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 especially trained to understand Portuguese well. It is able to follow instructions in a zero-shot manner, like ChatGPT.
This Google Colab contains an example showing how to use the API, which is similar to ChatGPT's API:
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
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.
At the moment, there is a limit of one request every 5 seconds 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.
Currently, we support a maximum sequence length of 2000 tokens, which is equivalent to approximately 1000 words in Portuguese. We will soon support 8000 tokens.
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.
Try the Web interface at: chat.maritaca.ai
- Instale o Python 3
- Execute o comando
pip install -r requirements.txt
ou clique no script install-requirements.bat para instalar as dependências - Execute o comando
python maritalk.py
ou clique no script run.bat