This package provides functionality developed to simplify interfacing with the Cohere API in Python 3.
We're nearing the release of our brand new cohere-python SDK! If you'd like to get a head start by installing the alpha version, please go ahead and use pip install --pre --upgrade cohere. Thanks for your patience as we make breaking changes where necessary. If you have any issues using it, we will respond to github issues as soon as possible.
See the migration guide for upgrading and the new README.md for more information and examples.
- SDK Documentation can be found here.
- You can build SDK documentation locally using
cd docs; make clean html.
- You can build SDK documentation locally using
- For more details on advanced parameters, you can also consult the API documentation.
- See the examples directory for examples, including some additional functionality for visualizations in Jupyter notebooks.
The package can be installed with pip:
pip install --upgrade cohereInstall from source:
pip install .- Python 3.8+
To use this library, you must have an API key and specify it as a string when creating the cohere.Client object. API keys can be created through the platform. This is a basic example of the creating the client and using the generate endpoint.
import cohere
# initialize the Cohere Client with an API Key
co = cohere.Client('YOUR_API_KEY')
# generate a prediction for a prompt
prediction = co.chat(message='Howdy! 🤠', model='command')
# print the predicted text
print(f'Chatbot: {prediction.text}')There is also an asyncio compatible client called cohere.AsyncClient with an equivalent interface. Consult the SDK Docs for more details.
Each SDK release is only compatible with the latest version of the Cohere API at the time of release. To use the SDK with an older API version, you need to download a version of the SDK tied to the API version you want. Look at the Changelog to see which SDK version to download.
For a full breakdown of endpoints and arguments, please consult the SDK Docs and Cohere Docs.
| Cohere Endpoint | Function |
|---|---|
| /chat | co.chat() |
| /generate | co.generate() |
| /embed | co.embed() |
| /rerank | co.rerank() |
| /classify | co.classify() |
| /tokenize | co.tokenize() |
| /detokenize | co.detokenize() |
| /detect-language | co.detect_language() |
When you call Cohere's APIs we decide on a good default model for your use-case behind the scenes. The default model is great to get you started, but in production environments we recommend that you specify the model size yourself via the model parameter. Learn more about the available models here(https://dashboard.cohere.com)
All of the endpoint functions will return a Cohere object corresponding to the endpoint (e.g. for generation, it would be Generation). The responses can be found as instance variables of the object (e.g. generation would be Generation.text). The names of these instance variables and a detailed breakdown of the response body can be found in the SDK Docs and Cohere Docs. Printing the Cohere response object itself will display an organized view of the instance variables.
Unsuccessful API calls from the SDK will raise an exception. Please see the documentation's page on errors for more information about what the errors mean.
To set up a development environment, first ensure you have poetry 1.7+ installed and run:
poetry shell # any time you want to run code or tests
poetry install # install and update dependencies in your environment, the first timeIn addition, to ensure your code is formatted correctly, install pre-commit hooks using:
pre-commit installYou can run tests locally using:
python -m pytestYou can configure a different base url with:
CO_API_URL="https://localhost:8050" python3 foo.pyor in code using one of:
cohere.COHERE_API_URL = "https://localhost:8050" # Place before client initilization
cohere.Client(...,api_url="https://localhost:8050") # Alternatively, as a parameter to the Client