1 00:00:00,080 --> 00:00:00,440 All right. 2 00:00:00,440 --> 00:00:01,920 So first of all I'm going to take my mouse. 3 00:00:01,920 --> 00:00:03,400 I'm going to hover around here. 4 00:00:03,440 --> 00:00:05,960 I'm going to press the plus code button. 5 00:00:06,240 --> 00:00:07,480 It's very like cursor. 6 00:00:07,720 --> 00:00:13,520 Uh and I'm going to go here and type two plus two and hold down shift and press enter. 7 00:00:13,760 --> 00:00:21,320 And we just used a Tesla T4 machine, uh, on the cloud in order to do this important operation. 8 00:00:21,520 --> 00:00:24,680 Uh, and, you know, we can we can do bananas. 9 00:00:25,120 --> 00:00:27,680 Uh, and uh, there we go. 10 00:00:27,720 --> 00:00:29,160 We get we get bananas. 11 00:00:29,320 --> 00:00:35,280 Um, and, um, yes, you you can imagine all the other great things we could do. 12 00:00:35,680 --> 00:00:42,800 Uh, a lot of stuff comes installed automatically on radios that you don't even need to pip install. 13 00:00:42,800 --> 00:00:46,240 So I can do like import gradio as GR. 14 00:00:46,280 --> 00:00:47,680 And I think that will just run. 15 00:00:48,120 --> 00:00:49,520 Doesn't need anything installed. 16 00:00:49,520 --> 00:00:52,040 Gradio just comes on the box. 17 00:00:52,040 --> 00:00:53,720 So it's as slow as it is on my machine. 18 00:00:53,720 --> 00:00:54,320 Look at that. 19 00:00:54,640 --> 00:00:55,960 Uh, you get used to this. 20 00:00:56,000 --> 00:00:57,200 Obviously that's the stop button. 21 00:00:57,200 --> 00:01:00,110 You can press the stop at any point if it's taking too long. 22 00:01:00,110 --> 00:01:01,030 But it did finish. 23 00:01:01,430 --> 00:01:08,030 Um, if you need to pip install something you can't just do like pip install pip install requests or 24 00:01:08,030 --> 00:01:08,550 something. 25 00:01:08,670 --> 00:01:13,830 Uh, you would have to put an exclamation mark in front of it, which is a way of saying don't run this 26 00:01:13,830 --> 00:01:17,350 as Python, run this as an underlying shell command. 27 00:01:17,510 --> 00:01:18,630 That's what that would do. 28 00:01:18,630 --> 00:01:21,310 But for this notebook we don't need to do any Pip installs. 29 00:01:21,310 --> 00:01:23,310 Everything is good to go. 30 00:01:23,830 --> 00:01:26,390 The first thing we're going to do is a little bit of code here. 31 00:01:26,790 --> 00:01:34,430 Uh, to check the GPU of this code, of this box, whatever this is running on out in the cloud by calling 32 00:01:34,430 --> 00:01:41,630 this this shell command Nvidia SMI to get the GPU info and find out what it is and print it if it is 33 00:01:41,630 --> 00:01:42,790 a Tesla T4. 34 00:01:42,830 --> 00:01:44,150 Let's run this code. 35 00:01:44,390 --> 00:01:44,990 Bam! 36 00:01:45,150 --> 00:01:46,270 Here's the results. 37 00:01:46,270 --> 00:01:49,070 It is indeed an Nvidia chip. 38 00:01:49,230 --> 00:01:50,110 And there we go. 39 00:01:50,110 --> 00:01:52,110 It is a Tesla T4. 40 00:01:52,150 --> 00:01:53,150 Good to know. 41 00:01:53,430 --> 00:01:57,510 And it has 15MB on it. 42 00:01:57,630 --> 00:02:02,590 And that makes sense because we can see just over here it does indeed, according to Colab, have that 43 00:02:02,590 --> 00:02:03,310 as well. 44 00:02:03,310 --> 00:02:07,350 So the code that's running is giving us a consistent story. 45 00:02:07,350 --> 00:02:08,230 That's good to see. 46 00:02:08,270 --> 00:02:11,870 Okay, next up comes the most important thing we're doing today. 47 00:02:12,190 --> 00:02:13,670 Pay attention please. 48 00:02:13,710 --> 00:02:15,630 There's a fiddly thing to do now. 49 00:02:15,750 --> 00:02:20,310 One fiddly thing one time only and then fiddly ness is over. 50 00:02:20,630 --> 00:02:27,790 Okay, we need to connect this colab to our hugging face account, and in fact, we're connecting any 51 00:02:27,790 --> 00:02:33,870 colab when we're logged in by using a feature that's secrets that are built into colab. 