Automatic Summarization using Deep Learning | Abstractive Summarization with Pegasus

[@soumyadeepnag6300]

Thanks for explaining the video nicely. But, does the pegasus model always generate one line of summary? Is there any way we can increase the number of summary lines?

[@user-og2cw5yv2m]

hey have you found how to generate multiple lines

[@naitikshah7969]

Whoever is getting an error while creating the tokenizer in step 1, run this command on your terminal: pip install sentencepiece Great video Nicholas, thanks mate!

[@satoshinakamoto5710]

Also restart kernel and run all cells if you get "None type error". Sometimes the cell that downloads the pegasus model fails.

[@ashikurrahman-ai8227]

@@satoshinakamoto5710 Thanks mate

[@vaishnavimore754]

Thanks

[@erfansthought]

How to make more than 1 sentence summary? It is possible to configure it to generate a summary of specified sentences like 10 sentence summary?

[@pradhansomu4100]

Bro please make a video on creating custom dataset for pose estimation and which architecture will be best to train

[@BS-ob3hp]

Hi Nicholas, thank you for your content here on youtube :)! I was just wondering if I can also use Preview or Stable, since LTS is not supported on a mac. Thanks!

[@NicholasRenotte]

Yup, suggest using stable!

[@mohammadabdulkadir8799]

Am getting this error when am trying to load the tokkenizer how can i resolve it TypeError Traceback (most recent call last) Cell In[25], line 2 1 # Load tokenize ----> 2 tokenizer = PegasusTokenizer.from_pretrained("google/pegasus-large") TypeError: 'NoneType' object is not callable

[@VladimirSkultetyOfficial]

Very good channel and videos. Thank you Nicholas!

[@mdrahatislamkhan9966]

cant install the torch. searched a lot but can't find the solution tried lots of things but got this error "No matching distribution found for torch" Please help me fix this problem.

[@Powercube7]

Pegasus? More like Pega SUS 😳 For real tho, this is actually really helpful, now i don't have to read anymore thanks fam 😂

[@NicholasRenotte]

I think you mean....Mega SUS?! Hahaha, hell yeah, bail on all that reading rubbish!

[@Powercube7]

@@NicholasRenotte sussy baka 😳😳😳

[@arshdeepsingh5950]

love you sir must say your brain is awesome.

[@Manideep.]

when i import the model i am getting AttributeError: 'Version' object has no attribute 'major'

[@priyamkakati4427]

Amazng content as usual!

[@anonymousking2053]

Nicholas why don't you start reinforcement learning for gaming, everybody wants it. by the way love your videos

[@NicholasRenotte]

Ik ik, just gotta get back to it.

[@Powercube7]

He is doing his best, we should have some patience. Quality over quantity!

[@urinater]

My results keep on repeating the first sentence. Will try other models

[@deepakkumarjha]

Hey, would you mind making a video on how the model could be fine tuned for a custom text dataset, because I read the paper and couldn't do it. It would also be a good continuation to this video.

[@rudreshmehta6510]

Hey, Did you find the way to do that?

[@shreevinayaka8219]

sir how to increase the number of words or how to keep it variable

[@e_souhil]

thanks, it was so helpful can you do a video on how we can fine-tune the pegasus model on a different dataset?

[@rudreshmehta6510]

Hey, Did you find the way to do that?

[@vent_srikar7360]

@@rudreshmehta6510 did you find it ?

[@shreeshaaithal-]

First I thought that you are talking the Pegasus virus then ooh ok summarize Pegasus 😂

[@NicholasRenotte]

😂

[@swishrsplitr]

great video, thanks - are there any summarization models that accept more than 1k tokens as input?

[@hosdvz1]

hi there. why do we bother ourselves to summarize just 512 tokens :)

[@satvikmanchanda7090]

reading research papers can really be hard, wish there was some trick

[@henkhbit5748]

Samenvattend, wederom een mooie introductie in NLP met de "vliegende paard" 👍

[@AutumnRainWish]

6:18 pip install transformers==4.11.3 if you are getting an error ImportError: cannot import name 'PegasusForConditionalGeneration' from 'transformers'

[@sanchitafusiondance]

thank u so much.. u have no idea how much u saved me..

[@asprakmantab]

can we set the length of summary?

[@akshay3361]

Can you make a text summarizer using gpt 3 or 2?

[@NicholasRenotte]

Could try!

[@rahil1575]

Is there any way to specify the length of summary

[@user-og2cw5yv2m]

hey have you found it?

[@idkidk1774]

Bro pls make a python chat bot with deep learning and actions like: join school class answer ok sir and join the class

[@NicholasRenotte]

Alrighty, will build it into the plan!

[@idkidk1774]

You are insane bro

[@trinitaroy5084]

Can you also show how do you fine tune the Pegasus model with a custom dataset for text summarization?

[@alexandregagne4151]

Like always, thank you. This channel is soooo good

[@NicholasRenotte]

Cheers @Alexandre!

[@shreevinayaka8219]

@@NicholasRenotte sir how to increase the number of words or how to keep it variable

[@hilariouzclips1235]

Great video, Nicholas.

[@vishnureddy7543]

Thanks Nick! Also how do I add a dataset to it instead of a piece of text?

[@NicholasRenotte]

For summarization? This is very much focused on NLP. For structured data I would be focused on using Pandas, got a crash course on the channel!

