1 00:00:00,000 --> 00:00:05,000 So guys, in this particular video we are going to discuss an end to end LM project with the help of 2 00:00:05,000 --> 00:00:07,000 AWS, bedrock and Lang Chain. 3 00:00:07,000 --> 00:00:11,000 This was also one of the most requested video from many people out there. 4 00:00:11,000 --> 00:00:15,000 What we are going to do in this particular project is that we are going to develop a document, Q and 5 00:00:15,000 --> 00:00:21,000 a application where specifically will be harnessing multiple models that are provided by AWS bedrock 6 00:00:21,000 --> 00:00:23,000 like Cloudy Llama two. 7 00:00:23,000 --> 00:00:29,000 And you can also use Amazon Titan, whatever models you specifically want that actually it provides. 8 00:00:29,000 --> 00:00:33,000 We can actually use them and we can implement this entire application. 9 00:00:33,000 --> 00:00:38,000 So to go ahead with, I will just go ahead and show you the quick demo over here. 10 00:00:38,000 --> 00:00:40,000 So what is this specific entire application. 11 00:00:40,000 --> 00:00:41,000 This is a rack right? 12 00:00:41,000 --> 00:00:46,000 We are trying to create a rack system wherein we have multiple PDFs. 13 00:00:46,000 --> 00:00:50,000 All those PDFs is stored in the form of vector embeddings inside vector store. 14 00:00:50,000 --> 00:00:57,000 And whenever we ask any query we will be able to harness the power of AWS bedrock, which provides different 15 00:00:57,000 --> 00:00:59,000 different models like cloudy, Llama two. 16 00:00:59,000 --> 00:00:59,000 And uh. 17 00:00:59,000 --> 00:01:03,000 Based on this particular question, we can retrieve that question from the PDF files. 18 00:01:03,000 --> 00:01:06,000 So let's say I'm asking what are Transformers? 19 00:01:06,000 --> 00:01:12,000 And if I go ahead and click on cloudy Output that basically is going to hit the cloudy API cloudy model. 20 00:01:12,000 --> 00:01:14,000 And I'm going to get the entire response okay. 21 00:01:14,000 --> 00:01:20,000 So currently this is running If you probably want llama two output, I can also go ahead and click on 22 00:01:20,000 --> 00:01:21,000 llama two. 23 00:01:21,000 --> 00:01:24,000 And I can actually get that specific response right. 24 00:01:24,000 --> 00:01:30,000 So whatever questions you you have with respect to this particular PDF documents what the set of PDF 25 00:01:30,000 --> 00:01:34,000 documents also will show you, and you will be able to get the entire response over here. 26 00:01:34,000 --> 00:01:34,000 Right? 27 00:01:34,000 --> 00:01:37,000 So, uh, why it is taking some time? 28 00:01:37,000 --> 00:01:42,000 Because I still need to write that optimized version of the code, but once it runs right for the first 29 00:01:42,000 --> 00:01:45,000 time, I think then after that it works absolutely fine. 30 00:01:45,000 --> 00:01:50,000 Now let me just go ahead and write what is YOLO, right? 31 00:01:50,000 --> 00:01:57,000 So if I go ahead and click on llama two output before see I showed you also llama two, uh, in my local 32 00:01:57,000 --> 00:01:58,000 system itself over there. 33 00:01:58,000 --> 00:02:01,000 It was taking a lot of time, but this is directly coming from the API model itself. 34 00:02:01,000 --> 00:02:05,000 So here is what is the response I am specifically getting. 35 00:02:05,000 --> 00:02:06,000 And I'll be able to get the answers. 36 00:02:06,000 --> 00:02:10,000 Now this is what I'm going to develop completely from scratch already. 37 00:02:10,000 --> 00:02:15,000 If you remember guys in our previous session in our yesterday, I had uploaded a video where I showed 38 00:02:15,000 --> 00:02:18,000 you the power of cloudy dot pi, how to invoke the model and all. 39 00:02:19,000 --> 00:02:21,000 Uh, then we also saw with respect to llama two and all. 40 00:02:21,000 --> 00:02:25,000 Now let me just go ahead and create a new file over here. 41 00:02:25,000 --> 00:02:28,000 So let me just go ahead and write this as app dot Pi. 42 00:02:28,000 --> 00:02:32,000 Now inside this app dot pi I'm going to write my entire code. 43 00:02:32,000 --> 00:02:38,000 Now before I go ahead, I really need to do some installation with respect to some of the requirements. 44 00:02:38,000 --> 00:02:45,000 As said, uh, here we are going to use not only Boto3 and aws CLI, we are also going to use some more 45 00:02:45,000 --> 00:02:47,000 libraries like pi PDF. 46 00:02:47,000 --> 00:02:50,000 So let me just go ahead and write this over here. 47 00:02:50,000 --> 00:02:55,000 So we require pi PDF along with pi PDF we We also need lang chain. 48 00:02:56,000 --> 00:03:01,000 And then we will also be requiring Streamlit write uh. 49 00:03:01,000 --> 00:03:06,000 Along with this we also require fierce fire CPU. 50 00:03:06,000 --> 00:03:07,000 Right. 51 00:03:07,000 --> 00:03:11,000 So this is what we are going to use because from our local environment we are going to do the vector 52 00:03:11,000 --> 00:03:12,000 embeddings. 53 00:03:12,000 --> 00:03:15,000 So for this I'm going to use the fierce DB right. 54 00:03:15,000 --> 00:03:19,000 So through this only we'll be able to create the vector embedding into the vector store. 55 00:03:19,000 --> 00:03:22,000 So these are some of the basic requirements that I actually require. 56 00:03:22,000 --> 00:03:27,000 Now uh aws cli is also there because we really need to configure it. 57 00:03:27,000 --> 00:03:30,000 Now let me quickly go ahead and open the terminal. 58 00:03:30,000 --> 00:03:33,000 Let me clear the screen because yesterday I was creating the images over here. 59 00:03:33,000 --> 00:03:37,000 The first step again you have to create a virtual environment. 60 00:03:37,000 --> 00:03:38,000 Do all this pip install. 61 00:03:38,000 --> 00:03:42,000 And all right I hope I don't have to tell you all those things now because I have repeated many number 62 00:03:42,000 --> 00:03:43,000 of times. 63 00:03:43,000 --> 00:03:47,000 So let me go ahead and write pip install minus our requirement dot txt. 64 00:03:47,000 --> 00:03:51,000 If you are not able to understand it guys, please make sure that I'll be providing you this entire 65 00:03:51,000 --> 00:03:53,000 playlist in the description of this particular video. 66 00:03:53,000 --> 00:03:55,000 You can go ahead and watch it. 67 00:03:55,000 --> 00:03:58,000 Okay, but uh, you have to follow the playlist, right? 68 00:03:58,000 --> 00:04:02,000 Again, I'm not going to repeat everything from scratch, otherwise it is just a waste of time. 69 00:04:02,000 --> 00:04:03,000 Right? 70 00:04:03,000 --> 00:04:07,000 So till this installation is basically taking place, let me just go ahead and talk about what all things 71 00:04:07,000 --> 00:04:09,000 we are going to basically develop. 72 00:04:09,000 --> 00:04:09,000 Right. 73 00:04:09,000 --> 00:04:11,000 So this is the document Q&A search. 74 00:04:11,000 --> 00:04:16,000 So in short my PDF will be stored in a stored in the form of my entire PDFs. 75 00:04:16,000 --> 00:04:18,000 We probably consider this. 76 00:04:18,000 --> 00:04:21,000 My entire PDF will be stored in vector store. 77 00:04:21,000 --> 00:04:26,000 And from this vector store we can specifically query any information that we want. 78 00:04:26,000 --> 00:04:30,000 Along with this, we are going to harness the power of Lang Chen. 79 00:04:30,000 --> 00:04:36,000 Chin Lang Chin along with the LM models from AWS bedrock. 