Confused which Transformer Architecture to use? BERT, GPT-3, T5, Chat GPT? Encoder Decoder Explained

This video explains all the major Transformer Architectures and differentiates between various important Transformer Models.
Which Transformer Architecture to use to solve a particular problem statement in Natural Language Understanding (NLU) and Natural Languages Generation (NLG) is explained in a simplified manner.
Over the past 6 years, Transformers, a neural network architecture, have completely transformed state-of-the-art natural language processing and the way we approach to different problem statements in NLG and NLU.
Chapters:
0:00 Introduction
1:21 Encoder Branch
1:57 BERT
2:37 DistilBERT
3:19 RoBERTa
3:59 XLM
4:50 XLM-RoBERTa
5:32 ALBERT
6:40 ELECTRA
7:19 DeBERTa
8:13 Decoder Branch
8:50 GPT
9:13 CTRL
9:54 GPT-2
10:31 GPT-3
11:30 GPT-Neo/GPT-J-6B
11:50 Encoder-Decoder Branch
12:00 T5
13:05 BART
13:46 M2M-100
14:22 BigBird
#datascience #neuralnetwork #machinelearning #naturallanguageprocessing

[@datafuseanalytics]

In this video, I tried to explain all the major Transformer architectures. I have also explained the differences and training objective of each one of them. If you feel this video adds value in your life then please like, share and comment on this video and subscribe to this channel. If any suggestions and feedback then please drop in comment box.

[@hemantwani4757]

Very nicely explained ❤👍

[@aurkom]

It would have been awesome if all the models had the release year mentioned along with it as well. Helps to get a birds eye view of the timeline.

[@datafuseanalytics]

Hello. Yes, I am making a separate video on similar topic. It will be uploaded soon. Stay tuned my friend.

[@kevon217]

thanks for the excellent, well-explained summary!

[@datafuseanalytics]

Thank you Kevin

[@santoshpanigrahi5711]

Thanks for sharing. It's very informative. Keep up with this work.

[@datafuseanalytics]

Thank you, Santosh, for watching the video.

[@milindkubal2738]

Amazing. Great work👍

[@datafuseanalytics]

Thanks Milind

[@SagarBhalke-td3vy]

Great explanation. Thank you very much

[@datafuseanalytics]

Glad it was helpful for you Sagar...

[@ajitkumar15]

Very nice and to the point video, thank you !!!

[@datafuseanalytics]

Hey thanks a lot Ajit 😃 🙏

[@falknfurter]

I just found this video and it's very good. I'm currently trying to understand when to use what type of model. Looking at Huggingface is just overwhelming. That's where this video jumps in and provides an excellent overview of the major models. I wish there would be a similiar video explaining the various pretraining objectives.

[@datafuseanalytics]

Hello. I will definitely make a video on the same. Thanks a lot. 😀

[@sagar3482]

Informative content Thanks for sharing this

[@datafuseanalytics]

Glad you liked it!

[@ahmedelsabagh6990]

Greate video!

[@datafuseanalytics]

Thanks a lot. Please do share it with your friends 😁

[@ganeshkharad]

this is really nice explaination!!!

[@datafuseanalytics]

Thanks a lot Ganesh 😃 🙏

[@adityakshirsagar1391]

Informative 👍

[@datafuseanalytics]

Glad it was helpful and informative for you Aditya. Please do share it with your friends. More interesting videos will be uploaded soon

[@SaketKumar-wy1wb]

This is good. Keep up the good work. 🙂

[@datafuseanalytics]

Thank you Saket, I will

[@sanjaybhalke8032]

Thanks for sharing

[@datafuseanalytics]

My pleasure

[@user-os1xi8qf4y]

thank you sir ! Fantastic method of explanation

[@datafuseanalytics]

Hey buddy. Thanks a lot. 😀

[@datafuseanalytics]

Hey buddy. Thanks a lot

[@rembautimes8808]

Excellent video and I joined as a sub. Like this style of going thru the various architectures and the use case. Maybe you can also update it with GPT 4 since it’s new out there.

[@datafuseanalytics]

Thanks a lot for this amazing comment. I have uploaded the latest video using ChatGPT model - kzbin.info/www/bejne/g3yrdpSumdSkrqM Please go through it and feel free to comment

[@exxzxxe]

Well done!

[@datafuseanalytics]

Thanks David.

[@d4munche3z]

Can you create a tutorial on Longformer and the concepts/code used to adapt an LLM for larger token sizes?

[@datafuseanalytics]

Hello David. I haven't made it yet. But I will definitely make one on Longformer etc which takes a whopping 4096 tokens as input. Thanks for your feedback.

[@WillBeebe]

Superb 🎉

[@datafuseanalytics]

Hey thanks William

[@sarc007]

Excellent

[@datafuseanalytics]

Thanks a lot Suhail.

[@amortalbeing]

thanks a lot❤

[@datafuseanalytics]

You are most welcome 😃 Do check other videos too on AI on this channel.

[@markfallu2389]

Great summary- would be good if you did an update

[@datafuseanalytics]

Sure. I will make an updated video comprising of all the possible model architectures

[@mhaya1]

Kudos🎉

[@datafuseanalytics]

Thank you 😃

[@projectbit2248]

Hello, how do I contact/ connect with you, with regards to a project?

[@datafuseanalytics]

Hello, please contact us via our email. [e-mail]

[@tilkesh]

Thx

[@datafuseanalytics]

Most welcome 😃 😊

[@ianboyles2425]

there's some new important ones like the newer gpt Neo models, alpaca, llama, cereus, vicuna

[@datafuseanalytics]

Hello Ian. Yes. At the time of this session, these models weren't available. Thank you for your feedback. I will definitely make one video (part 2) which will encompass these models in a more simpler fashion

[@chenpeter7428]

It seems it does not cover BERT in computer vision.

[@datafuseanalytics]

Yes you are right Chen Peter

[@ko-Daegu]

this sounds like copy pasted from online articles and just reading from them without extra info at all

[@datafuseanalytics]

Hey Ko-Jap. I referred multiple books for the same and then wrote the content in my language. But I did not refer to any online blogs or articles. Only books are the reference. But thank you for your valuable feedback. I will improve so that it doesn't sound as I am reading. 🙏😀

[@yosup125]

for the algo

[@datafuseanalytics]

Thank you

[@gregbugaj]

Nice overview

[@datafuseanalytics]

Hey Thanks a lot 😃

[@saketkr]

This is good. Keep up the good work. 🙂

[@datafuseanalytics]

Hey Thanks Saket