AI Text and Code Generation with GPT Neo and Python | GPT3 Clone

what's happening guys my name is nicholas ronatt and in this video we're going to be taking a look at a ipad text generation using a gpt3 clone called gpt neo in just four lines of code let's take a deeper look at what we'll be going through so first up what is gpt3 well gpt3 is a deep learning powered language model trained on 175 billion parameters by open ai now this means that it would take a huge amount of time to train something as sophisticated as this on a consumer gpu the amazing thing about this particular model is that it's really really flexible and performs really really well on a range of natural language tasks specifically like text generation q a sentiment and classification now the unfortunate thing about gpt3 is that it is only available through a closed beta the great thing however is that you're able to leverage a gpg3 clone called gpt neo now the model that we're going to be taking a look in this particular video is trained on 2.7 billion parameters so it's not exactly the 175 billion parameter model the open ai built but it is really really sophisticated and you'll see that a little bit later so let's take a look as to what we'll actually be going through so first up what we're going to be doing is installing gpt neo so gpt neo is going to give us a gpt3 clone which is trained on 2.7 billion parameters and can definitely be fine-tuned so if you'd like to see a video on fine-tuning this by all means do let me know so once we've installed it we're going to be able to generate text in just four lines of code so it's really really straightforward and what we'll also do is we'll save that output to disk so you can pick it up as a text file and leverage it later on now you're probably thinking how is this all going to work well let's take a look so first up what we're going to be doing is we're going to be leveraging the hugging face transformers library to load our pre-trained gpt neo model then we'll pass through a text prompt so say for example we might ask what is the meaning to life import pandas is pd if we wanted to generate python code we could even generate some sql statements as well and then once we've generated that text or once our models generated that text will then be able to output that result to a text file using a standard python with statement ready to do it let's get to it alrighty guys so in order to get started with gpt nero there's going to be four things that we need to do now we're going to keep this tutorial relatively short and sharp we're just going to power through it and ideally by the end of this you should be able to get up to speed with gpt neo and start generating some of your own text so the four things that we need to do are install and import our dependencies set up our pipeline or our generator then pass through some text and generate text using a prompt and then what we're going to do is output that text and save it to a file so first things first let's go ahead and install and import our dependencies now the first dependency that you're going to need is pytorch so the easiest way to install this is if you just go over to the pytorch website so pytorch.org and again all the code for this as well as links to relevant documentation will be in the description below so you can pick this up and run with it so what we're going to go on ahead and do is go to the pytorch website and then towards the bottom you're going to see this section called install pytorch now all we need to do here is choose the build that we want so we're going to choose stable in this case choose your operating system in this case i'm on a windows machine so i'll choose windows choose what package you want to install so i'm going to use pip but you could use conda as well if you wanted to in this case we're going to be working with python so we're going to choose python we're leveraging gpu acceleration using cuda 11.1 now gpu acceleration really isn't going to kick in unless you've got a massive gpu that's able to handle this particular model but that's fine we'll install it using cuda 11.1 anyway so once we've got that you're sort of going to get this command down the bottom here so i'm just going to copy this and paste it into my notebook now in order to run a command line within your notebook you just need to include an exclamation mark and i'm just going to remove the three from the pip command because i'm just using standard pip and then all we need to do is just run this cell so that's going to go ahead and install pi torch now i've already got it installed so it went relatively quickly so that's our first install done so installing pytorch remember all you need to do is go to pytorch.org select your different options and copy that command run it in a cell now the next thing that we need to do is actually install transformers so transformers is a really powerful and probably one of my favorite natural language processing libraries and the beautiful thing about transformers is that you get a bunch of these pipelines so you get a conversational pipeline feature extraction pipeline text generation pipeline a translation pipeline a whole bunch of pipelines that allow you to leverage really really sophisticated nlp models relatively easily so we're going to be using transformers exactly for that so let's go ahead and install transform so this one's relatively straightforward so in order to install transformers all we've gone and done is run exclamation mark pip install transformers and that's going to go on ahead and install all of this good stuff here for us so those are our two dependencies now installed now all we need to do is import our main dependency which is going to be transform so let's do it okay so that is our transformers library now imported so in order to do that i've written from transformers import pipeline so this is our first line of code so first line now this pipeline over here is effectively going to be giving us all of these pipelines available now we're going to be using the text generation pipeline so in a second you'll see that we pass through the text generation argument to that particular pipeline to be able to leverage our text generation using gpt near but again if you want to deeper dive into this by all means do let me know in the comments below i was thinking a mega tutorial might be useful for gpt might also be useful