1 00:00:00,000 --> 00:00:03,000 So guys, now let's continue the discussion with respect to this particular project. 2 00:00:03,000 --> 00:00:05,000 Now what we are going to implement over here. 3 00:00:05,000 --> 00:00:09,000 So here you will be able to see that I have a research paper folder okay. 4 00:00:09,000 --> 00:00:12,000 And inside this folder I have this two amazing research paper. 5 00:00:12,000 --> 00:00:16,000 That is attention is all you need and overview of large language models. 6 00:00:16,000 --> 00:00:20,000 So I have kept all both these PDFs inside this folder. 7 00:00:21,000 --> 00:00:24,000 And we will consider this as an external data source. 8 00:00:24,000 --> 00:00:26,000 And we will try to ask any question. 9 00:00:26,000 --> 00:00:30,000 And probably we should be able to get the response from this particular research paper through this 10 00:00:30,000 --> 00:00:31,000 document. 11 00:00:31,000 --> 00:00:32,000 Q&A okay. 12 00:00:32,000 --> 00:00:38,000 So how we will be executing this again over here we are going to be specifically use this chat grok 13 00:00:38,000 --> 00:00:38,000 okay. 14 00:00:38,000 --> 00:00:45,000 And uh, along with this you'll also be seeing that, uh, we will go ahead and uh, you know, uh, 15 00:00:45,000 --> 00:00:49,000 implement open source LM model instead of just using open AI. 16 00:00:49,000 --> 00:00:49,000 Okay. 17 00:00:50,000 --> 00:00:53,000 Uh, so that you will also everybody will be able to execute this, Okay. 18 00:00:53,000 --> 00:00:56,000 So quickly, let's go ahead and do this now. 19 00:00:56,000 --> 00:01:04,000 First thing first I will go ahead and write from dot env import load underscore dot env okay. 20 00:01:04,000 --> 00:01:12,000 And then uh we will now just go ahead and execute this load underscore dot env. 21 00:01:12,000 --> 00:01:16,000 Now this is specifically used to load all the environment variables. 22 00:01:16,000 --> 00:01:21,000 Right now I'm going to load the grok grok API. 23 00:01:22,000 --> 00:01:24,000 And this will be very much important. 24 00:01:24,000 --> 00:01:28,000 And for some number of hits grok completely provides you free API itself. 25 00:01:28,000 --> 00:01:32,000 So anybody can probably go ahead and create your account okay. 26 00:01:32,000 --> 00:01:34,000 You have to go ahead and create your account. 27 00:01:34,000 --> 00:01:37,000 And based on that you can actually go ahead and create your keys. 28 00:01:37,000 --> 00:01:43,000 Now here I will go ahead and write OS dot environment dot environ. 29 00:01:43,000 --> 00:01:46,000 And here I'm going to use my grok API key. 30 00:01:46,000 --> 00:01:50,000 So I'll go ahead and copy this I'll go ahead and paste it over here. 31 00:01:50,000 --> 00:01:51,000 Okay. 32 00:01:51,000 --> 00:01:56,000 Along with this what I will specifically do is that I will just go ahead and write is equal to OS dot 33 00:01:56,000 --> 00:01:57,000 get env. 34 00:01:57,000 --> 00:01:57,000 Okay. 35 00:01:58,000 --> 00:02:07,000 Once I do this I will quickly go ahead and write my open uh sorry grok grok API key itself. 36 00:02:07,000 --> 00:02:12,000 Okay, so once I get this grok API key then I will be able to interact with any model that I want okay. 37 00:02:12,000 --> 00:02:18,000 Now uh, once we are setting this up, I can also make sure that I can get the direct key from here. 38 00:02:18,000 --> 00:02:19,000 Okay. 39 00:02:19,000 --> 00:02:22,000 Um, so let's go to the next step. 40 00:02:22,000 --> 00:02:24,000 And here, uh, let me just go ahead and write. 41 00:02:24,000 --> 00:02:28,000 This is my grok API underscore key, which is nothing. 42 00:02:28,000 --> 00:02:34,000 But if I just go ahead and write os dot get env, it is just going to get it from the environment variable. 43 00:02:34,000 --> 00:02:34,000 Okay. 44 00:02:34,000 --> 00:02:36,000 So this way also you can actually get your API key. 45 00:02:36,000 --> 00:02:40,000 Now is the most important step is creating your LM model. 