80 00:04:36,000 --> 00:04:39,000 Okay, so this is what we are going to do in this project. 81 00:04:39,000 --> 00:04:44,000 So please make sure that you prepare well, uh, understand the architecture, what we are trying to 82 00:04:44,000 --> 00:04:44,000 do. 83 00:04:44,000 --> 00:04:49,000 So our entire project deals with two important steps one, two. 84 00:04:49,000 --> 00:04:55,000 And before this, we are also going to do a very important step, which is called as data ingestion 85 00:04:55,000 --> 00:05:01,000 Right now in data ingestion, what we are going to do is that we are going to read from the entire folder 86 00:05:01,000 --> 00:05:03,000 how many PDFs are there, all the PDFs. 87 00:05:03,000 --> 00:05:04,000 We are going to read it. 88 00:05:04,000 --> 00:05:05,000 Okay. 89 00:05:05,000 --> 00:05:10,000 So once I probably get this entire PDF, then my next step actually starts over here. 90 00:05:11,000 --> 00:05:16,000 I'm going to take all those documents, the PDFs over there, split this into chunks, create the embeddings. 91 00:05:16,000 --> 00:05:19,000 And here we are going to use this for this database. 92 00:05:19,000 --> 00:05:23,000 So here we are going to use this fast right. 93 00:05:23,000 --> 00:05:25,000 The vector database. 94 00:05:25,000 --> 00:05:28,000 And through that only we will try to create this vector store. 95 00:05:28,000 --> 00:05:28,000 Right. 96 00:05:28,000 --> 00:05:31,000 So step by step I'll show you how to do this. 97 00:05:31,000 --> 00:05:32,000 What embeddings. 98 00:05:32,000 --> 00:05:34,000 Also we are going to specifically use here. 99 00:05:34,000 --> 00:05:36,000 See understand we are going to create the embeddings. 100 00:05:36,000 --> 00:05:43,000 So for embeddings we are going to use today I will show you a new model which is basically called as 101 00:05:43,000 --> 00:05:44,000 Amazon Titan. 102 00:05:45,000 --> 00:05:45,000 Okay. 103 00:05:45,000 --> 00:05:48,000 So that you understand you have multiple options. 104 00:05:48,000 --> 00:05:51,000 You have you may have implemented multiple options with respect to that. 105 00:05:51,000 --> 00:05:56,000 Because tomorrow in companies if you go definitely they are going to use this AWS bedrock because it 106 00:05:56,000 --> 00:05:58,000 provides a lot of features, right. 107 00:05:59,000 --> 00:06:02,000 So for creating this embeddings I will specifically use Amazon Titan. 108 00:06:03,000 --> 00:06:06,000 Other than this, if you don't want to use this, you can also use OpenAI embeddings. 109 00:06:06,000 --> 00:06:06,000 Right? 110 00:06:06,000 --> 00:06:07,000 It is up to you. 111 00:06:07,000 --> 00:06:12,000 I've also shown with Google generative AI embeddings, multiple embeddings techniques in my playlist 112 00:06:12,000 --> 00:06:14,000 with respect to Google, Gemini and all. 113 00:06:14,000 --> 00:06:19,000 So after we create this vector store, now how we are going to use the LM. 114 00:06:19,000 --> 00:06:24,000 So in the second step, whenever we ask any question, first of all the similarity search will happen 115 00:06:24,000 --> 00:06:29,000 from the vector store whatever relevant documents and chunks we get. 116 00:06:29,000 --> 00:06:30,000 We have to take this chunks. 117 00:06:30,000 --> 00:06:33,000 Give it to my LM model along with the prompt. 118 00:06:34,000 --> 00:06:35,000 Like let's say that I say that, okay. 119 00:06:35,000 --> 00:06:40,000 Please summarize this entire information based on the query that I've asked in 250 words. 120 00:06:40,000 --> 00:06:45,000 So this LLM model along with this prompt is going to take this particular data and is going to give 121 00:06:45,000 --> 00:06:46,000 the answer itself. 122 00:06:46,000 --> 00:06:46,000 Right. 123 00:06:46,000 --> 00:06:51,000 So this both I'm going to develop completely from scratch step by step. 124 00:06:51,000 --> 00:06:52,000 We'll try to see it okay. 125 00:06:52,000 --> 00:06:57,000 So uh let's quickly go ahead and let's start our coding without wasting any time. 126 00:06:57,000 --> 00:06:58,000 Okay. 127 00:06:58,000 --> 00:07:00,000 So here is my app dot Pi. 128 00:07:00,000 --> 00:07:02,000 I said that I'm going to do it completely from scratch. 129 00:07:02,000 --> 00:07:05,000 So um, there will be some installation. 130 00:07:05,000 --> 00:07:08,000 There will be some, uh, errors that may be probably coming up. 131 00:07:08,000 --> 00:07:14,000 So we will try to also import all the libraries based on the pipeline that we have created, based on 132 00:07:14,000 --> 00:07:17,000 the steps or architectures that we have already discussed. 133 00:07:17,000 --> 00:07:22,000 So first of all, quickly let's go ahead and import JSON. 134 00:07:23,000 --> 00:07:25,000 So I'm going to import JSON. 135 00:07:25,000 --> 00:07:27,000 I'm going to import OS. 136 00:07:27,000 --> 00:07:34,000 Uh I'm going to import uh sis I think I'll not be using sis but let it be. 137 00:07:34,000 --> 00:07:37,000 These are some of the common libraries that we will specifically use. 138 00:07:37,000 --> 00:07:39,000 Along with this we are going to use Boto3. 139 00:07:39,000 --> 00:07:44,000 And again guys, in my previous video I have shown you how to configure the aws cli. 140 00:07:44,000 --> 00:07:48,000 Please make sure that you watch the playlist, otherwise you will not be able to understand. 141 00:07:48,000 --> 00:07:53,000 Okay, if you directly go into and jump in this particular video, no, you'll not be able to understand. 142 00:07:53,000 --> 00:07:55,000 So in the description I will give you the playlist. 143 00:07:55,000 --> 00:07:56,000 The first video only. 144 00:07:56,000 --> 00:07:59,000 I've shown you how you can actually configure the aws cli. 145 00:08:00,000 --> 00:08:00,000 Okay. 146 00:08:00,000 --> 00:08:01,000 So please make sure this okay. 147 00:08:01,000 --> 00:08:04,000 The next thing is that as I said right. 148 00:08:04,000 --> 00:08:08,000 We will be we will be using. 149 00:08:10,000 --> 00:08:19,000 using Titan embedding small model right for creating vectors or to generate embeddings. 150 00:08:19,000 --> 00:08:19,000 Okay. 151 00:08:19,000 --> 00:08:21,000 So we will create this model. 152 00:08:21,000 --> 00:08:24,000 The reason why I'm showing you because I've never used in any of my videos. 153 00:08:24,000 --> 00:08:30,000 Now this Titan embedding what I will do I will call from the long chain library a long chain framework. 154 00:08:30,000 --> 00:08:37,000 Long chain provides you multiple functionalities, multiple multiple options to properly interact with 155 00:08:37,000 --> 00:08:39,000 bedrock, right AWS bedrock. 156 00:08:39,000 --> 00:08:42,000 So as I said, this framework is compulsory for you all to know. 157 00:08:42,000 --> 00:08:43,000 One is launching and Lama index. 158 00:08:43,000 --> 00:08:44,000 Right. 159 00:08:44,000 --> 00:08:53,000 So what I'm going to do I'm going to right from lang chain dot embeddings I'm going to import bedrock. 160 00:08:54,000 --> 00:08:56,000 Let's see bed. 161 00:08:59,000 --> 00:08:59,000 Bed. 162 00:08:59,000 --> 00:09:01,000 Rock embeddings. 163 00:09:01,000 --> 00:09:01,000 Right. 164 00:09:01,000 --> 00:09:03,000 So I'm going to specifically use this. 165 00:09:03,000 --> 00:09:05,000 This will be small letter okay. 166 00:09:05,000 --> 00:09:08,000 So guys I'm also verifying it from the documentation okay. 167 00:09:08,000 --> 00:09:13,000 So there is a documentation that is given not very much clearly but I've implemented this entire project 168 00:09:13,000 --> 00:09:13,000 also. 169 00:09:13,000 --> 00:09:14,000 Right. 