm, prompt))

# Sync streaming complete response = llm.stream_complete(prompt) for chunk in response: print(chunk.delta, end="", flush=True) # Async streaming complete async def classify_review_streaming(llm, prompt): async for chunk in await llm.astream_complete(prompt): print(chunk.delta, end="", flush=True) asyncio.run(classify_review_streaming(llm, prompt))