46 00:02:40,000 --> 00:02:42,000 For this we will be using this Chad Grok. 47 00:02:42,000 --> 00:02:50,000 And here we have to first of all provide our grok API key which will be equal to nothing, but it will 48 00:02:50,000 --> 00:02:52,000 be equal to this grok API key. 49 00:02:52,000 --> 00:02:52,000 Right. 50 00:02:52,000 --> 00:02:55,000 So this is the first parameter that is required over here. 51 00:02:55,000 --> 00:02:57,000 And the second parameter is nothing but my model name. 52 00:02:58,000 --> 00:03:03,000 So in order to provide the model name I will go back to my website of grok. 53 00:03:03,000 --> 00:03:07,000 And here you'll be able to see if I go to the first home page. 54 00:03:07,000 --> 00:03:11,000 Here are different different models that you can specifically use right. 55 00:03:11,000 --> 00:03:15,000 So you can provide this llama three 8,000,000,008 192. 56 00:03:15,000 --> 00:03:18,000 You can provide llama three 7,000,000,008 192 gamma seven. 57 00:03:18,000 --> 00:03:19,000 Be it. 58 00:03:19,000 --> 00:03:23,000 So let's go ahead and select this I will go ahead and write like this on itself. 59 00:03:23,000 --> 00:03:27,000 Gamma dash seven B dash it okay. 60 00:03:27,000 --> 00:03:29,000 So I will just go ahead and write in the same way. 61 00:03:29,000 --> 00:03:33,000 So here you have this gamma. 62 00:03:35,000 --> 00:03:38,000 7BIT okay. 63 00:03:38,000 --> 00:03:41,000 So we are basically going to use this LLM model like this. 64 00:03:41,000 --> 00:03:44,000 You can also go ahead and specify your llama three model if you want. 65 00:03:44,000 --> 00:03:47,000 So let's say I want to go ahead and use llama three model. 66 00:03:47,000 --> 00:03:48,000 I can go ahead and specify like this. 67 00:03:48,000 --> 00:03:52,000 So let's continue with llama three so that uh it's okay. 68 00:03:52,000 --> 00:03:53,000 You can use any of the open source models. 69 00:03:53,000 --> 00:03:54,000 It is up to you. 70 00:03:54,000 --> 00:03:58,000 Now I'll go ahead and define my chat prompt template. 71 00:03:58,000 --> 00:03:58,000 Okay. 72 00:03:58,000 --> 00:04:03,000 Now inside this chat prompt template I'm just going to use dot from underscore template. 73 00:04:04,000 --> 00:04:06,000 Now is the most important thing. 74 00:04:06,000 --> 00:04:09,000 Like see the previous ways how we used to create chat prompt template. 75 00:04:09,000 --> 00:04:14,000 Now this will be a little bit different because here we need to provide some context information also. 76 00:04:14,000 --> 00:04:19,000 So here I'm just going to say hey um I'll go ahead and write it over here. 77 00:04:19,000 --> 00:04:32,000 I'll say, hey, answer the questions based on the provided context only Okay. 78 00:04:32,000 --> 00:04:35,000 And then I will say next sentence. 79 00:04:35,000 --> 00:04:43,000 Please provide the most accurate response. 80 00:04:43,000 --> 00:04:48,000 Response based on the question. 81 00:04:48,000 --> 00:04:48,000 Right. 82 00:04:48,000 --> 00:04:52,000 And here I will just go ahead and define my context. 83 00:04:52,000 --> 00:04:54,000 This will basically be my context. 84 00:04:54,000 --> 00:04:55,000 Here I will give a placeholder. 85 00:04:55,000 --> 00:04:58,000 Now now why I am just using this keyword. 86 00:04:58,000 --> 00:05:02,000 We will try to understand okay, because this is important. 87 00:05:02,000 --> 00:05:07,000 The libraries that we have used this create stuff document chain and create retrieval chain. 88 00:05:07,000 --> 00:05:12,000 This has a separate meaning for this kind of context message that is seen in the prompt template. 89 00:05:12,000 --> 00:05:14,000 Okay we'll discuss more about it. 90 00:05:14,000 --> 00:05:17,000 And then here I will just go ahead and write my question. 91 00:05:17,000 --> 00:05:20,000 This question will basically have my input. 92 00:05:20,000 --> 00:05:21,000 Okay. 93 00:05:22,000 --> 00:05:24,000 Now here is my prompt. 