1 00:00:00,000 --> 00:00:04,000 So guys, we are going to continue the generative AI on cloud series. 2 00:00:04,000 --> 00:00:10,000 And in this video I'm going to probably discuss about the gen AI project lifecycle. 3 00:00:10,000 --> 00:00:16,000 Now, since you already know that, we are definitely going to develop a lot of applications specifically 4 00:00:16,000 --> 00:00:20,000 in cloud from the basic data ingestion till the deployment. 5 00:00:20,000 --> 00:00:28,000 So it is very much necessary that you actually understand a generic workflow of the gen AI project lifecycle. 6 00:00:28,000 --> 00:00:34,000 So let me quickly go ahead and let me go ahead and write some amazing things for you in this particular 7 00:00:34,000 --> 00:00:41,000 notebook, and I will be explaining you completely, step by step, how you can probably see or follow 8 00:00:41,000 --> 00:00:42,000 a project life cycle. 9 00:00:43,000 --> 00:00:48,000 And uh, as we go ahead, there will be a lot many things that will be coming, like LM ops platform. 10 00:00:48,000 --> 00:00:56,000 Uh, and uh, we will be working specifically with Azure, uh, AI studio, AWS SageMaker studio and 11 00:00:56,000 --> 00:00:56,000 all. 12 00:00:56,000 --> 00:00:59,000 So definitely both the clouds will get covered. 13 00:00:59,000 --> 00:01:04,000 So before I go ahead, please make sure that you keep the like target of all this kind of videos till 14 00:01:04,000 --> 00:01:04,000 1000. 15 00:01:04,000 --> 00:01:06,000 That will definitely motivate me. 16 00:01:06,000 --> 00:01:11,000 And I've been exploring many more things so that you get the right kind of guidance and knowledge. 17 00:01:12,000 --> 00:01:16,000 So let me go ahead and let me start the gen AI project lifecycle. 18 00:01:17,000 --> 00:01:23,000 Uh, with respect to this gen AI project life cycle, I would like to make this entire life cycle into 19 00:01:23,000 --> 00:01:24,000 4 to 5 steps. 20 00:01:24,000 --> 00:01:25,000 Okay. 21 00:01:25,000 --> 00:01:28,000 The first step, that is nothing. 22 00:01:28,000 --> 00:01:37,000 But here I'm just going to write over here it is basically defining the use case okay. 23 00:01:37,000 --> 00:01:40,000 So what kind of use case are you solving. 24 00:01:40,000 --> 00:01:44,000 Then this use cases can be a drag application can be a text summarization. 25 00:01:44,000 --> 00:01:46,000 Application can be a chat bot. 26 00:01:46,000 --> 00:01:53,000 So based on different different use cases that actually depends on your requirements your company requirements. 27 00:01:53,000 --> 00:01:54,000 So this is the first step. 28 00:01:54,000 --> 00:01:57,000 You really need to define the use case that you're specifically doing. 29 00:01:57,000 --> 00:02:05,000 Now with respect to this particular use case we usually take this entire module into the scope part 30 00:02:05,000 --> 00:02:05,000 okay. 31 00:02:05,000 --> 00:02:08,000 So this is basically my scope, right? 32 00:02:08,000 --> 00:02:10,000 If I basically use a generic term. 33 00:02:10,000 --> 00:02:16,000 Now once you define a use case, you let's say that I am going to probably develop a Rag application 34 00:02:16,000 --> 00:02:18,000 in that I'm going to definitely use vector databases. 35 00:02:18,000 --> 00:02:21,000 I may have a lot of PDF files. 36 00:02:21,000 --> 00:02:26,000 I also need to probably convert that into a vectors and store it in some kind of vector store DB. 37 00:02:26,000 --> 00:02:31,000 So some kind of use case you really need to define and all the requirements that is required in that 38 00:02:31,000 --> 00:02:33,000 particular use case. 39 00:02:33,000 --> 00:02:40,000 Coming to the next step, coming to the next step, which is super important because this steps will 40 00:02:40,000 --> 00:02:43,000 be involving two important things okay. 41 00:02:43,000 --> 00:02:46,000 And that is nothing but choosing. 