Last April, NLP Cloud, a Grenoble-based start-up, launched its NLP programming interfaceallowing companies and their developers and/or data scientists to use NLP in their projects. The French start-up announced the integration of the open source GPT-J model in their NLP tool. Recently published by a team of researchers from EleutherAI, this model is intended to be equivalent to the famous GPT-3 of Open AI.

An innovative NLP start-up and an open source NLP model that competes with GPT-3

Founded in 2020 by Julien Salinas, the French start-up NLP Cloud, based in Grenoble, provides high-performance automatic language processing (NLP or TAL in French) models based on spaCy and Hugging Face Transformers, meeting production requirements. Last April, this new player in the NLP market launched its own NLP programming interface for data scientists and developers.

In parallel, EleutherAI’s teams (especially Aran Komatsuzaki, a PhD student in machine learning, and Ben Wang, an electrical engineering student at the University of Pensylavania) are developing GPT-J, an open source automatic language processing model trained with 6 billion parameters, which is 4 times larger than GPT-2 but almost 30 times smaller than Open AI’s GPT-3 model. Nevertheless, EleutherAI claims that its NLP model is one of the most advanced and would be equivalent to GPT-3.

The arrival of GPT-J in NLP Cloud: between ethical issues and interface performance

If the NLPcloud interface already supported GPT-3, the start-up announces that it will also support GPT-J. The company states:

“GPT-3 and GPT-J have recently shaken up the NLP world. Due to their size, these AI models are both highly effective at generating fluent human language and efficiently handle most modern automatic language processing use cases. With GPT-J, developers can perform a variety of advanced operations such as chatbots, data extraction from unstructured documents (such as web pages, emails, contracts, financial statements…), code generation, sentiment analysis, or marketing content creation, in many languages. New use cases are discovered almost every day.”

For more than three months, the NLP Cloud team has been working to make the GPT-J integration as successful as possible in order to offer a production-ready solution to its customers. Julien Salinas, founder and CEO of NLP Cloud talks about the arrival of this new NLP model in the interface that his company has designed:

“At NLP Cloud, we are happy to offer an open source alternative to the OpenAI-Microsoft monopoly, as we believe that everyone should be able to understand – if they want to – how an AI model works from the inside.”

Finally, NLP Cloud states that in order for advanced NLP models to be as ethical and responsible as possible, they should be able to remain open source and continually audited by the natural language processing community.

Translated from La start-up NLP Cloud intègre le modèle open-source GPT-J dans son interface de programmation NLP