94 00:05:24,000 --> 00:05:27,000 So here in this prompt it is just like a document Q&A. 95 00:05:27,000 --> 00:05:30,000 In case of document Q&A I need to provide a context over here. 96 00:05:30,000 --> 00:05:33,000 What this context will be replaced with. 97 00:05:33,000 --> 00:05:35,000 I will just discuss about it once we proceed. 98 00:05:35,000 --> 00:05:36,000 Okay. 99 00:05:36,000 --> 00:05:39,000 So this is my prompt, a simple prompt that I have actually written. 100 00:05:40,000 --> 00:05:45,000 Now the first thing, as you know, I have to create my vectors, right? 101 00:05:45,000 --> 00:05:47,000 Create my vectors. 102 00:05:47,000 --> 00:05:49,000 And this time I'll be creating my vectors. 103 00:05:49,000 --> 00:05:51,000 I'll say vector embeddings. 104 00:05:51,000 --> 00:05:52,000 Okay. 105 00:05:52,000 --> 00:05:59,000 Now inside this function, my main aim will be that I should be able to read all the documents from 106 00:05:59,000 --> 00:05:59,000 here. 107 00:05:59,000 --> 00:06:04,000 After reading the documents, I should be able to read the content inside this. 108 00:06:04,000 --> 00:06:10,000 I mean, read this research paper, read the content inside this and after reading it make sure apply 109 00:06:10,000 --> 00:06:14,000 recursive character text splitter along with that, split all this particular documents and just by 110 00:06:14,000 --> 00:06:22,000 using files, we can proceed further and we can go ahead and probably store them in some kind of vector 111 00:06:22,000 --> 00:06:22,000 database. 112 00:06:22,000 --> 00:06:23,000 Okay. 113 00:06:23,000 --> 00:06:29,000 So here what I will do, I will quickly go ahead and use uh, define this function which is called as 114 00:06:29,000 --> 00:06:30,000 create vector embedding. 115 00:06:30,000 --> 00:06:35,000 But here since my vector store will be used in some other, uh, code. 116 00:06:35,000 --> 00:06:39,000 Also, what I will do is that I will try to use session state over here. 117 00:06:39,000 --> 00:06:45,000 Session states actually helps you to remember uh, your vector store DB specifically this is for vector 118 00:06:45,000 --> 00:06:45,000 store DB. 119 00:06:45,000 --> 00:06:48,000 That is the reason I'm trying to create some session state. 120 00:06:48,000 --> 00:06:52,000 This session state variables will be able to access in other functions also whenever it is required. 121 00:06:52,000 --> 00:06:58,000 Okay, this is just to make sure that your entire application has some kind of memory with respect to 122 00:06:58,000 --> 00:06:59,000 this particular chatbot. 123 00:06:59,000 --> 00:07:09,000 Okay, so here I will say if my vector's not in s t dot session state, I'm just going to try to create 124 00:07:09,000 --> 00:07:14,000 a vectors variable, which will be my session state variable. 125 00:07:14,000 --> 00:07:17,000 So I'll say TX dot session underscore state. 126 00:07:18,000 --> 00:07:19,000 Okay. 127 00:07:19,000 --> 00:07:27,000 Then I will just go ahead and write it over here, and I will just go ahead and say TT dot session underscore 128 00:07:27,000 --> 00:07:28,000 state. 129 00:07:28,000 --> 00:07:31,000 First of all, I will go ahead and create my embeddings. 130 00:07:31,000 --> 00:07:36,000 Now in this case my embeddings that I'm actually going to use is nothing but my embeddings. 131 00:07:36,000 --> 00:07:36,000 Okay. 132 00:07:36,000 --> 00:07:42,000 Now whatever embedding I'm basically using, I will also keep this in the session state variable right. 133 00:07:42,000 --> 00:07:45,000 Specifically in this particular session state. 134 00:07:45,000 --> 00:07:49,000 Then we will quickly go ahead and write SD dot session state. 135 00:07:49,000 --> 00:07:52,000 Another session state we will go ahead and create. 136 00:07:53,000 --> 00:07:56,000 And inside this I will go ahead and write a loader. 137 00:07:56,000 --> 00:07:57,000 Right. 138 00:07:57,000 --> 00:08:00,000 This loader will be my data ingestion step right now. 139 00:08:00,000 --> 00:08:06,000 In this case I'm going to use this pi pdf directory loader okay pi pdf directory loader. 