42 00:02:48,000 --> 00:02:53,000 Choosing the right model okay. 43 00:02:53,000 --> 00:02:55,000 The right model. 44 00:02:56,000 --> 00:03:02,000 When I say choosing the right model, here they are two different things that you can probably split 45 00:03:02,000 --> 00:03:03,000 this into. 46 00:03:04,000 --> 00:03:12,000 One, whether you are using or you are probably using some kind of foundation models. 47 00:03:12,000 --> 00:03:14,000 So here I'm going to probably right. 48 00:03:14,000 --> 00:03:19,000 Whether you are using foundation model and solving a use case. 49 00:03:19,000 --> 00:03:24,000 So this is the one category that I would like to divide this particular module into. 50 00:03:24,000 --> 00:03:32,000 The other category is that whether you want to build your own custom LLM right custom LLM. 51 00:03:32,000 --> 00:03:34,000 Now see there are two things over here. 52 00:03:34,000 --> 00:03:35,000 Right? 53 00:03:35,000 --> 00:03:37,000 When I say foundation model. 54 00:03:37,000 --> 00:03:41,000 Foundation models are already those larger models like OpenAI llama two. 55 00:03:41,000 --> 00:03:42,000 Llama three. 56 00:03:42,000 --> 00:03:43,000 Right. 57 00:03:43,000 --> 00:03:44,000 You have Google Gemini Pro. 58 00:03:44,000 --> 00:03:47,000 So these all are very huge foundation models. 59 00:03:47,000 --> 00:03:52,000 And first, for most of the generic use cases, you can directly use those kind of foundation models 60 00:03:52,000 --> 00:03:54,000 and you can solve the use case itself. 61 00:03:54,000 --> 00:04:01,000 Right now with respect to this foundation models, we can also further go ahead and do fine tuning. 62 00:04:01,000 --> 00:04:02,000 Right. 63 00:04:03,000 --> 00:04:08,000 So let's say I have a fine I have a foundation model which I am specifically using to solve my business 64 00:04:08,000 --> 00:04:09,000 use cases. 65 00:04:09,000 --> 00:04:15,000 On top of that, if I really want to make this foundation model behave well for my own custom data, 66 00:04:15,000 --> 00:04:20,000 then what I can do on top of this foundation model, I can use Laura Laura techniques and I can probably 67 00:04:20,000 --> 00:04:22,000 fine tune all this kind of models. 68 00:04:22,000 --> 00:04:27,000 Okay, so this is one of the step, the second step, the second step that I have written over here 69 00:04:27,000 --> 00:04:28,000 as custom LM. 70 00:04:28,000 --> 00:04:33,000 Custom LM is nothing, but it is building your LM from scratch. 71 00:04:33,000 --> 00:04:36,000 Building your lm from scratch. 72 00:04:36,000 --> 00:04:38,000 From scratch. 73 00:04:38,000 --> 00:04:39,000 Right. 74 00:04:39,000 --> 00:04:43,000 And obviously, uh, this one, there's a lot of benefit. 75 00:04:43,000 --> 00:04:49,000 Also, if a company is building an LM model completely from scratch for this specific use cases, but 76 00:04:49,000 --> 00:04:52,000 a lot of resources will definitely be required. 77 00:04:52,000 --> 00:04:56,000 We have to really take care of model hallucination, many things and all as we go ahead. 78 00:04:56,000 --> 00:05:01,000 But yes, uh, I've also seen many, many companies developing their own custom LM model. 79 00:05:01,000 --> 00:05:02,000 Okay. 80 00:05:02,000 --> 00:05:08,000 So choosing the right model or what kind of models you're specifically using to solve this particular 81 00:05:08,000 --> 00:05:14,000 use cases, that becomes the second important module with respect to this gen AI project life cycle. 82 00:05:14,000 --> 00:05:18,000 And obviously I've spoken about foundation models both in AWS. 83 00:05:18,000 --> 00:05:22,000 You have in Google, you have in Microsoft Azure, you have currently Microsoft Azure. 84 00:05:23,000 --> 00:05:26,000 How AI studio specifically have all the access of open AI services. 85 00:05:27,000 --> 00:05:29,000 Obviously it is investing a huge amount of money over there. 