170 00:09:14,000 --> 00:09:17,000 So from lang chain dot embeddings import bedrock embeddings. 171 00:09:17,000 --> 00:09:22,000 The next thing that I'm going to use from lang chain dot LMS. 172 00:09:22,000 --> 00:09:26,000 Like we can also use Lange chain to call this LM models that are present inside bedrock. 173 00:09:26,000 --> 00:09:35,000 So that is also integrated right from LMS dot bedrock import bedrock. 174 00:09:35,000 --> 00:09:38,000 Okay I'm going to specifically use this tool. 175 00:09:38,000 --> 00:09:43,000 One is from Lange dot embeddings bedrock embeddings and from Lange chain dot lms dot bedrock. 176 00:09:44,000 --> 00:09:46,000 Uh I'm going to import the bedrock. 177 00:09:46,000 --> 00:09:46,000 Okay. 178 00:09:46,000 --> 00:09:49,000 So this is specifically for the embedding part okay. 179 00:09:49,000 --> 00:09:55,000 Now they will also be some libraries that I need to import for data ingestion because I really need 180 00:09:55,000 --> 00:09:57,000 to load the data set right. 181 00:09:57,000 --> 00:10:04,000 So over here you'll be able to see I will write import numpy as np. 182 00:10:04,000 --> 00:10:05,000 Okay. 183 00:10:05,000 --> 00:10:08,000 Uh this is numpy I'm going to specifically use. 184 00:10:08,000 --> 00:10:11,000 I think by default numpy will be available. 185 00:10:11,000 --> 00:10:16,000 Import numpy as np perfect, so numpy as np. 186 00:10:16,000 --> 00:10:26,000 Then from long chain dot text splitter, I'm going to use text splitter so that as soon as I load the 187 00:10:26,000 --> 00:10:31,000 document, I need to probably use this recursive character text splitter. 188 00:10:31,000 --> 00:10:34,000 Okay, so this is what we specifically use. 189 00:10:34,000 --> 00:10:37,000 And in my lecture series I've discussed everything about this. 190 00:10:37,000 --> 00:10:48,000 Then again I will go ahead and write from Lange chain dot document on document loaders import. 191 00:10:49,000 --> 00:10:52,000 Um we will specifically use this pi. 192 00:10:53,000 --> 00:10:56,000 So let me just copy this pi pdf directly. 193 00:10:56,000 --> 00:10:57,000 Loaders. 194 00:10:57,000 --> 00:11:01,000 So this is what we are specifically going to use py pdf directly loaders. 195 00:11:01,000 --> 00:11:05,000 And it is actually present inside long chained or document loaders. 196 00:11:05,000 --> 00:11:05,000 Okay. 197 00:11:05,000 --> 00:11:09,000 So this is specifically required for the data ingestion. 198 00:11:09,000 --> 00:11:14,000 With this what I will do I have already created one folder over here which is called as data which has 199 00:11:14,000 --> 00:11:15,000 this two PDFs. 200 00:11:15,000 --> 00:11:16,000 Right. 201 00:11:16,000 --> 00:11:21,000 I need to load all this PDFs from this particular folder and then perform all the vector embeddings 202 00:11:21,000 --> 00:11:22,000 that I require. 203 00:11:22,000 --> 00:11:24,000 So in data ingestion, first of all we will load it. 204 00:11:24,000 --> 00:11:29,000 We will split the entire documents by using recursive character text splitter. 205 00:11:29,000 --> 00:11:33,000 And then after this we will convert this into vector embeddings. 206 00:11:33,000 --> 00:11:40,000 So let me just go ahead and write vector embeddings and vector store here I'm going to specifically 207 00:11:40,000 --> 00:11:43,000 use for this index okay fish db or chroma db. 208 00:11:43,000 --> 00:11:44,000 Also you can use it is up to you. 209 00:11:44,000 --> 00:11:47,000 Again I have shown both the ways in my playlist. 210 00:11:47,000 --> 00:11:51,000 So from Long Chain let me just see this. 211 00:11:51,000 --> 00:11:54,000 Okay, so bedrock embeddings. 212 00:11:54,000 --> 00:12:00,000 Is there something called as from long chain underscore community. 213 00:12:00,000 --> 00:12:00,000 It was right. 214 00:12:00,000 --> 00:12:02,000 Let's see okay. 215 00:12:02,000 --> 00:12:06,000 From from long chain underscore community. 216 00:12:06,000 --> 00:12:08,000 Let's copy this. 217 00:12:08,000 --> 00:12:10,000 Is this correct or not. 218 00:12:10,000 --> 00:12:11,000 We'll try to see Yeah. 219 00:12:11,000 --> 00:12:13,000 Bedrock embeddings was there. 220 00:12:13,000 --> 00:12:16,000 Uh, let's copy this here also. 221 00:12:16,000 --> 00:12:18,000 So py PDF directory loader is also there. 222 00:12:18,000 --> 00:12:20,000 I think there will be some warnings that will be coming. 223 00:12:20,000 --> 00:12:24,000 Uh, I think most of the, some of the libraries and community. 224 00:12:24,000 --> 00:12:25,000 Okay. 225 00:12:25,000 --> 00:12:29,000 So here what I'm actually going to do from Lang Chain. 226 00:12:30,000 --> 00:12:33,000 Let's see if we get an error again I'll revert it okay. 227 00:12:34,000 --> 00:12:36,000 From Lang chain dot text splitter. 228 00:12:36,000 --> 00:12:37,000 Sorry, not text splitter. 229 00:12:37,000 --> 00:12:37,000 Why? 230 00:12:37,000 --> 00:12:39,000 I'm using text splitter already. 231 00:12:39,000 --> 00:12:40,000 I've actually done it. 232 00:12:40,000 --> 00:12:47,000 So I'm also going to use for this vector embedding specifically uh, I have to use vector uh files. 233 00:12:47,000 --> 00:12:47,000 Right. 234 00:12:47,000 --> 00:12:53,000 So I will write vector stores import files. 235 00:12:54,000 --> 00:12:54,000 Okay. 236 00:12:54,000 --> 00:12:57,000 So fierce is one library I'll be using. 237 00:12:57,000 --> 00:12:57,000 Again. 238 00:12:57,000 --> 00:13:00,000 This files will be present inside community I guess. 239 00:13:00,000 --> 00:13:04,000 Okay, so Vector Langton underscore community. 240 00:13:04,000 --> 00:13:07,000 Uh, dot vector stores import files. 241 00:13:07,000 --> 00:13:09,000 Let's see if this is not accessible. 242 00:13:09,000 --> 00:13:12,000 So I think it is inside Langton only okay. 243 00:13:12,000 --> 00:13:14,000 Otherwise you can see the documentation okay. 244 00:13:14,000 --> 00:13:15,000 At the end of the day no worries. 245 00:13:15,000 --> 00:13:22,000 Then after doing this I will also write from lang che dot indexes. 246 00:13:22,000 --> 00:13:27,000 Okay I will not use this Also, let's just use fires because we have already installed fire CPU. 247 00:13:27,000 --> 00:13:28,000 Okay, then. 248 00:13:28,000 --> 00:13:35,000 Um, after doing this, uh, what we are specifically going to do is that we are now this is for the 249 00:13:35,000 --> 00:13:35,000 vector embedding. 250 00:13:35,000 --> 00:13:38,000 Now we need to go ahead with the LM models. 251 00:13:38,000 --> 00:13:38,000 Right. 252 00:13:38,000 --> 00:13:46,000 So for LM models there Lang chain already provides ways to load models from the AWS bedrock okay. 253 00:13:46,000 --> 00:13:49,000 So for here I'll write from long chain. 254 00:13:51,000 --> 00:14:00,000 From long chain underscore community dot one I will specifically use prompts oops from long chain dot 255 00:14:00,000 --> 00:14:00,000 prompts. 256 00:14:01,000 --> 00:14:05,000 I'm going to import prompt template okay. 257 00:14:05,000 --> 00:14:11,000 And then the other one is from long chain dot chain. 258 00:14:11,000 --> 00:14:17,000 So I'm going to specifically use one chain where I'll be using import retrieval QA. 259 00:14:17,000 --> 00:14:18,000 Right. 