140 00:08:06,000 --> 00:08:11,000 And inside this I'm going to give my folder name. 141 00:08:11,000 --> 00:08:12,000 What is the folder name that you have? 142 00:08:12,000 --> 00:08:16,000 It is nothing but research underscore paper right. 143 00:08:16,000 --> 00:08:20,000 So this is the folder name that we will be giving over here okay. 144 00:08:20,000 --> 00:08:21,000 So let's go ahead and give this. 145 00:08:21,000 --> 00:08:26,000 So here you can see this is my research research papers papers I should right okay. 146 00:08:27,000 --> 00:08:32,000 Now this is nothing but this is my data ingestion step. 147 00:08:33,000 --> 00:08:33,000 Perfect. 148 00:08:33,000 --> 00:08:36,000 We are loading something, loading some information. 149 00:08:36,000 --> 00:08:37,000 This can also come from Streamlit directly. 150 00:08:37,000 --> 00:08:39,000 It is up to you right? 151 00:08:39,000 --> 00:08:39,000 How? 152 00:08:39,000 --> 00:08:41,000 You basically want to go ahead and upload this. 153 00:08:42,000 --> 00:08:42,000 Now. 154 00:08:42,000 --> 00:08:49,000 After this I will just go ahead and say st dot session, underscore state dot documents. 155 00:08:49,000 --> 00:08:53,000 And uh from this loader I'm going to load all the documents. 156 00:08:53,000 --> 00:09:00,000 So for that again I have to call this st dot session state dot loader dot load. 157 00:09:01,000 --> 00:09:03,000 I will load all the documents over here. 158 00:09:03,000 --> 00:09:09,000 And uh this will basically be my complete document loading. 159 00:09:09,000 --> 00:09:12,000 I'll just write this comment so that you'll be able to understand it okay. 160 00:09:12,000 --> 00:09:18,000 Then uh, what we are going to do is that I'll go ahead and write s t dot session underscore state dot 161 00:09:18,000 --> 00:09:19,000 text splitter. 162 00:09:19,000 --> 00:09:21,000 I have to use some kind of text splitter over here. 163 00:09:21,000 --> 00:09:27,000 And this text splitter will be nothing, but it will be assigned to a recursive character text splitter, 164 00:09:27,000 --> 00:09:29,000 which we have already discussed. 165 00:09:29,000 --> 00:09:33,000 Inside this there will be a chunk size of 1000, let's say. 166 00:09:34,000 --> 00:09:36,000 And then I'm going to use a chunk. 167 00:09:39,000 --> 00:09:42,000 Overlap of 200 okay. 168 00:09:43,000 --> 00:09:47,000 I'm just going to use this chunk uh, with respect to this okay. 169 00:09:47,000 --> 00:09:47,000 Okay. 170 00:09:47,000 --> 00:09:48,000 Perfect. 171 00:09:48,000 --> 00:09:51,000 Uh, so this is my recursive character text splitter, which I'm actually doing it. 172 00:09:51,000 --> 00:09:59,000 And then finally it is just dot session underscore state dot final underscore documents. 173 00:09:59,000 --> 00:10:04,000 This will be my final documents because I have to apply this text splitting technique again I will go 174 00:10:04,000 --> 00:10:06,000 ahead and save everything in the session. 175 00:10:06,000 --> 00:10:10,000 That is what I'm actually doing, so that my entire application has some kind of memory whenever it 176 00:10:10,000 --> 00:10:12,000 has to refer to. 177 00:10:12,000 --> 00:10:17,000 Okay, now I will be using this session state dot text splitter. 178 00:10:19,000 --> 00:10:21,000 And I will use this split text splitter. 179 00:10:21,000 --> 00:10:23,000 Dot split documents. 180 00:10:23,000 --> 00:10:27,000 It is called as split documents. 181 00:10:27,000 --> 00:10:31,000 This way we are actually splitting all our documents right. 182 00:10:31,000 --> 00:10:35,000 So for this I will again where my documents is basically present. 183 00:10:35,000 --> 00:10:36,000 It is present over here. 184 00:10:36,000 --> 00:10:37,000 Okay. 185 00:10:37,000 --> 00:10:43,000 Just to make sure that this will not take much time, I will just take, uh, this documents and I'll 186 00:10:43,000 --> 00:10:46,000 say, hey, starting 50 documents, just go ahead and do it right. 187 00:10:46,000 --> 00:10:51,000 You can also do the remaining one so that it will not take much time for doing this entire processing. 