You can now try JAIS 30B Chat, Arabic-language large-language model (LLM), on Microsoft Azure

The model was pre-trained on 1.6T tokens, 475B of them were Arabic.

Reading time icon 2 min. read


Readers help support MSpoweruser. We may get a commission if you buy through our links. Tooltip Icon

Read our disclosure page to find out how can you help MSPoweruser sustain the editorial team Read more

Key notes

  • Microsoft announced that JAIS 30B Chat is now available on Azure AI.
  • It’s an Arabic-language LLM from the United Arab Emirates, launched last year.
  • A lot of recent models, like GPT-4o and Phi-3-vision, are also available on Azure.

When JAIS first arrived last year in the United Arab Emirates, the large-language model (LLM) had a promising application. It still does, and now, it expands even further as it’s now available on the Microsoft Azure AI model catalog.

Announced just in time for the AI-centric Build 2024 conference, Microsoft says that JAIS 30B Chat is available to try out on Azure “for increased flexibility in model choice and addresses the multilingual needs of companies.”

Based on OpenAI’s GPT-3 architecture and pre-trained on a total of 1.6 trillion tokens in both Arabic and English, JAIS is a bilingual LLM for both languages, so much so that it was also pre-trained on 475 billion Arabic tokens. It’s now offered as an API through Models as a Service in the model catalog in both Azure AI Studio and Azure Machine Learning Studio.

So far, however, the only available model from the family on Azure is the JAIS 30B Chat, an instruction-tuned variant with 30 billion parameters. It features advanced architecture for handling longer inputs and efficient processing, with a custom tokenizer that enhances text generation quality and speed.

A lot of recently launched models, like OpenAI’s latest GPT-4oCommand R+, and Phi-3-vision, have also been made available to try on Microsoft Azure. The Redmond-based tech giant has also poured $1,5 billion into cloud & AI infrastructure in the UAE through G42, an Emirati AI holding company that developed this model.

As for JAIS 30B Chat, Microsoft says that you’ll be billed based on the number of prompts and completion tokens. Paygo-inference-input tokens cost $0.0032 per 1k tokens, and paygo-inference-output tokens cost $0.00971 per 1k tokens.

User forum

0 messages