[@davidzhang4825]

Is Pegasus the best model for text summarization? if not which model is the best?

[@debatradas9268]

thank you so much

[@ryanw8765]

When companies build text summarization models like this one, do they create their own model and launch it for their app or do they generally use pre-existing models?

[@ahmedalameldien765]

hi nicholas, i really appreciate your video. thank you for this very informative video. could you make another one of how to fine tuning a custom text dataset ?

[@srishtigureja6534]

Great video! thanks 🎉 I was unable to understand the last part, what fine tuning can be done exactly for the model to perform better?

[@NicholasRenotte]

Can fine tune the underlying model on a dataset of your choice! Google did it on a bunch of different text corpuses, e.g. for Journals you could use this model: huggingface.co/google/pegasus-pubmed

[@srishtigureja6534]

@@NicholasRenotte Thanks!

[@rohankherath]

Thank You Nicho...

[@lahiruneranjanrathnayaka3665]

great video :D. How can we implement this for another language ?

[@NicholasRenotte]

Would probably look at converting to english first, summarizing then converting back. One of the other subscribers mentioned the summarization in other languages sucks, would try that approach instead!

[@lahiruneranjanrathnayaka3665]

@@NicholasRenotte Thanks. I'll give a try :D

[@ashleysami1640]

Don't lie. Bet you love Jane Austin.

[@NicholasRenotte]

M8 😂

[@alexandregagne4151]

I have a little question. I would like to create a model to recognize a person, but everything I find online and on youtube uses Face-recognition. However, I would like my model to be able to recognize a person, not necessarily by their face but also by a tattoo or a feature of their body and etc. What do you think would be the best technique to accomplish this task? Would a simple image classifier do the trick?

[@NicholasRenotte]

Aside from the ethical implications, you could look at using a siamese network. Keep in mind it requires a ton of data if you're to do it on more than just faces!

[@alexandregagne4151]

@@NicholasRenotte That's what I understood, I need a lot of data. To practice a little, is it possible to use the landmarks (face) to detect and recognize a person with mediapipe? I'm trying to use your method on the sign languages video, but I get an accuracy of around 27% after 4000 epochs, no good :(

[@NicholasRenotte]

@@alexandregagne4151 might be a bit late to this, have you checked out the facial rec tutorial?

[@mahdimahdavi1341]

thank's

[@aliffnabil5542]

yooooo this is sick!

[@TejrajParab]

I remember that nightmarish novel.

[@NicholasRenotte]

Don't even get me started. Honestly I hated every minute of that class and I definitely made it known. What a complete waste of time....if only i knew back then I'd end up coding, would've bailed completely!

[@utkar1]

Hey thanks man! I'm wondering is there is way we can create book summaries as well with one of these transformers?

[@TKollaKid]

I learned so much from this video. Liked and subscribed. Thank you, Nicholas!

[@idkidk1774]

love it

[@NicholasRenotte]

🙏🙏🙏

[@litttlemooncream5049]

mark! gonna take a try. many thanks

[@Brian5094]

Really dope! I was looking for the turtorial to guide me through the summarization model and your video has extremely high quality and super practical! I have a question that is abstractive summarization need to be fine-tuned? If so, how can we do it? :D

[@detour1055]

Just found you today...absolutely love your content and wide range of projects. I'm not a programmer but I'm looking to complete some projects very similar to what you've showcased in your videos. Are you available to hire?

[@erfansthought]

I am getting 'NoneType' object is not callable after this code in colab tokens = tokenizer(text, truncation=True, padding="longest", return_tensors="pt") Solution please. Great Videos. Keep it up.

[@NicholasRenotte]

What's in the text?

[@erfansthought]

@@NicholasRenotte TypeError Traceback (most recent call last) in () ----> 1 tokens = tokenizer(text, truncation=True, padding="longest", return_tensors="pt") TypeError: 'NoneType' object is not callable

[@qwertl99]

running into same problem

[@qwertl99]

reloading jupyter fixed my issue :)

[@erfansthought]

@@qwertl99 Didn't work for me in colab

[@guimaraesalysson]

Great video

[@mannatsiddiqui5499]

Getting error TypeError Traceback (most recent call last) in () 4 model = PegasusForConditionalGeneration.from_pretrained("google/pegasus-xsum") 5 # Load tokenizer ----> 6 tokenizer = PegasusTokenizer.from_pretrained("google/pegasus-xsum") TypeError: 'NoneType' object is not callable Please tell how should I resolve it

[@jetakota]

PegasusTokenizer.from_pretrained('google/pegasus-xsum') returns None. PegasusTokenizer also returns None for 'google/pegasus-large'

[@vent_srikar7360]

hey iam getting this error of -: tokenizer = PegasusTokenizer.from_pretrained("google/pegasus-xsum") TypeError: 'NoneType' object is not callable any idea whats the mistake ?

[@laurentprea7746]

Who else is having issues installing pytorch? The error is shown below ERROR: No matching distribution found for torch==1.8.2+cu111

[@NicholasRenotte]

If copied directly from the PT site, you should be good to go. Possibly another version of Python required? I used 3.7.3 in the video with no issues.

[@joshuaemmanuelaudu4799]

I really hope you reply to this. Thanks so much for.this project. It worked before.. now for the autotokenizer.from_pretrained(google/pegasus-xsum) .. it's giving an error that filenotfound