188 00:10:51,000 --> 00:10:51,000 Okay. 189 00:10:51,000 --> 00:10:53,000 That is the reason I have just given top 50. 190 00:10:53,000 --> 00:11:00,000 Uh, now I will go ahead and write s t dot session underscore state dot vectors. 191 00:11:00,000 --> 00:11:03,000 And here we are going to basically use for this. 192 00:11:04,000 --> 00:11:08,000 See this last step is all about converting your final documents. 193 00:11:08,000 --> 00:11:11,000 First of all converting that into vectors. 194 00:11:11,000 --> 00:11:16,000 And after you convert everything into vectors, all the text or documents into vector, then you try 195 00:11:16,000 --> 00:11:18,000 to store it in some kind of vector database. 196 00:11:18,000 --> 00:11:19,000 Okay. 197 00:11:19,000 --> 00:11:25,000 And this vectors that you are seeing SD dot session dot vectors sessions underscore state dot vectors. 198 00:11:25,000 --> 00:11:27,000 This is your vector database. 199 00:11:27,000 --> 00:11:31,000 So I will go ahead and write files dot from underscore documents. 200 00:11:32,000 --> 00:11:39,000 And here you have SD dot session state dot final underscore documents. 201 00:11:40,000 --> 00:11:44,000 It is final underscore documents. 202 00:11:44,000 --> 00:11:49,000 And here I'm also going to the second parameter we need to give is our embedding technique. 203 00:11:49,000 --> 00:11:50,000 Right. 204 00:11:50,000 --> 00:11:57,000 So if I go ahead and write SD dot session underscore state dot embedding whatever embedding techniques 205 00:11:57,000 --> 00:11:59,000 I have actually used. 206 00:11:59,000 --> 00:12:00,000 Right. 207 00:12:00,000 --> 00:12:01,000 That's it. 208 00:12:01,000 --> 00:12:02,000 You know. 209 00:12:02,000 --> 00:12:07,000 So finally after executing all these things, you'll be seeing that you're converting this into vectors. 210 00:12:07,000 --> 00:12:12,000 You don't even have to return this vectors because already we are saving this in the session state. 211 00:12:12,000 --> 00:12:12,000 Perfect. 212 00:12:12,000 --> 00:12:15,000 Now quickly, let's go ahead and do this. 213 00:12:15,000 --> 00:12:17,000 I will go ahead and create my prompt. 214 00:12:17,000 --> 00:12:22,000 And here I will write SD dot text underscore input okay. 215 00:12:22,000 --> 00:12:30,000 And here we are going to basically use enter your enter your query okay. 216 00:12:30,000 --> 00:12:34,000 From the documents or from the research paper I'll say okay. 217 00:12:36,000 --> 00:12:39,000 And then finally, uh, I will go ahead and do this. 218 00:12:39,000 --> 00:12:44,000 I'll go ahead and just write if s t dot button, okay? 219 00:12:44,000 --> 00:12:50,000 And I'll say, hey, document embedding and I'll just create a button, okay. 220 00:12:50,000 --> 00:12:50,000 Over there. 221 00:12:51,000 --> 00:12:55,000 This button will be responsible in running this particular function and creating our vector store. 222 00:12:55,000 --> 00:12:56,000 Okay. 223 00:12:56,000 --> 00:13:00,000 So as soon as we run this this is basically going to read all the research papers. 224 00:13:00,000 --> 00:13:02,000 It is going to probably convert that into documents. 225 00:13:02,000 --> 00:13:05,000 And then it is finally going to convert into a vectors. 226 00:13:05,000 --> 00:13:05,000 Okay. 227 00:13:05,000 --> 00:13:11,000 So here, uh, what I will do is that I'll just go ahead and call this and I'll say hey go ahead and 228 00:13:11,000 --> 00:13:13,000 call this create vector embedding. 229 00:13:13,000 --> 00:13:18,000 And after doing that, if once it is successful I'll go ahead and say read dot write. 230 00:13:19,000 --> 00:13:22,000 Your vector database is ready okay. 231 00:13:22,000 --> 00:13:25,000 So that you will be able to query from it okay. 232 00:13:25,000 --> 00:13:29,000 Now it's time to check the Power of Grok API. 233 00:13:29,000 --> 00:13:35,000 Okay, so once your vector database is ready, uh, the next step will be that, uh, we will go ahead 234 00:13:35,000 --> 00:13:42,000 and check our, uh, you know, the specifically we'll check, uh, that how powerful is the grok API? 235 00:13:42,000 --> 00:13:43,000 So that is the reason I'm importing time. 236 00:13:43,000 --> 00:13:47,000 So here, uh, I will just go ahead and rename this to user underscore prompt. 