86 00:05:29,000 --> 00:05:35,000 Okay, now once you probably select the right kind of model, there are main three tasks that you probably 87 00:05:35,000 --> 00:05:36,000 do for going forward. 88 00:05:36,000 --> 00:05:39,000 Okay, so main three tasks okay. 89 00:05:39,000 --> 00:05:48,000 The first task is nothing, but you can specifically use prompt engineering and solve a use case Prompt 90 00:05:48,000 --> 00:05:49,000 engineering and solve a use case. 91 00:05:49,000 --> 00:05:55,000 The second task that you actually can do is nothing but fine tuning, right? 92 00:05:55,000 --> 00:05:58,000 Fine tuning, fine tuning. 93 00:05:58,000 --> 00:06:02,000 So with the help of fine tuning, also, you can probably develop your own custom LM model. 94 00:06:02,000 --> 00:06:05,000 And on top of that you can basically do it. 95 00:06:05,000 --> 00:06:08,000 Let's say you're completely creating your LM model from scratch. 96 00:06:08,000 --> 00:06:14,000 One more important mechanism that you have is nothing but, uh, aligning. 97 00:06:14,000 --> 00:06:18,000 Uh, or you can probably say training with human feedback. 98 00:06:19,000 --> 00:06:21,000 Training with human feedback. 99 00:06:21,000 --> 00:06:28,000 And this is one of the very important step that is actually used while you are training your LM models. 100 00:06:28,000 --> 00:06:30,000 How are LM model is basically trained. 101 00:06:30,000 --> 00:06:32,000 I've already created a video in my playlist. 102 00:06:32,000 --> 00:06:37,000 Uh, with respect to the link chain and all generative AI playlist, you can probably go head over there 103 00:06:37,000 --> 00:06:40,000 fine tuning, how to specifically do fine tuning and all that. 104 00:06:40,000 --> 00:06:45,000 Also, I've actually shown you the reason why I'm showing you this generative AI project life cycle. 105 00:06:45,000 --> 00:06:50,000 Because tomorrow when I'm probably creating videos, um, now let's say in the upcoming videos, when 106 00:06:50,000 --> 00:06:55,000 I'm creating videos related to this series over there, you'll be seeing all this particular steps going 107 00:06:55,000 --> 00:06:55,000 ahead. 108 00:06:55,000 --> 00:06:56,000 Okay. 109 00:06:56,000 --> 00:07:01,000 Now, once you probably do all the steps, uh, the further step is something called as evaluation. 110 00:07:02,000 --> 00:07:02,000 Okay. 111 00:07:02,000 --> 00:07:08,000 Evaluation is basically seeing that how your model is performing by performing all this particular steps. 112 00:07:08,000 --> 00:07:13,000 There are also different different performance metrics which we are probably going to follow. 113 00:07:13,000 --> 00:07:13,000 Okay. 114 00:07:13,000 --> 00:07:19,000 This two steps I would like to combine and say something like this okay. 115 00:07:19,000 --> 00:07:24,000 So I'll say probably adapt and align models okay. 116 00:07:24,000 --> 00:07:29,000 So this will be the specific model that, uh, we specifically use for this purpose. 117 00:07:29,000 --> 00:07:32,000 Now over here, your model will be ready. 118 00:07:32,000 --> 00:07:33,000 Everything is perfect. 119 00:07:33,000 --> 00:07:36,000 Or you are able to solve the use cases. 120 00:07:36,000 --> 00:07:38,000 Let's say your performance metrics is increasing over here. 121 00:07:38,000 --> 00:07:41,000 So your metrics is specifically increasing. 122 00:07:41,000 --> 00:07:43,000 And it is saying that now your model is ready. 123 00:07:43,000 --> 00:07:45,000 Now it comes to the deployment part. 124 00:07:45,000 --> 00:07:46,000 Right. 125 00:07:46,000 --> 00:07:52,000 So with respect to the deployment part I would definitely say deployment. 126 00:07:52,000 --> 00:07:57,000 Uh, further, you also need to do a lot of integration with different different applications. 127 00:07:57,000 --> 00:08:01,000 So I will probably say application integration okay. 128 00:08:02,000 --> 00:08:04,000 Application integration. 