52 00:02:34,150 --> 00:02:36,870 And this is the alternative to the EMV file. 53 00:02:36,870 --> 00:02:39,830 You don't have a dot env file on Colab because we don't. 54 00:02:39,910 --> 00:02:42,590 It's not on our file system, it's out in the cloud. 55 00:02:42,790 --> 00:02:44,230 Here's how it works. 56 00:02:44,230 --> 00:02:49,910 First of all, you need to have got yourself a hugging face account, which you did in the last video 57 00:02:49,910 --> 00:02:50,750 or two videos ago. 58 00:02:50,750 --> 00:02:51,830 So you're good there. 59 00:02:51,870 --> 00:02:52,550 Thank you. 60 00:02:52,710 --> 00:02:54,750 You you you listen to me. 61 00:02:54,750 --> 00:02:56,420 So at this point, please. 62 00:02:56,420 --> 00:02:59,300 Let's go to your hugging face account. 63 00:02:59,300 --> 00:02:59,460 Here. 64 00:02:59,460 --> 00:03:01,900 I have mine open, conveniently hugging face. 65 00:03:02,100 --> 00:03:08,940 So this is me signed in, and, uh, you go to your avatar, you go to your profile. 66 00:03:08,940 --> 00:03:10,820 That is my hugging face profile. 67 00:03:10,860 --> 00:03:13,060 And I have two notifications. 68 00:03:13,060 --> 00:03:13,820 I don't know what they are. 69 00:03:14,020 --> 00:03:18,460 Uh, here are his his, my two apps and some fun things from another course. 70 00:03:18,900 --> 00:03:25,340 Uh, and, uh, then you come along here and you go to settings. 71 00:03:25,620 --> 00:03:31,660 So you're looking at your avatar page, you press settings and you go to API to access tokens. 72 00:03:31,700 --> 00:03:32,580 Access tokens. 73 00:03:32,580 --> 00:03:34,540 This page here, look at this. 74 00:03:34,820 --> 00:03:36,980 Now here is the thing. 75 00:03:37,060 --> 00:03:37,740 Here is the thing. 76 00:03:37,740 --> 00:03:38,420 Listen up. 77 00:03:38,700 --> 00:03:40,700 You press create new token. 78 00:03:40,940 --> 00:03:42,180 Here comes the new token. 79 00:03:42,180 --> 00:03:43,980 You see this page for the new token. 80 00:03:44,020 --> 00:03:47,060 The next thing you do is you find this word right. 81 00:03:47,060 --> 00:03:50,420 You need to press that word right and give it a token name. 82 00:03:50,420 --> 00:03:51,740 Give it whatever token name you like. 83 00:03:51,740 --> 00:03:53,900 This can be, you know, my my favorite token. 84 00:03:53,900 --> 00:03:56,060 And then you press Create token. 85 00:03:56,100 --> 00:04:00,940 Now the reason you need to write permissions is first of all that gives you access to download models 86 00:04:00,940 --> 00:04:04,940 from from any repo which you'll need when we download open source models. 87 00:04:04,940 --> 00:04:08,820 But also we are going to be writing our own models at future weeks. 88 00:04:08,860 --> 00:04:11,500 And if you don't have this permission, it will cause lots of troubles later. 89 00:04:11,500 --> 00:04:13,100 So do have write permissions. 90 00:04:13,340 --> 00:04:17,020 Give it a name, create the token and copy it into your clipboard. 91 00:04:17,020 --> 00:04:21,660 It should start h underscore and that should now be in your clipboard. 92 00:04:21,660 --> 00:04:24,300 And now we go back to Colab okay. 93 00:04:24,340 --> 00:04:30,540 And in Colab you now go to the left here to this little key, and you press the key and it opens up 94 00:04:30,540 --> 00:04:38,060 this thing that says secrets and secrets are they are just like the dot env file. 95 00:04:38,060 --> 00:04:44,220 They have a name and a value, just like the dot env file has something equals something. 96 00:04:44,460 --> 00:04:49,060 And the name for hugging face is underscore token. 