260 00:14:18,000 --> 00:14:23,000 So since I'm going to create a Q&A chat bot with the documents, uh, Q&A in short. 261 00:14:23,000 --> 00:14:25,000 So that is the reason I'm using this. 262 00:14:25,000 --> 00:14:27,000 And this is basically to create my own prompt template. 263 00:14:28,000 --> 00:14:34,000 Now let's call the bedrock client so that we get the access of all the models okay. 264 00:14:34,000 --> 00:14:38,000 So I will go ahead and set up the bedrock clients over here. 265 00:14:38,000 --> 00:14:39,000 So here let me go ahead and write. 266 00:14:39,000 --> 00:14:43,000 Bedrock is equal to Boto3 dot client. 267 00:14:43,000 --> 00:14:53,000 And here I'm going to give my service name is equal to bed the bed bedrock dash runtime okay. 268 00:14:53,000 --> 00:14:59,000 So I've already shown you yesterday also uh in in my playlist of this bedrock, I've also shown you 269 00:14:59,000 --> 00:15:02,000 how you can actually call the client itself and access the models. 270 00:15:02,000 --> 00:15:08,000 Okay, now as said, as I as I already told that the embedding that we are going to specifically use 271 00:15:08,000 --> 00:15:11,000 is called as bedrock embeddings, right. 272 00:15:11,000 --> 00:15:14,000 So let me quickly go ahead and call this. 273 00:15:14,000 --> 00:15:16,000 I will go ahead and write Bedrock Embedding. 274 00:15:16,000 --> 00:15:21,000 And let me show you how you can actually call this embedding from the bedrock. 275 00:15:21,000 --> 00:15:23,000 So I will copy paste this. 276 00:15:23,000 --> 00:15:25,000 Let me give the model ID. 277 00:15:25,000 --> 00:15:28,000 Now this model ID you'll be able to see it over here right. 278 00:15:28,000 --> 00:15:36,000 So if you probably go ahead in AWS itself let's say I will go ahead and write embedding okay. 279 00:15:37,000 --> 00:15:41,000 So here only in Titan I think you will be able to find it out. 280 00:15:41,000 --> 00:15:45,000 So foundation model base models. 281 00:15:45,000 --> 00:15:51,000 So if you go ahead and search inside this any of the embedding models you can specifically take. 282 00:15:51,000 --> 00:15:54,000 But I am going to use this Titan embedding model okay. 283 00:15:54,000 --> 00:16:00,000 So if you probably click over here and see down you'll be getting the entire details like what is the 284 00:16:00,000 --> 00:16:02,000 embedding, what is the model ID and all. 285 00:16:02,000 --> 00:16:04,000 Okay, so I am going to specifically use this. 286 00:16:04,000 --> 00:16:07,000 So let me quickly go ahead and call this model ID. 287 00:16:07,000 --> 00:16:13,000 So for this model ID I've already copied it from that from this particular website. 288 00:16:13,000 --> 00:16:19,000 So here you can see I will have copied this entire thing and I will paste it over here on my model ID. 289 00:16:19,000 --> 00:16:21,000 So you also have to do the same step okay. 290 00:16:21,000 --> 00:16:26,000 So once we do this the next thing in this that we really need to give is the client. 291 00:16:26,000 --> 00:16:27,000 And client. 292 00:16:27,000 --> 00:16:31,000 You know that we have not yet called any client or what. 293 00:16:31,000 --> 00:16:32,000 So we have called this bedrock client. 294 00:16:32,000 --> 00:16:35,000 So this will basically be my client over here. 295 00:16:35,000 --> 00:16:35,000 Right. 296 00:16:35,000 --> 00:16:41,000 So once we give this bedrock client, that basically means it knows we are going to call this particular 297 00:16:41,000 --> 00:16:43,000 embedding model from this particular client. 298 00:16:43,000 --> 00:16:44,000 That is bedrock itself. 299 00:16:44,000 --> 00:16:48,000 In short, we are going to use the AWS, lm, uh, AWS bedrock. 300 00:16:48,000 --> 00:16:51,000 Now let me quickly go ahead. 301 00:16:51,000 --> 00:16:53,000 And this is what is all step by step will be going. 302 00:16:53,000 --> 00:16:57,000 First we will create this data ingestion model. 303 00:16:57,000 --> 00:16:59,000 See if this step is done. 304 00:16:59,000 --> 00:17:00,000 We have created our client. 305 00:17:00,000 --> 00:17:04,000 Now we will go ahead and implement this data ingestion so quickly. 306 00:17:04,000 --> 00:17:07,000 Let's go ahead and implement this data ingestion. 307 00:17:07,000 --> 00:17:10,000 And please make sure that you follow this. 308 00:17:10,000 --> 00:17:14,000 So here I'm going to basically write data ingestion okay. 309 00:17:14,000 --> 00:17:20,000 And uh I will create a step which is called as data ingestion a function. 310 00:17:20,000 --> 00:17:25,000 Now inside this data ingestion what things we are specifically going to do. 311 00:17:25,000 --> 00:17:26,000 Right I have a folder. 312 00:17:26,000 --> 00:17:27,000 So let me go ahead and write. 313 00:17:27,000 --> 00:17:31,000 Loader is equal to pi PDF directory loader. 314 00:17:31,000 --> 00:17:34,000 And here I'm going to give my folder which is called as data. 315 00:17:34,000 --> 00:17:38,000 Right from that data folder only I need to pick up all the PDF files right. 316 00:17:38,000 --> 00:17:39,000 So this is the first step. 317 00:17:39,000 --> 00:17:45,000 Then I will go ahead and write documents is equal to loader dot load. 318 00:17:45,000 --> 00:17:46,000 Right. 319 00:17:46,000 --> 00:17:54,000 So from that this loader we are going to load this entire documents Now in our testing we will specifically 320 00:17:54,000 --> 00:17:57,000 use something called as characters text split. 321 00:17:57,000 --> 00:17:57,000 Right. 322 00:17:57,000 --> 00:17:59,000 So that is what we are. 323 00:17:59,000 --> 00:18:01,000 We have already implemented right. 324 00:18:01,000 --> 00:18:02,000 Recursive character text split. 325 00:18:02,000 --> 00:18:04,000 Now how do you implement this? 326 00:18:04,000 --> 00:18:05,000 It is very much simple. 327 00:18:05,000 --> 00:18:07,000 I will go ahead and write text splitter. 328 00:18:07,000 --> 00:18:08,000 Right. 329 00:18:08,000 --> 00:18:11,000 And this with the action I've already shown you. 330 00:18:11,000 --> 00:18:13,000 So it will be recursive text splitter. 331 00:18:13,000 --> 00:18:18,000 And here first of parameter will be my chunk size. 332 00:18:20,000 --> 00:18:22,000 Chunk underscore size. 333 00:18:22,000 --> 00:18:23,000 Chunk size. 334 00:18:23,000 --> 00:18:26,000 I will give it to 1000 or 10,000. 335 00:18:27,000 --> 00:18:27,000 Right. 336 00:18:27,000 --> 00:18:33,000 And along with this chunk size the second parameter I would like to give as chunk overlap. 337 00:18:33,000 --> 00:18:39,000 Please give this number a little bit more value so that you will be able to understand it Okay. 338 00:18:39,000 --> 00:18:43,000 So this two are their text splitter recursive character text splitter. 339 00:18:43,000 --> 00:18:45,000 And this right. 340 00:18:45,000 --> 00:18:52,000 Once we get this text splitter I will go ahead and write Dox is equal to text splitter dot from oh sorry 341 00:18:52,000 --> 00:18:55,000 dot split underscore documents okay. 342 00:18:55,000 --> 00:18:58,000 And then here I'm going to give my entire document. 343 00:18:58,000 --> 00:19:02,000 So in short we are going to split it right based on this recursive character text splitter. 