What language model is Azure Question Answering using?

Daniel Hjelm 50 Reputation points
2023-02-27T13:54:25.98+00:00

Hi,
I want to understand how Question Answering works in more detail. In this post https://techcommunity.microsoft.com/t5/ai-cognitive-services-blog/qna-maker-is-being-retired-hello-question-answering/bc-p/3753638#M396 from May 30 2022, the writer states "Powered by state-of-the-art transformer models and Turing natural language model, Question Answering is Microsoft Azure’s latest intelligent Q&A offering with marked improvements in relevance and quality over QnA Maker". Does this mean that Question Answering uses T-ULRv5 (or maybe T-ULRv6) https://www.microsoft.com/en-us/research/blog/microsoft-turing-universal-language-representation-model-t-ulrv5-tops-xtreme-leaderboard-and-trains-100x-faster/ as it is underlying language model which is fine-tuned to answer the custom questions which is put as Sources of the Question Answering resource?
If there are any other blog posts, research papers, etc., which talk about this topic, please link them.
Thanks in advance!

Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
2,354 questions
{count} votes

Accepted answer
  1. YutongTie-MSFT 46,091 Reputation points
    2023-03-01T15:55:52.62+00:00

    Hello Daniel Hjelm

    Thanks for your follow up. Just got the confirmation today and the model we are using is Turing NLP models. Product team can not provide more details about what model is using for which feature exactly since it is confidential, but there is a website talking about those models, you may be interested in -

    https://turing.microsoft.com/

    It does mention T-ULRv6 is the latest version of the model -

    User's image

    And also there are pages you may want to take a look at -

    Introduce for Turing Bletchley

    https://www.microsoft.com/en-us/research/blog/turing-bletchley-a-universal-image-language-representation-model-by-microsoft/

    One informatic page -

    https://www.microsoft.com/en-us/research/blog/using-deepspeed-and-megatron-to-train-megatron-turing-nlg-530b-the-worlds-largest-and-most-powerful-generative-language-model/

    I am sorry due to the confidential policy we can not share a lot, I hope above website and pages help.

    Regards,

    Yutong

    -Please kindly accept the answer if you feel helpful to support the community, thanks a lot.


0 additional answers

Sort by: Most helpful