97 00:04:49,220 --> 00:04:56,330 And in that value cell you paste the the lowercase HF underscore and that key. 98 00:04:56,450 --> 00:05:02,410 And you put in exactly as it is with no spaces and no nothing random on the beginning or end, just 99 00:05:02,410 --> 00:05:07,210 as it should be, so that you have that in there and you make sure it's switched on here for notebook 100 00:05:07,290 --> 00:05:12,570 access, which means that this notebook here will have access to that secret. 101 00:05:12,730 --> 00:05:15,130 And once you've done that, you're golden. 102 00:05:15,130 --> 00:05:16,890 You've done the fiddly part. 103 00:05:16,930 --> 00:05:22,010 You can now access this hugging face token from your code, and it even tells you right there that it's 104 00:05:22,010 --> 00:05:29,050 super simple over here to have access to any of these keys with code that looks a bit like that, but 105 00:05:29,090 --> 00:05:34,210 instead of secret name, you would put this word here, that word would go there, and then you'd be 106 00:05:34,210 --> 00:05:36,730 able to access this value. 107 00:05:36,890 --> 00:05:40,090 It's just a way to have secrets stored separately. 108 00:05:40,090 --> 00:05:43,850 And these secrets here are associated with your Google account. 109 00:05:43,850 --> 00:05:47,610 If you share this with someone else, they won't get access to your secrets. 110 00:05:47,610 --> 00:05:50,170 The secrets, of course, are secret to you. 111 00:05:50,170 --> 00:05:54,890 And as an added benefit, anytime you're logged in, you'll have access to all of these secrets, which 112 00:05:54,890 --> 00:05:56,410 makes it super convenient. 113 00:05:56,410 --> 00:06:02,290 It's like a EMV file that's everywhere on Google Colab for you and only for you when you're logged in. 114 00:06:02,930 --> 00:06:04,370 So that's the fiddly part. 115 00:06:04,410 --> 00:06:05,410 Done, sorted. 116 00:06:05,410 --> 00:06:06,490 You've got it set up. 117 00:06:06,530 --> 00:06:07,330 Brilliant. 118 00:06:07,730 --> 00:06:08,650 Let's go on. 119 00:06:08,850 --> 00:06:09,490 Okay. 120 00:06:10,050 --> 00:06:11,010 Game time. 121 00:06:11,250 --> 00:06:12,530 You've filled in your secrets. 122 00:06:12,530 --> 00:06:14,130 You've connected to a T4. 123 00:06:14,370 --> 00:06:16,490 You're looking all good there. 124 00:06:16,690 --> 00:06:18,010 Scroll down. 125 00:06:18,170 --> 00:06:23,650 So first of all, this is exactly the code that you may remember you saw just here. 126 00:06:23,850 --> 00:06:28,650 It's the code which is going to allow us to access our HRF token. 127 00:06:28,650 --> 00:06:33,890 And it's going into a something called HRF token, a variable called that. 128 00:06:34,050 --> 00:06:37,530 And that's coming from Google Colab importing user data. 129 00:06:37,530 --> 00:06:40,050 So user data is basically the secrets on the left. 130 00:06:40,050 --> 00:06:42,130 We're getting HRF token and there it is. 131 00:06:42,170 --> 00:06:45,770 And then we're calling something called login with HRF token. 132 00:06:45,770 --> 00:06:46,890 And what is that? 133 00:06:46,890 --> 00:06:52,040 That that is a function from a package called Huggingface hub. 134 00:06:52,200 --> 00:06:55,280 And this is the first of those libraries I told you about. 135 00:06:55,280 --> 00:07:00,360 This is the Hugging face library that's related to connecting to the Hugging Face Hub. 136 00:07:00,400 --> 00:07:01,320 Okay, let's do it. 137 00:07:01,320 --> 00:07:02,240 Let's run this. 138 00:07:02,440 --> 00:07:06,600 And if all is well, if you've done the key properly, if you've copied and pasted, if it has the right 139 00:07:06,600 --> 00:07:09,680 permissions and everything, you'll have a green tick there. 140 00:07:09,880 --> 00:07:12,760 And this running on the cloud is now connected. 141 00:07:12,760 --> 00:07:14,800 It's logged in to Huggingface. 