Getting Started with AI powered Q&A using Hugging Face Transformers | HuggingFace Tutorial

This video gives an introduction into how to use existing pre-trained AI models in your own solutions with your own data. I give an introduction to Hugging Face and their AI Model Hub and how you can use that to test Q&A (question and answer) functionality with your own data. We even pull some random content from the internet and ask questions of that content that the model has never seen before. I then give an introduction to Transfer Learning and BERT (a state of the art Natural Language Processing (NLP) model that powers many AI functionality today).. We look at how BERT works and how it is pre-trained with wikipedia and BookCorpus and why it's advantageous just to use a pre-trained model rather than training your own model. We then look at BERT is then fine tuned on a Stanford Q&A dataset called SQuAD 2.0 so it learns how to answer questions (so that you can use it with your own data).
Finally we code up some of our Q&A routines in Python in Jupyter Notebooks hosted in Google Colab so that you can get an idea on how easy it is to embed pre-trained AI's models in your own solutions with your own data.
00:00 - Intro
01:20 - Hugging Face Model Hub
03:49 - Using a BERT model on HuggingFace
08:49 - Introduction to Transfer Learning
10:12 - Understanding BERT
15:55 - Datasets used to build pre-train BERT (Wikipedia and BookCorpus)
18:41 - Fine Tuning BERT to understand Q&A with SQuAD 2.0
22:24 - Coding our model with HuggingFace Pipelines using Google Colab
29:34 - Coding our model with TensorFlow using Google Colab

[@zhuchencao2527]

Thx! Although I have been learning transformers for a while now, I still think this video is worth watching.

[@chrishayuk]

Glad it’s useful

Hey! Thanks for the content. It's really helpful for me. Btw, you can use cmd+shift+v to paste copied content with no formatting 🙌 So no need to paste it to text editor etc and convert it to plain text.

[@rafalyp73]

Mega pro tip: Ctrl+Shift+V pastes plain text.

[@MN-fl3hb]

Great work!

[@chrishayuk]

thank you

[@sirrr.9961]

Best tutorial for us noobs so far on the KZbin !

[@chrishayuk]

Glad it’s useful

[@evegarcianz]

Nice video

[@omertaub3198]

Which model have you used when you wrote the huggingface's pipeline code?

[@Twi1ightFr0st]

Hello, have you tried using WASM for ML or DL inferencing especially for NLP? Videos on this topic would be great!

[@chrishayuk]

it's coming 😀

[@srijamukherjee5215]

Could you please display the full line of code: answer=tokenizer.convert_tokens_to_string()....the last part is cannot be seen

[@hrishikeshparmar]

which algorithm are you using for qa model?

[@binitapriya4976]

Hi, is there any way to create a dataset( question and answer format) from a random web scraped data?

[@phily8020]

Have you come across any that can summarize a whole book in pdf format effortlessly? They all seem limited to a small pdf

[@chrishayuk]

Anthropic Claude has a 100K token limit, so that should work

[@abhishekprakash9803]

its generate only one line of answers,,,,,,,,suppose we have big text......like we have data visulization text, but when u aksed question like what is data visulization.....its only response in one line , i want more how we do that..

[@iva1389]

please attach the code in the future videos, I can't be coding along with every video, and sometimes to have the notes that I can acess later would be great. thanks!

[@chrishayuk]

great shout, i usually create the code examples and some slide examples when creating the videos. i will start including them soon (and try and back update some of the previous videos with it)

[@iva1389]

@@chrishayuk sure, please do, you are covering important and interesting topics. it would be nice to include the code, just for the sake of completion.

[@abhishekprakash9803]

@@chrishayuk wha is logits here,

[@husseinmenkam8350]

great video sir, but next time please can you out your face in a place out of the code so wile we are trying it we can see what you typed it will be veri good

[@chrishayuk]

Will share the code

[@testingemailstestingemails4245]

how to do that trained huggingface model on my own dataset? how i can start ? i don't know the structure of the dataset? help.. very help how I store voice and how to link with its text how to organize that I an looking for any one help me in this planet

[@lightyagami6823]

i'll help you.I am an AI developer

[@taposiahmed9745]

Hi..is there any way to get answers without adding any content? I want to pull answers from web

[@microgamawave]

Connect it to Wikipedia api

[@thebicycleman8062]

dont ever put that kind of music in the background again - makes 0 sense to have that level of intense music with many frequencies that directly clash with your voice - we don't need a WWE hype up song to watch a tutorial about A.I - respect your viewers please

[@sathwikmadhusudan5063]

Chill

[@kachrooabhishek]

its not practical for production level applications . We cant pass the context every single time to the model and every possible person has failed to handle to show case that problem

[@chicopaisagem]

Very comprehensive!! I am planning of creating a QnA bot using mosaicml/mpt-7b, and I am planning to fine tune it to health and care. Let me know if you would like to exchange some thoughts, thanks 🙂

[@chrishayuk]

I literally just released a video on fine tuning on colab, hope it’s useful

[@chicopaisagem]

@@chrishayuk outstanding I will definitely have a look, thanks 👍