for transformers particularly transformers because there's a bunch of stuff in here now in this case that's our pipeline now imported now what we need to do is set up our generator so we're going to write a second line of code alrighty that's our second line of code now done and dusted so this line of code here is actually going on ahead and loading up our gpt neo model which is going to give us our gpt3 clone so in this case what we've written is generator equals pipeline and then to that pipeline we've passed what type of pipeline we want in this case we're going to be leveraging the text generation pipeline so to do that we've written inside of a string text dash generation and then i've passed through a comma and then i've pass through a keyword parameter so this next keyword parameter is what defines what type of natural language model or what type of language model you're going to be loading into that pipeline now in this particular case we're going to be leveraging the gpt neo model with 2.7 billion parameters so to do that we've written eleuther a i forward slash gpt dash neo dash 2.7 b so this means that we're going to be leveraging the eluther ai which are the guys that have actually gone and built the gpt neo model and then by passing through slash gpt neo dash 2.7 b we're importing the 2.7 billion parameter model now if you're doing this for the first time it's actually going to download the model so i believe it is about 10 gigabytes so it might take a little bit of time to download but it will do everything that you need to do for you so you don't actually need to go and do anything else it will just download it for you onto disk so you're able to leverage it there is also a 1.3 billion parameter model so if your computer is running a little bit slow and you wanted to run a smaller model you can do that so let's let that go on ahead and load and then we'll be able to run our model okay so that is our generator now loaded so you can see here that that particular cell is run so we're no longer have a little asterisk in there so we're now able to run it within our notebook the next thing that we're actually going to go on ahead and do and so this is our second line of code second line next thing that we're going to go ahead and do is start generating some text using a prompt now this prompt could really be whatever you wanted to so if you wanted to write a love story write a poem and get some answers to what is the meaning of life this is where you could pass it through so in this case what i'm actually going to do is let's start off with that so we'll type in what is the meaning of life and what will actually happen so let's just make sure this is prompt so what will actually happen is when we pass this prompt to our gpt neo model it's actually going to try to generate text using that prompt so you'll actually see that it's trying to answer that maybe write a bit of a blog post and you'll actually get some answers generated using the gpt model which i think is absolutely fascinating but anyway let's actually go on ahead and do it so what i've done is i've created a new variable called prompt so p-r-o-m-p-t and then i've just set that equal to a string and you'll see in a second we'll change the string we'll try generating some other styles of text maybe some code as well then what we need to do is actually pass that prompt to our generator to be able to actually generate code so let's do that alrighty before we go ahead and run that next line of code let's take a look at what we actually wrote let's actually include some additional cells so we can see this so this is our third line of code up here this is our fourth line so after this we'll actually have our generated text of fourth line now in this particular case what i've gone and done is created a new variable called res for results and then i've set that equal to our generator so in this case you can see that we set up our generator here what we're doing now is passing some arguments and keyword arguments to that particular generator to generate text so here what we've gone and written is generator and then to that we've passed through our prompt which is this string over here then we've just gone and set some keyword arguments so you can play around with these if you wanted to and this will influence the results you get back so the first argument is max underscore length equals 50. so this defines how long your output is going to be in this case i've set it to a limit of 50. then i've passed all 50 words then the next keyword argument is do underscore sample equals true so this is going to allow us to leverage sampling within our model and then we've also gone and set our temperature to 0.9 so this influences how it actually determines results going forward from a particular token that it's actually had returned back so again you can play around with these parameters tune them play around take a look at the documentation again this tutorial is going to be super fast and we're not going to delve into it too much but on that note let's go ahead and run this line and generate our text okay so that line of code is now run now the speed of which that actually generates is really dependent on your machine so in my case it took about 20 30 seconds to generate a string with a max length of 50 words but in this case if you were to pass through a higher maximum length number it will take longer now what we can actually do is take a look at our results if i type in res down here you can actually see that it's gone and generated this big block of text so you can see that it's gone and says what is the meaning of life why does it matter oh it's not exactly that happy at the moment where do i go to find the meaning or at least some sort of explanation for what i'm experiencing here and now i've been asking this a dozen times now now if we actually go and pass this result so if we grab the first result and type in dot generated text this is actually going to give us our raw string so you can see and if we print this out we'll get our string formatting so you can see we've actually got our line spaces so in this particular case just by passing through this short prompt we've gone and got this entire result out now if we wanted to generate more text we can just change the maximum length so say i set it to 100 run that again we're going to get a longer string generated so let's let that go