142 00:07:15,120 --> 00:07:21,400 So what I'm now going to do is run a piece of code that I'm not going to explain how it works, because 143 00:07:21,400 --> 00:07:23,000 we'll be dealing with that another time. 144 00:07:23,000 --> 00:07:25,000 It's using something called pipelines. 145 00:07:25,280 --> 00:07:27,640 And we are going to be first. 146 00:07:27,680 --> 00:07:31,000 We are going to be, uh, let's see what order do I want to do these in? 147 00:07:31,040 --> 00:07:31,520 Yeah okay. 148 00:07:31,520 --> 00:07:32,080 Okay. 149 00:07:32,120 --> 00:07:37,640 So first of all, we are going to be using, uh, something that isn't going to be using a model, a 150 00:07:37,640 --> 00:07:39,480 model from the Hugging Face library. 151 00:07:39,560 --> 00:07:41,640 But it is not a transformer. 152 00:07:41,640 --> 00:07:43,240 It is actually a different kind of model. 153 00:07:43,240 --> 00:07:46,240 It is a diffusion model that you may have heard of. 154 00:07:46,280 --> 00:07:49,600 It is a kind of model that's good at generating images. 155 00:07:49,600 --> 00:07:56,240 And we're making one from a company called stability, uh, which is, uh, the, uh, the, the stability 156 00:07:56,280 --> 00:07:59,640 diffusion model, and it's called Excel Turbo. 157 00:07:59,880 --> 00:08:08,640 And we are going to ask it to generate a class of students learning AI engineering in a vibrant pop 158 00:08:08,680 --> 00:08:09,880 art style. 159 00:08:10,040 --> 00:08:11,520 So that is our prompt. 160 00:08:11,760 --> 00:08:15,040 We are using something called pipelines that we will tackle. 161 00:08:15,040 --> 00:08:21,720 We'll actually explain this tomorrow, but I'm doing it because this is now hitting our our T4. 162 00:08:22,040 --> 00:08:25,040 So we should start to see work happening down here. 163 00:08:25,040 --> 00:08:30,080 It says it's been executing for 22 seconds on the T4 and things are happening. 164 00:08:30,080 --> 00:08:30,280 Whoa. 165 00:08:30,320 --> 00:08:30,880 Look at all this. 166 00:08:30,880 --> 00:08:32,400 Lots of status things. 167 00:08:32,400 --> 00:08:40,320 So a lot of the time that you spend on Colab is actually spent downloading stuff, because the downside 168 00:08:40,320 --> 00:08:46,400 of it being a completely clean box every time is that it doesn't cache any files you're downloading 169 00:08:46,400 --> 00:08:49,480 things from the Hugging Face Hub every single time. 170 00:08:49,640 --> 00:08:51,720 So all of this is downloading. 171 00:08:51,720 --> 00:08:57,000 You can see here it says that's almost a 1.4GB of data coming there. 172 00:08:57,240 --> 00:09:03,400 Uh, and uh, well that's, that's that's five gigabytes of data associated with this model. 173 00:09:03,400 --> 00:09:05,440 And that's going to take a minute to download. 174 00:09:05,480 --> 00:09:11,560 You can see the system memories coming up, but so far the GPU memory hasn't really moved at all. 175 00:09:11,720 --> 00:09:15,160 Uh, and this is looking like we're filling a lot of CPU. 176 00:09:15,360 --> 00:09:16,520 But fear not. 177 00:09:16,560 --> 00:09:21,920 Once this thing is loaded, it's going to slam over to the GPU and we're going to run against the GPU. 178 00:09:21,960 --> 00:09:24,840 Now hopefully you're doing this too because it's free, so why not? 179 00:09:25,080 --> 00:09:26,960 So you are currently running this. 180 00:09:27,000 --> 00:09:28,280 You press shift enter. 181 00:09:28,320 --> 00:09:35,600 You are seeing the stability AI model running for a class of students learning AI engineering in a vibrant 182 00:09:35,600 --> 00:09:36,840 pop art style. 183 00:09:37,280 --> 00:09:40,120 All right, I'm gonna I'm gonna vanish for a second. 184 00:09:40,160 --> 00:09:41,080 It'll be a few minutes for me. 185 00:09:41,080 --> 00:09:42,400 It'll be a second for you. 186 00:09:42,400 --> 00:09:45,080 And when I'm back, we'll see what comes of this.