maritalk 0.0.3

Client library for the MariTalk API

Introduction

This repository contains the code and documentation explaining how to use the MariTalk API. MariTalk is a chatbot based on a language model that has been specially trained to understand Portuguese well. It is able to follow instructions in a zero-shot manner, like ChatGPT.

Usage Example

This Google Colab contains an example showing how to use the API, which is similar to ChatGPT's API:

Google Colab Example

You will also find in the colab an example of how to use the model in a few-shot manner.

You can find more details about the API at https://chat.maritaca.ai/docs

Technical Aspects

Is it free?

The API is available for free to all users. We plan to introduce a paid version in the future that will offer increased throughput and reliability.

Rate Limit

At the moment, there is a limit of 4 requests per minute to ensure that everyone has the opportunity to test the model. If your request is denied due to exceeding this limit, you will receive an HTTP 429 error.

Maximum sequence length

Currently, we support a maximum sequence length of 2048 tokens, which is equivalent to approximately 1000 words in Portuguese. We will soon support 8196 tokens.

Throughput

It takes about 1 to 2 seconds to generate the first token given a sequence of 1000 tokens as input. After that, new tokens are generated at a rate of 10 to 15 tokens/sec.

Web Interface

Try the Web interface at: chat.maritaca.ai