on ahead and run and then we'll ideally get a longer answer to what is the meaning of life there you go so it's now gone and generated a new block of text so all we really did there is we changed the max length parameter and in this particular case it's gone and generated a bigger block of text so what is the meaning of life so it still starts out with that prompt in a recent conversation with a friend i heard him say you must go through life with the attitude that you are never going to leave i had to check on whether he actually said this to me or if it was in the form of a quote or something similar i decided to find the original source of this quote after reading about it in a book on self-help quickly found a website so in this case it's writing a slightly different it's going down a completely different route but again this is just a random property say for example we wanted something a little bit more contextual where so i'm going to set our max length back to 50 and what if we said something about bitcoin so if we said bitcoin is going to and then rend that particular prompt so if we run that now let's go and see what gpt neo now generates there you go so it's now gone and generated a new text so if you take a look it's gone and written bitcoin is going to be the most important currency in 2020. the crypto market is on the rise and bitcoin has gone from a joke in late 2016 to the world's favorite cryptocurrency in 2018. well it's not wrong i mean 2021 and it's super popular there's been a massive rally in 2018 and 2019 and again we've cut it off at 50 but you can see that it's actually generating some really really interesting text results now you could use this to generate blog posts maybe write a song write a poem so on and so forth but my particular favorite thing is actually using it to write some code now i was testing this out earlier this week and i thought it was super fascinating so instead of writing our prompt as text what happens if we wrote one of the most common lines of code that i wrote so import pandas as pd and run that now so let's go and see if it actually goes and generates some valid python code and there you go so what it's actually gone and written is import pandas is pd which was our original prompt but then it's gone and passed through a bunch of additional imports so from collections import counter import os import sys import numpy is np so it's almost like it's importing some pretty standard data science libraries i mean the fact that it's got numpy it's got time it's got nltk down here so you can see that it's actually really really contextually aware and it's able to generate a whole bunch of stuff i'm probably going to be playing with this all weekend but you can sort of see what's possible with gpt neo now if you wanted to go on ahead and export this text you could do that really easily so let's go ahead and do that okay so before we go on ahead and run that so i just want to sort of show what i've written so with open and then to that we've passed through the name of the file that we want to generate or output so in this case i've called it gpt text dot txt so you can see that there and then we've passed through the w flag because we want to write it out and then we specified as f so this means that we're going to be able to work with our file as a variable called f then in order to write out our text i've just written f dot write lines and then i've passed through this text block here so this text block is exactly the same as what we were printing out up here now we're just outputting it so if we actually go and run that block of text and if we take a look at the folder that we're working in let me bring it up you can see that we've now got this file called gpt text if we zoom in a little bit so you can see that better so you can see that we've got gpt text there and if we actually go and open that up you can see that it's exported all of our python code so maybe you could write that use this to write an entire data science workflow who knows what's possible and where you can take this to and again if you wanted to use a different prompt so say for example we wrote something about the current stock market current stock market and just run that we could now go and export different blocks of text so again whatever you pass around that prompt this particular workflow is going to output that to a text file so let's let that run and we'll take a look just to make sure that it's now outputting our new block of text so there you go so what we've now gone or what gpt's now gone and written so is the current stock market is a unique market in that it incorporates some fundamental market characteristics correct uh together with market psychology human judgment and human biases such as market timers price momentum and crowd psychology the stock market is like any market for goods and services and if we go and check out our file so you can see that it's now gone and exported the text that we had exactly as we had inside of our notebook pretty cool right so this sort of shows you what's possible with gpt neo in a nutshell now again all we did is we installed and imported our dependencies so we went and installed pytorch and transformers and then we wrote four lines of code to allow us to go and generate our new blocks of text using gpt neo so we wrote our import our generator our prompt and then we actually went and ran our prompt so again really really quickly you're able to get up to speed with this and then you're able to output this text so again you might take this further you might use this to automatically post to a wordpress website you could generate uh different articles you could generate blog posts i don't know research papers there's a whole bunch of different use cases that are possible for gpt neo on that note that about wraps it up thanks so much for joining guys hopefully you enjoyed this video if you did be sure to give it a thumbs up hit subscribe and tick that bell and let me know how you went with leveraging gpt neo i'd love to hear your thoughts thanks again for tuning in peace

Channel: Nicholas Renotte

Views: 17,798

Rating: 4.9413333 out of 5

Keywords: gpt 3 tutorial, gpt 3, gpt 3 demo, gpt 3 alternative, gpt neo demo, gpt neo tutorial, gpt neo huggingface, gpt neo example, gpt neo api, gpt neo parameters

Id: 6MI0f6YjJIk

Length: 18min 5sec (1085 seconds)

Published: Wed Apr 28 2021