344 00:19:02,000 --> 00:19:06,000 Once we do this, then we will return the docs. 345 00:19:06,000 --> 00:19:08,000 All the docs we're going to specifically return the docs. 346 00:19:08,000 --> 00:19:13,000 So data ingestion is done right with respect to this particular data ingestion. 347 00:19:13,000 --> 00:19:16,000 Here what we are doing we are reading all the PDFs from the data folder. 348 00:19:16,000 --> 00:19:20,000 We are doing recursive character text splitter with this chunk overlapping. 349 00:19:20,000 --> 00:19:23,000 And then we split all those documents right. 350 00:19:23,000 --> 00:19:26,000 So this becomes a this completes our data ingestion. 351 00:19:26,000 --> 00:19:32,000 Now the next thing is that we will go ahead with vector embeddings. 352 00:19:33,000 --> 00:19:35,000 Vector embedding and vector store. 353 00:19:35,000 --> 00:19:36,000 Right. 354 00:19:36,000 --> 00:19:40,000 And this is where we are going to specifically use that title embedding that we have imported. 355 00:19:40,000 --> 00:19:44,000 Along with that, what we are specifically going to do is that we are going to use this files. 356 00:19:44,000 --> 00:19:45,000 Right. 357 00:19:45,000 --> 00:19:48,000 So let me quickly go ahead and write it down. 358 00:19:48,000 --> 00:19:49,000 So here I'm going to write my definition. 359 00:19:50,000 --> 00:19:53,000 And quickly let's go ahead and write Get Vector Store. 360 00:19:53,000 --> 00:19:54,000 So I'll create one more function. 361 00:19:55,000 --> 00:19:59,000 Now inside this vector store I'm going to first of all give the documents whatever documents is coming 362 00:19:59,000 --> 00:19:59,000 from here. 363 00:19:59,000 --> 00:20:04,000 Because we take this documents and then we do the embedding techniques and we perform all the embedding 364 00:20:04,000 --> 00:20:04,000 techniques. 365 00:20:04,000 --> 00:20:12,000 So in my next step what I will go ahead and write I'll write a vector store underscore files okay. 366 00:20:12,000 --> 00:20:14,000 So this will basically be my variable. 367 00:20:14,000 --> 00:20:18,000 And then I'm going to use fire start. 368 00:20:18,000 --> 00:20:18,000 Okay. 369 00:20:19,000 --> 00:20:23,000 From underscore documents right. 370 00:20:23,000 --> 00:20:26,000 And we are going to implement this right. 371 00:20:26,000 --> 00:20:30,000 So in the first parameter inside fires we give docs. 372 00:20:30,000 --> 00:20:34,000 And the second parameter that we specifically give is my bedrock embeddings. 373 00:20:34,000 --> 00:20:34,000 Right. 374 00:20:34,000 --> 00:20:36,000 So this is the embedding. 375 00:20:36,000 --> 00:20:42,000 Specifically we are going to use okay one One is the bedrock embedding which we have initialized over 376 00:20:42,000 --> 00:20:42,000 here. 377 00:20:42,000 --> 00:20:45,000 See over here in the bedrock embedding we have initialized. 378 00:20:45,000 --> 00:20:46,000 And that is what we are going to use it over here. 379 00:20:46,000 --> 00:20:53,000 Now after we get that vector surface this is basically my vector store I will go ahead and save it in 380 00:20:53,000 --> 00:20:54,000 my local disk. 381 00:20:54,000 --> 00:20:59,000 So let me quickly write dot save underscore local. 382 00:21:00,000 --> 00:21:05,000 And inside this we are going to basically write phi s underscore index. 383 00:21:05,000 --> 00:21:05,000 Right. 384 00:21:05,000 --> 00:21:11,000 So this will basically be saved in my uh folder over here itself right in my hard disk. 385 00:21:11,000 --> 00:21:14,000 You can also save this in any database as such if you want. 386 00:21:14,000 --> 00:21:17,000 So this is my vector embedding I've done it. 387 00:21:17,000 --> 00:21:19,000 And this step is also completed. 388 00:21:19,000 --> 00:21:22,000 Now let's talk about the next step. 389 00:21:22,000 --> 00:21:26,000 In the next step we based on the import we have to work with LM models. 390 00:21:26,000 --> 00:21:30,000 Now what we are going to do over here is that I'm going to create some LM models. 391 00:21:30,000 --> 00:21:35,000 Let's say I will go ahead and write first LM model that I'm going to work with is cloudy. 392 00:21:35,000 --> 00:21:35,000 Right. 393 00:21:35,000 --> 00:21:45,000 So I will say cloudy LM and inside this cloudy we will create create the anthropic model. 394 00:21:45,000 --> 00:21:52,000 Because I should say already AWS bedrock is giving you the power to harness multiple models to use the 395 00:21:52,000 --> 00:21:53,000 multiple models. 396 00:21:53,000 --> 00:21:56,000 So we can specifically develop in this application itself. 397 00:21:56,000 --> 00:21:59,000 It's just like having different different models in different, different ways. 398 00:21:59,000 --> 00:22:01,000 But here I'll try to make it much more generic. 399 00:22:01,000 --> 00:22:05,000 So here we are going to write quite the anthropic model and I'm going to create my LLM. 400 00:22:06,000 --> 00:22:08,000 Along with this we are going to use this bedrock. 401 00:22:08,000 --> 00:22:15,000 So bedrock um, before if I probably show you in my previous tutorial, the way of invoking a model 402 00:22:15,000 --> 00:22:18,000 was something like bedrock dot invoke model. 403 00:22:19,000 --> 00:22:23,000 But now, since we are already using Lang chain, we don't have to basically use invoke model. 404 00:22:23,000 --> 00:22:27,000 So what we specifically do, we just call this bedrock. 405 00:22:27,000 --> 00:22:32,000 And you can probably see how where this bedrock is present, it is present in a Lang chain. 406 00:22:32,000 --> 00:22:34,000 So they have actually created a wrapper. 407 00:22:34,000 --> 00:22:37,000 And internally they will invoke that specific model. 408 00:22:37,000 --> 00:22:42,000 But this is a way that how you can use this frameworks like Lang in a generic way. 409 00:22:42,000 --> 00:22:45,000 So now I'm going to use this model as bedrock. 410 00:22:45,000 --> 00:22:49,000 Now with respect to this I will go ahead and write my model ID. 411 00:22:50,000 --> 00:22:55,000 Already I have searched my model ID so this is what is my model ID over here. 412 00:22:55,000 --> 00:23:01,000 How you do you get this information again from this particular examples, go and click on any model 413 00:23:01,000 --> 00:23:02,000 you will be able to. 414 00:23:02,000 --> 00:23:05,000 Let's say I want to generate cloud I generate code. 415 00:23:05,000 --> 00:23:08,000 So with respect to this this will be my model ID right. 416 00:23:08,000 --> 00:23:12,000 So it is a generic way of finding the model ID so you don't have to even worry about it. 417 00:23:12,000 --> 00:23:13,000 Right. 418 00:23:13,000 --> 00:23:17,000 So I have my model ID which I did it in my previous tutorial also. 419 00:23:17,000 --> 00:23:20,000 And then I will go ahead and write my client. 420 00:23:21,000 --> 00:23:24,000 My client will be nothing but bedrock. 421 00:23:25,000 --> 00:23:27,000 Bedrock. 422 00:23:27,000 --> 00:23:30,000 And the next thing that I'm going to specifically give after this, right? 423 00:23:30,000 --> 00:23:32,000 One is model arguments. 424 00:23:32,000 --> 00:23:32,000 Right. 425 00:23:32,000 --> 00:23:35,000 So model arguments with respect to this. 426 00:23:35,000 --> 00:23:38,000 Now this arguments, this arguments that you'll be seeing. 427 00:23:38,000 --> 00:23:38,000 Right. 428 00:23:38,000 --> 00:23:42,000 Where do you find it again if you probably go here inside the body. 429 00:23:42,000 --> 00:23:48,000 And if you go at the last right, there will be some arguments that will be added like this. 430 00:23:48,000 --> 00:23:50,000 Max tokens to sample temperature. 