237 00:13:47,000 --> 00:13:48,000 Okay. 238 00:13:48,000 --> 00:13:53,000 So finally if I give this user underscore prompt if user is giving any prompt okay. 239 00:13:54,000 --> 00:13:56,000 Now is the most important step. 240 00:13:56,000 --> 00:13:58,000 What should basically happen. 241 00:13:58,000 --> 00:14:05,000 You know that my vectors this is t dot session state dot vectors has my vector db vector store db right 242 00:14:05,000 --> 00:14:08,000 now when as soon as my user prompt. 243 00:14:08,000 --> 00:14:13,000 Basically user is writing some input and pressing some enter or pressing enter right? 244 00:14:13,000 --> 00:14:16,000 It should first of all create this stuff document chain. 245 00:14:16,000 --> 00:14:18,000 And now let's discuss about this stuff. 246 00:14:18,000 --> 00:14:19,000 Document change what it is. 247 00:14:19,000 --> 00:14:22,000 Here we are going to give two important parameters. 248 00:14:22,000 --> 00:14:27,000 One is LM model and the second one is nothing but our prompt okay. 249 00:14:27,000 --> 00:14:29,000 So this should be the prompt. 250 00:14:29,000 --> 00:14:33,000 Now let me just go ahead and define this variable document underscore chain. 251 00:14:33,000 --> 00:14:36,000 Now see what this exactly does. 252 00:14:36,000 --> 00:14:38,000 Create stuff underscore document chain. 253 00:14:38,000 --> 00:14:42,000 It creates a chain for passing a list of documents to the model. 254 00:14:42,000 --> 00:14:42,000 Okay. 255 00:14:42,000 --> 00:14:47,000 What it is basically doing, it creates a chain for passing a list of documents to the model. 256 00:14:47,000 --> 00:14:51,000 And here the two important parameters are LM and prompt. 257 00:14:52,000 --> 00:14:56,000 As soon as I give this here, you'll be able to see that in the prompt you have this variable called 258 00:14:56,000 --> 00:14:57,000 as context. 259 00:14:57,000 --> 00:15:03,000 Right inside this context, all that list of documents will specifically go and whatever specific input 260 00:15:03,000 --> 00:15:06,000 you are giving it will go over here, right how it will go. 261 00:15:06,000 --> 00:15:11,000 I will just combine it further with one more important retrieval chain, which will discuss about okay, 262 00:15:11,000 --> 00:15:16,000 but in short, this is providing a list of documents in specifically in this particular context. 263 00:15:16,000 --> 00:15:17,000 Like it will get appended over here. 264 00:15:17,000 --> 00:15:21,000 And this entire chat prompt template will be created from here itself. 265 00:15:21,000 --> 00:15:22,000 Okay. 266 00:15:22,000 --> 00:15:26,000 Now further, we will not keep quiet with this document chain here. 267 00:15:26,000 --> 00:15:33,000 What I will do is that I will go ahead and write st dot session underscore state okay dot vectors okay. 268 00:15:33,000 --> 00:15:35,000 You know this it is a vectors is basically there. 269 00:15:36,000 --> 00:15:41,000 I will go ahead and write this as as retriever. 270 00:15:41,000 --> 00:15:44,000 Now you know that we need to convert. 271 00:15:44,000 --> 00:15:49,000 Uh, if I want to retrieve any information from my vector db, I need to first of all convert that into 272 00:15:49,000 --> 00:15:50,000 a retriever. 273 00:15:50,000 --> 00:15:51,000 Right. 274 00:15:51,000 --> 00:15:56,000 So if I go ahead and write as retriever, let me just check the spelling. 275 00:15:56,000 --> 00:15:57,000 I hope it is fine. 276 00:16:00,000 --> 00:16:09,000 Okay so just a second as underscore retriever. 277 00:16:09,000 --> 00:16:09,000 Okay. 278 00:16:09,000 --> 00:16:16,000 So I'm just going to convert this into retriever and finally save this particular value in the form 279 00:16:16,000 --> 00:16:18,000 of a retriever variable. 