129 00:08:04,000 --> 00:08:17,000 And here uh, what we do we specifically perform two major step one we optimize models Okay, uh, I'll 130 00:08:17,000 --> 00:08:20,000 just write, optimize and deploy models. 131 00:08:23,000 --> 00:08:25,000 Optimize and deploy models. 132 00:08:25,000 --> 00:08:25,000 Okay. 133 00:08:25,000 --> 00:08:30,000 And this deployment is specifically done for inferencing. 134 00:08:30,000 --> 00:08:31,000 Okay. 135 00:08:31,000 --> 00:08:35,000 And here is where most of your cloud platforms. 136 00:08:36,000 --> 00:08:39,000 Here is where your LM ops. 137 00:08:40,000 --> 00:08:42,000 LM ops is used. 138 00:08:42,000 --> 00:08:45,000 You know, different, different inferencing techniques are there. 139 00:08:45,000 --> 00:08:49,000 One technique I've already covered with respect to a platform, which is called as grok. 140 00:08:49,000 --> 00:08:50,000 Right. 141 00:08:50,000 --> 00:08:53,000 It uses a inferencing technique which is called as LP. 142 00:08:53,000 --> 00:08:58,000 So it is always good idea that you should definitely know multiple ways of inferencing. 143 00:08:58,000 --> 00:09:02,000 See, at the end of the day, whatever models you specifically create, unless and until the inferencing 144 00:09:02,000 --> 00:09:05,000 is not fast, you definitely cannot use those things, right? 145 00:09:05,000 --> 00:09:13,000 So it is very much necessary that you know the idea of this module extensively, because tomorrow building 146 00:09:13,000 --> 00:09:15,000 all these things is very easy. 147 00:09:15,000 --> 00:09:16,000 Fine tuning is very easy, right? 148 00:09:16,000 --> 00:09:20,000 You definitely have a template, a framework, a data set, preparation and all. 149 00:09:20,000 --> 00:09:22,000 And you can perform this particular step. 150 00:09:23,000 --> 00:09:27,000 So that is the reason in this series of videos you'll be seeing that how much I will be focusing on 151 00:09:27,000 --> 00:09:31,000 this lot of um, uh, LMS platform. 152 00:09:31,000 --> 00:09:31,000 Right. 153 00:09:31,000 --> 00:09:36,000 And I will also show you multiple platforms, like, which can definitely make your inferencing very 154 00:09:36,000 --> 00:09:37,000 much good. 155 00:09:37,000 --> 00:09:40,000 So this is the most important thing here. 156 00:09:40,000 --> 00:09:41,000 Uh, definitely. 157 00:09:41,000 --> 00:09:44,000 We'll be using AWS Azure. 158 00:09:44,000 --> 00:09:44,000 Right. 159 00:09:44,000 --> 00:09:46,000 You can use all these things GCP. 160 00:09:47,000 --> 00:09:52,000 And we'll see what all services they have specifically provided probably for the inferencing purpose 161 00:09:52,000 --> 00:09:53,000 again. 162 00:09:53,000 --> 00:09:56,000 But initially our focus will definitely be on AWS. 163 00:09:56,000 --> 00:09:57,000 Okay. 164 00:09:57,000 --> 00:09:59,000 Then the second step. 165 00:09:59,000 --> 00:10:05,000 Ah, um, after we do the deployment, uh, in the application integration, integration, what we do 166 00:10:05,000 --> 00:10:13,000 next is that we build, uh, LLM powered application, LLM powered application powered application. 167 00:10:13,000 --> 00:10:16,000 Because your integration is done, your API is created. 168 00:10:16,000 --> 00:10:17,000 Now. 169 00:10:17,000 --> 00:10:20,000 It's all how well you can actually build the solutions. 170 00:10:20,000 --> 00:10:22,000 You can solve different different use cases and all. 171 00:10:22,000 --> 00:10:27,000 So this overall gives a brief idea about the entire AI project life cycle. 172 00:10:28,000 --> 00:10:31,000 Um, since, uh, we have already started this journey on cloud. 173 00:10:31,000 --> 00:10:35,000 So this is necessary to know and uh, you should probably follow all the steps. 174 00:10:35,000 --> 00:10:41,000 And whenever I create any videos with respect to any AI on AWS, all these steps will be considered 175 00:10:41,000 --> 00:10:43,000 in mind and it will be shown to you.