431 00:23:50,000 --> 00:23:50,000 This this. 432 00:23:50,000 --> 00:23:52,000 You can also add it over here. 433 00:23:52,000 --> 00:23:57,000 Now with respect to this particular model that I am actually using, this was the argument that was 434 00:23:57,000 --> 00:23:59,000 present inside that particular body. 435 00:23:59,000 --> 00:24:01,000 So that is the reason I had created this. 436 00:24:01,000 --> 00:24:01,000 Right. 437 00:24:01,000 --> 00:24:04,000 All this JSON file separately so that I can refer it. 438 00:24:04,000 --> 00:24:05,000 Okay. 439 00:24:05,000 --> 00:24:07,000 So I hope you are able to understand it. 440 00:24:07,000 --> 00:24:09,000 If you are able to understand, please make sure that you hit like till now. 441 00:24:09,000 --> 00:24:11,000 And now let's go ahead. 442 00:24:11,000 --> 00:24:15,000 And here what we are specifically doing whenever I call this function, that basically means my cloud 443 00:24:15,000 --> 00:24:17,000 model has got loaded. 444 00:24:17,000 --> 00:24:20,000 So I'm going to return the LM model. 445 00:24:20,000 --> 00:24:25,000 Now suppose similarly you want to probably go ahead and call the llama two model. 446 00:24:25,000 --> 00:24:28,000 So I will go ahead and paste it over here. 447 00:24:28,000 --> 00:24:30,000 I'll say get llama two LM like this. 448 00:24:30,000 --> 00:24:34,000 You can actually create any number of functions as you want okay. 449 00:24:34,000 --> 00:24:36,000 So this will basically be my llama two. 450 00:24:36,000 --> 00:24:40,000 Now inside my llama two let me just copy this again I'm going to see my. 451 00:24:41,000 --> 00:24:44,000 So this is the llama two model over here okay. 452 00:24:44,000 --> 00:24:48,000 Only the model ID change and whatever is the argument that will change. 453 00:24:48,000 --> 00:24:54,000 So in the case of arguments uh, with respect to this, this is nothing but max gen length, right? 454 00:24:54,000 --> 00:24:55,000 How I'm getting it. 455 00:24:55,000 --> 00:24:57,000 Let me show you again okay. 456 00:24:57,000 --> 00:24:59,000 So let's say this is a chain of thoughts. 457 00:24:59,000 --> 00:25:01,000 Let's say that I'm going to use this okay. 458 00:25:01,000 --> 00:25:05,000 So here you can probably see this is my model ID okay. 459 00:25:06,000 --> 00:25:11,000 And then inside my arguments this is from where we'll be able to get it right. 460 00:25:11,000 --> 00:25:15,000 So this API request is pretty much important with respect to this okay. 461 00:25:15,000 --> 00:25:17,000 So return a model. 462 00:25:17,000 --> 00:25:21,000 And this is where is my model that I'm getting with respect to llama two. 463 00:25:21,000 --> 00:25:26,000 So now you know how to create how to call any uh models itself right. 464 00:25:26,000 --> 00:25:27,000 Data ingestion is done. 465 00:25:27,000 --> 00:25:29,000 Vector store is done. 466 00:25:29,000 --> 00:25:29,000 Everything is done. 467 00:25:29,000 --> 00:25:32,000 Now let's go ahead and create my prompt template. 468 00:25:32,000 --> 00:25:37,000 Now this prompt template that I'm actually going to create I'm going to use something called as Lang 469 00:25:37,000 --> 00:25:37,000 chain. 470 00:25:37,000 --> 00:25:41,000 So in my prompt template I have written see this okay. 471 00:25:44,000 --> 00:25:48,000 So I've given a simple prompt template which you can also use it. 472 00:25:52,000 --> 00:25:54,000 So I have written human in this format. 473 00:25:54,000 --> 00:26:00,000 Use the following piece of context to provide a concise answer to the and at least summarize with 250 474 00:26:00,000 --> 00:26:01,000 words with detail explanation. 475 00:26:01,000 --> 00:26:03,000 If you don't know the answer, just say that you don't know. 476 00:26:03,000 --> 00:26:05,000 Just don't try to make up the answer. 477 00:26:05,000 --> 00:26:06,000 So context and question is there. 478 00:26:06,000 --> 00:26:10,000 And based on this assistant, whatever assistant output is there, it will get appended over here. 479 00:26:11,000 --> 00:26:14,000 Now with respect to this particular thing we use the login prompt template. 480 00:26:14,000 --> 00:26:17,000 We write context and question and we use it okay. 481 00:26:17,000 --> 00:26:18,000 Now this is fine. 482 00:26:18,000 --> 00:26:19,000 Everything is good. 483 00:26:19,000 --> 00:26:21,000 Now let's focus on the response part. 484 00:26:21,000 --> 00:26:26,000 So here I will go ahead and write get underscore response underscore lm. 485 00:26:26,000 --> 00:26:28,000 And here the first parameter is my lm. 486 00:26:28,000 --> 00:26:31,000 The second parameter is my vector store. 487 00:26:31,000 --> 00:26:32,000 Vector store fires. 488 00:26:33,000 --> 00:26:35,000 The third parameter that I'm going to give is my query. 489 00:26:35,000 --> 00:26:35,000 Right. 490 00:26:35,000 --> 00:26:38,000 So this three parameters we are going to play with. 491 00:26:38,000 --> 00:26:44,000 One is obviously if I want to get the response from a specific LM then I have to give this three information. 492 00:26:44,000 --> 00:26:50,000 Okay, get response LM the LM model, which I'm calling this is my first index and this is my query 493 00:26:50,000 --> 00:26:51,000 okay. 494 00:26:51,000 --> 00:26:57,000 Now once we get this, we as you know that we have imported something called as a retrieval cue. 495 00:26:57,000 --> 00:26:58,000 So retrieval cue. 496 00:26:58,000 --> 00:27:00,000 And here also we are going to use it. 497 00:27:00,000 --> 00:27:02,000 And this is how you basically call it. 498 00:27:02,000 --> 00:27:03,000 Right. 499 00:27:03,000 --> 00:27:05,000 So I'm just going to copy it paste it over here. 500 00:27:05,000 --> 00:27:07,000 Again I'm seeing the lantern documentation. 501 00:27:07,000 --> 00:27:11,000 So retrieval QA from chain type I'm giving my LM model. 502 00:27:11,000 --> 00:27:13,000 Chain type will be stuff I've already shown you. 503 00:27:13,000 --> 00:27:15,000 What is text summarization. 504 00:27:15,000 --> 00:27:17,000 Different different text summarization techniques. 505 00:27:17,000 --> 00:27:18,000 So stuff is there over here. 506 00:27:18,000 --> 00:27:20,000 This is the most important thing. 507 00:27:20,000 --> 00:27:25,000 Retriever how from where the similarity search will basically happen. 508 00:27:25,000 --> 00:27:26,000 So this is where we are doing right. 509 00:27:26,000 --> 00:27:31,000 Office index basically has the entire index itself right. 510 00:27:31,000 --> 00:27:31,000 Right. 511 00:27:31,000 --> 00:27:36,000 And then we are trying to do the similarity search along with the argument top k three or top three 512 00:27:36,000 --> 00:27:37,000 prompts. 513 00:27:37,000 --> 00:27:41,000 And then we are also saying return source document with respect to this one. 514 00:27:41,000 --> 00:27:43,000 And then I have my prompt over here. 515 00:27:43,000 --> 00:27:44,000 Right. 516 00:27:44,000 --> 00:27:45,000 So all these things are basically done. 517 00:27:45,000 --> 00:27:48,000 And then after this we will be able to get the response. 518 00:27:48,000 --> 00:27:52,000 So now once we specifically write this chain type uh arguments. 519 00:27:52,000 --> 00:27:54,000 And here we are specifically giving the prompt. 520 00:27:54,000 --> 00:27:57,000 The prompt is getting created by the prompt template over here. 521 00:27:57,000 --> 00:28:04,000 And then in the next step we will go ahead and right answer I'll create a variable called as answer. 522 00:28:04,000 --> 00:28:06,000 And we can use this q a variable. 523 00:28:07,000 --> 00:28:13,000 And inside this we will give in the form of query colon. 524 00:28:13,000 --> 00:28:16,000 And here we are going to give the query itself. 525 00:28:16,000 --> 00:28:21,000 So this query is the query that is coming from the as As an input okay. 526 00:28:21,000 --> 00:28:27,000 And once we get the answer see when we when this QA is basically using right it is nothing but retrieval 527 00:28:27,000 --> 00:28:27,000 QA. 528 00:28:28,000 --> 00:28:33,000 So it will retrieve the response and it will store it in this particular variable. 529 00:28:33,000 --> 00:28:40,000 Now inside this variable there will be something called as result as a key which will give you the output. 530 00:28:41,000 --> 00:28:42,000 So if you print it right. 531 00:28:42,000 --> 00:28:44,000 Initially I printed this answer. 532 00:28:44,000 --> 00:28:46,000 I've already implemented this. 533 00:28:46,000 --> 00:28:49,000 So there there is a variable called as result which has the entire answer. 534 00:28:50,000 --> 00:28:55,000 So this is my response LM where I'm getting the entire result. 535 00:28:55,000 --> 00:28:58,000 Now let's go ahead and create our Streamlit app. 536 00:28:59,000 --> 00:29:02,000 Now quickly there are two important things with respect to this Streamlit app. 537 00:29:03,000 --> 00:29:09,000 One place I have to make sure that whenever a document is updated, you know that it should get converted 538 00:29:09,000 --> 00:29:11,000 into a vector embeddings. 539 00:29:11,000 --> 00:29:17,000 So what I'm actually going to do quickly over here, I'll create a main function and let me show you 540 00:29:17,000 --> 00:29:19,000 what I will be doing okay. 541 00:29:20,000 --> 00:29:23,000 So quickly first of all I will import Streamlit as st. 542 00:29:26,000 --> 00:29:27,000 Streamlit as st. 543 00:29:30,000 --> 00:29:31,000 Fine. 544 00:29:31,000 --> 00:29:32,000 We are going to use Streamlit. 545 00:29:32,000 --> 00:29:33,000 Now see this? 546 00:29:33,000 --> 00:29:35,000 The thing that I've copied and pasted. 547 00:29:35,000 --> 00:29:39,000 So I've written chat, PDF chat with PDF using AWS bedrock. 548 00:29:39,000 --> 00:29:40,000 This is my user question. 549 00:29:40,000 --> 00:29:42,000 Like what kind of question I want from the pdf file. 550 00:29:42,000 --> 00:29:53,000 Then a sidebar is basically created and I will say update or create vector store, right? 551 00:29:53,000 --> 00:29:59,000 That basically means inside this I will create a button saying that vectors update. 552 00:29:59,000 --> 00:30:05,000 That basically means once I click click right then it should go ahead and call this data ingestion. 553 00:30:05,000 --> 00:30:11,000 Now data ingestion, what it is going to do it is this data ingestion is going to read all the files 554 00:30:11,000 --> 00:30:12,000 from the data folder. 555 00:30:12,000 --> 00:30:15,000 Then it is going to take this loader documents. 556 00:30:15,000 --> 00:30:18,000 It is going to perform this recursive character text splitter. 557 00:30:18,000 --> 00:30:22,000 And then after providing after doing this it is going to return this entire documents. 558 00:30:22,000 --> 00:30:25,000 Right after this we are going to call this vector store. 559 00:30:25,000 --> 00:30:27,000 So one by one we will be calling it. 560 00:30:27,000 --> 00:30:30,000 So here you will be able to see data injection and vector store. 561 00:30:30,000 --> 00:30:30,000 We are calling. 562 00:30:30,000 --> 00:30:31,000 We get this docs. 563 00:30:32,000 --> 00:30:33,000 Now inside that vector store. 564 00:30:33,000 --> 00:30:39,000 What is happening is that we will save this in our local, uh, folder itself in the hard disk in the 565 00:30:39,000 --> 00:30:40,000 form of files index. 566 00:30:40,000 --> 00:30:47,000 So as soon as I click this button that is vector update Update, which will be available in the sideways 567 00:30:47,000 --> 00:30:48,000 inside the sidebar sidebar. 568 00:30:49,000 --> 00:30:52,000 Means in Streamlit it is in the left hand side. 569 00:30:52,000 --> 00:30:57,000 So the vectors will get created, the vector stores will get created, and it will also saved in a local 570 00:30:57,000 --> 00:30:58,000 folder itself. 571 00:30:58,000 --> 00:30:58,000 Right. 572 00:30:58,000 --> 00:31:03,000 So over here a folder will come with this particular name that is called as files index. 573 00:31:03,000 --> 00:31:04,000 This is one step. 574 00:31:04,000 --> 00:31:10,000 Now the second step is that I can go ahead and create a cloudy output. 575 00:31:10,000 --> 00:31:14,000 Okay, I will create a cloud, a button which is called as cloudy output. 576 00:31:14,000 --> 00:31:14,000 Now see this. 577 00:31:14,000 --> 00:31:18,000 This is important to understand okay so this is my button. 578 00:31:18,000 --> 00:31:20,000 I've created another button. 579 00:31:20,000 --> 00:31:25,000 When I say cloudy output that basically means I have to use the cloudy model API. 580 00:31:25,000 --> 00:31:32,000 So the first thing, as soon as I click on the cloudy output button, my face should get loaded from 581 00:31:32,000 --> 00:31:33,000 the local. 582 00:31:33,000 --> 00:31:38,000 So so that is the reason I'm writing dot load underscore local with this file index and the bedrock 583 00:31:38,000 --> 00:31:40,000 embedding, the same embedding which I've actually used. 584 00:31:41,000 --> 00:31:45,000 Then I will go ahead and call this cloudy LLM right. 585 00:31:45,000 --> 00:31:48,000 I'm going to call this cloudy LLM because I will get the LLM itself. 586 00:31:48,000 --> 00:31:54,000 Now if I go inside this function, if I go ahead and write f 12 okay. 587 00:31:54,000 --> 00:31:58,000 So here you will be able to see where it is. 588 00:31:58,000 --> 00:31:59,000 Uh, cloudy. 589 00:31:59,000 --> 00:32:00,000 LM so cloudy. 590 00:32:00,000 --> 00:32:05,000 LM it is going to call this particular model and return our LM over here. 591 00:32:05,000 --> 00:32:05,000 Right. 592 00:32:05,000 --> 00:32:08,000 So here I'm going to get my LM with respect to this. 593 00:32:08,000 --> 00:32:12,000 And this LM will specifically be used inside this function. 594 00:32:12,000 --> 00:32:17,000 That is get response underscore LM where I have given my LM vector store files and query right. 595 00:32:17,000 --> 00:32:20,000 When I get this three information over here, I'll get my response. 596 00:32:20,000 --> 00:32:23,000 And I am writing this particular response over here. 597 00:32:24,000 --> 00:32:25,000 Very much simple. 598 00:32:25,000 --> 00:32:25,000 Very much easy. 599 00:32:25,000 --> 00:32:26,000 Right. 600 00:32:26,000 --> 00:32:30,000 So let's execute this and let's see whether this is working fine or not okay. 601 00:32:30,000 --> 00:32:32,000 So this is my main function. 602 00:32:32,000 --> 00:32:37,000 So what I will do I will create this if underscore underscore name with respect to this mean. 603 00:32:37,000 --> 00:32:40,000 Now let's see whether we will get any error okay. 604 00:32:41,000 --> 00:32:42,000 So I will clear my screen. 605 00:32:43,000 --> 00:32:46,000 But in short we have executed each and every thing right. 606 00:32:46,000 --> 00:32:50,000 So I will go ahead and write Streamlit run app dot Pi. 607 00:32:50,000 --> 00:32:51,000 Let's see whether we will get any errors. 608 00:32:51,000 --> 00:32:55,000 First of all, uh, okay. 609 00:32:55,000 --> 00:32:56,000 It's working fine. 610 00:32:56,000 --> 00:32:57,000 I will close this. 611 00:32:57,000 --> 00:32:58,000 Okay. 612 00:32:58,000 --> 00:33:03,000 Now this on the left hand side, I will get this vector update update or create vector store. 613 00:33:03,000 --> 00:33:06,000 And here I have chat PDF using AWS bedrock. 614 00:33:06,000 --> 00:33:06,000 Okay. 615 00:33:06,000 --> 00:33:11,000 Now the first step as I said what I'm going to do here is my data folder. 616 00:33:11,000 --> 00:33:15,000 It has this PDF file attention dot pdf, yolo dot pdf. 617 00:33:15,000 --> 00:33:18,000 So what I'm going to do I'm going to probably click on Vector Update. 618 00:33:18,000 --> 00:33:20,000 As soon as I click on Vector Update. 619 00:33:20,000 --> 00:33:21,000 Now what is going to happen. 620 00:33:21,000 --> 00:33:22,000 See over here. 621 00:33:22,000 --> 00:33:26,000 As soon as I click on Vector update, what is going to happen. 622 00:33:26,000 --> 00:33:29,000 So if this is clicked that basically means the data ingestion folder. 623 00:33:29,000 --> 00:33:30,000 It will go. 624 00:33:30,000 --> 00:33:36,000 It will read both this PDF and then it will convert that into vectors And we will save that vector over 625 00:33:36,000 --> 00:33:37,000 here. 626 00:33:37,000 --> 00:33:40,000 Right now you cannot see any folder in the name of files index. 627 00:33:40,000 --> 00:33:43,000 But now if I go ahead and click on vector update. 628 00:33:43,000 --> 00:33:44,000 Now what is going to happen. 629 00:33:44,000 --> 00:33:47,000 See as soon as this vector update will finish. 630 00:33:47,000 --> 00:33:52,000 Now processing is basically happening and this will probably come up with like done okay done status. 631 00:33:52,000 --> 00:33:54,000 Unless until we don't get any errors. 632 00:33:54,000 --> 00:33:56,000 Let's see with respect to this right what will happen. 633 00:33:57,000 --> 00:34:00,000 So it is now loading all the PDF it is performing. 634 00:34:00,000 --> 00:34:02,000 It is converting those into vectors. 635 00:34:02,000 --> 00:34:03,000 Right. 636 00:34:03,000 --> 00:34:06,000 And then it has now stored that as a fixed index. 637 00:34:06,000 --> 00:34:12,000 Now if you probably go over here and see this folder my files index has created, it has index dot files 638 00:34:12,000 --> 00:34:13,000 and index dot pickle. 639 00:34:13,000 --> 00:34:15,000 Right now this is the first step. 640 00:34:15,000 --> 00:34:19,000 Now the second step is another button which is basically cloudy output. 641 00:34:19,000 --> 00:34:21,000 Now cloudy output is over here. 642 00:34:21,000 --> 00:34:27,000 Now if I give any prompt what is attention is all you need. 643 00:34:27,000 --> 00:34:29,000 What is attention is all you need. 644 00:34:30,000 --> 00:34:33,000 Okay, if I go ahead and click on Cloudy Output. 645 00:34:33,000 --> 00:34:36,000 Now it is going to call the cloudy model right. 646 00:34:36,000 --> 00:34:40,000 So if you probably go ahead and see the code again what is basically happening. 647 00:34:40,000 --> 00:34:43,000 First we will load this file index from the local. 648 00:34:43,000 --> 00:34:46,000 Then we will call the cloudy LLM model. 649 00:34:46,000 --> 00:34:52,000 Then we will call this Get response LM where we are giving all these three information the user question. 650 00:34:52,000 --> 00:34:56,000 And now finally you'll be able to see I'm getting the output right now. 651 00:34:56,000 --> 00:34:58,000 This is with respect to cloudy output right. 652 00:34:58,000 --> 00:35:03,000 So similarly if I go ahead and write what is YOLO I will go ahead and click on Cloudy Output. 653 00:35:03,000 --> 00:35:06,000 Now again see follow this pattern inside it will go again. 654 00:35:06,000 --> 00:35:08,000 It will load this files index. 655 00:35:08,000 --> 00:35:09,000 Right. 656 00:35:09,000 --> 00:35:12,000 And then we will call this particular cloud model. 657 00:35:12,000 --> 00:35:14,000 Now how the performance can be improved. 658 00:35:14,000 --> 00:35:16,000 This Phi C index I will not call each and every time. 659 00:35:16,000 --> 00:35:18,000 So let me do one thing. 660 00:35:18,000 --> 00:35:22,000 Let me call this once okay. 661 00:35:23,000 --> 00:35:25,000 Let me do one thing. 662 00:35:28,000 --> 00:35:34,000 For the first time I will just keep it outside this okay. 663 00:35:34,000 --> 00:35:36,000 But again it will give us an error. 664 00:35:36,000 --> 00:35:38,000 But I will give you that as an assignment. 665 00:35:38,000 --> 00:35:40,000 First of all, let's work on this okay. 666 00:35:40,000 --> 00:35:42,000 So here you can see my YOLO is also coming. 667 00:35:42,000 --> 00:35:44,000 Now see if I want to add llama two and all. 668 00:35:44,000 --> 00:35:46,000 What I need to do I will go over here. 669 00:35:46,000 --> 00:35:49,000 My llama two function is already created if you remember. 670 00:35:49,000 --> 00:35:54,000 See over here my llama two function is where get llama two right? 671 00:35:54,000 --> 00:35:56,000 So I will create another button. 672 00:35:56,000 --> 00:35:58,000 I will copy it like this. 673 00:35:58,000 --> 00:36:03,000 If my start button is llama two, right? 674 00:36:03,000 --> 00:36:07,000 So this will be my llama two output. 675 00:36:08,000 --> 00:36:12,000 Now instead of calling get cloud lm, I will say get llama two LM. 676 00:36:12,000 --> 00:36:16,000 So now we will go ahead and check whether for llama two it is working fine or not. 677 00:36:16,000 --> 00:36:19,000 So I will save this quickly. 678 00:36:19,000 --> 00:36:22,000 I will see now I'll click on llama two. 679 00:36:22,000 --> 00:36:24,000 Now let's see whether we'll get the output or not. 680 00:36:24,000 --> 00:36:26,000 Again, the same thing is going to happen. 681 00:36:26,000 --> 00:36:29,000 First of all, it is going to load the files index. 682 00:36:29,000 --> 00:36:32,000 After that it is going to take the LM model. 683 00:36:32,000 --> 00:36:33,000 It is going to give all the information. 684 00:36:33,000 --> 00:36:35,000 And then finally you'll be getting the answer. 685 00:36:35,000 --> 00:36:42,000 Similarly I can go ahead and write what is attention is all you need okay. 686 00:36:42,000 --> 00:36:46,000 So if I go ahead and click on llama two, then again I will be able to get the response. 687 00:36:46,000 --> 00:36:50,000 So this entirely everything is basically happening with the AWS bedrock. 688 00:36:50,000 --> 00:36:55,000 So step by step I have shown you almost everything the data ingestion step, what our libraries are 689 00:36:55,000 --> 00:36:58,000 specifically getting used, all this code, I have written it in front of you. 690 00:36:58,000 --> 00:37:03,000 The Streamlit file we have actually created, I would suggest go ahead and try it from your side, and 691 00:37:03,000 --> 00:37:05,000 this is how you are able to get the response. 692 00:37:05,000 --> 00:37:11,000 But the best thing is that all these models are available in AWS bedrock and it is already scalable. 693 00:37:11,000 --> 00:37:13,000 It you can actually use it according to your wish.