280 00:16:18,000 --> 00:16:24,000 And what this retriever acts as, it acts as an interface to pass to query your vector database. 281 00:16:24,000 --> 00:16:25,000 In short. 282 00:16:25,000 --> 00:16:25,000 Right. 283 00:16:25,000 --> 00:16:27,000 So this is my vectors over here. 284 00:16:27,000 --> 00:16:30,000 I've used this t dot embeddings techniques that I've used over here. 285 00:16:30,000 --> 00:16:31,000 This is perfect. 286 00:16:31,000 --> 00:16:38,000 Now once I get this as a retriever you know what we do is that obviously this retriever is like querying 287 00:16:38,000 --> 00:16:39,000 a specific vector DB. 288 00:16:39,000 --> 00:16:43,000 This document chain has the context of a list of entire documents. 289 00:16:43,000 --> 00:16:49,000 So once we combine both of them, in short we are going to create a retrieval chain that is create retriever 290 00:16:49,000 --> 00:16:49,000 chain. 291 00:16:49,000 --> 00:16:51,000 And here two parameters are basically given. 292 00:16:51,000 --> 00:16:54,000 One is retriever and the second one is document chain. 293 00:16:54,000 --> 00:16:55,000 Okay. 294 00:16:55,000 --> 00:16:59,000 Once we do this we will be able to invoke this entire thing right. 295 00:16:59,000 --> 00:17:02,000 So here I will go ahead and write retrieval underscore chain. 296 00:17:02,000 --> 00:17:02,000 Right. 297 00:17:02,000 --> 00:17:07,000 So this actually is my chain which has document chain which has retrieval combined together. 298 00:17:07,000 --> 00:17:08,000 Okay. 299 00:17:08,000 --> 00:17:10,000 Now let's go ahead and see the power of this grok API. 300 00:17:10,000 --> 00:17:11,000 You'll just understand. 301 00:17:11,000 --> 00:17:19,000 So first of all I will just go ahead and write a start timer over here I'll say time dot process time 302 00:17:19,000 --> 00:17:21,000 process underscore time. 303 00:17:21,000 --> 00:17:25,000 And once we process this you will be able to see my response over here. 304 00:17:25,000 --> 00:17:29,000 And I'll go ahead and write my retrieval chain. 305 00:17:29,000 --> 00:17:29,000 Okay. 306 00:17:29,000 --> 00:17:32,000 Dot invoke okay. 307 00:17:32,000 --> 00:17:34,000 And you know what kind of input I'm actually giving. 308 00:17:34,000 --> 00:17:37,000 It is nothing but input over here. 309 00:17:37,000 --> 00:17:37,000 Right. 310 00:17:37,000 --> 00:17:39,000 This is the same input that we are giving over here. 311 00:17:39,000 --> 00:17:39,000 Right. 312 00:17:39,000 --> 00:17:42,000 The same variable, the same input we are giving over here. 313 00:17:42,000 --> 00:17:46,000 And here we are going to just replace this with user prompt. 314 00:17:47,000 --> 00:17:51,000 Now we will be able to get the answer after this. 315 00:17:51,000 --> 00:17:55,000 But before that we will go ahead and calculate our response time. 316 00:17:55,000 --> 00:17:55,000 Okay. 317 00:17:56,000 --> 00:17:57,000 So let's go ahead and print. 318 00:18:00,000 --> 00:18:02,000 I'll say, hey, uh, I'll use a f string. 319 00:18:02,000 --> 00:18:06,000 I'll say response time is nothing, but it is equal to. 320 00:18:08,000 --> 00:18:10,000 It is equal to. 321 00:18:11,000 --> 00:18:18,000 I will go ahead and write over here time dot process underscore time minus start time. 322 00:18:18,000 --> 00:18:25,000 Okay, once I do this I will be able to get my, uh, the time how much it time how much time it has 323 00:18:25,000 --> 00:18:28,000 basically taken for this retrieval chain to get executed. 324 00:18:28,000 --> 00:18:31,000 And then finally we go ahead and write our response. 325 00:18:31,000 --> 00:18:38,000 So I'll go ahead and write TX dot write let's say response off answer okay. 326 00:18:38,000 --> 00:18:39,000 Done. 327 00:18:40,000 --> 00:18:47,000 Finally we will go ahead and write uh, with uh c you also need to know that what kind of response, 328 00:18:47,000 --> 00:18:49,000 what kind of other context we usually get. 329 00:18:49,000 --> 00:18:49,000 Right. 330 00:18:49,000 --> 00:18:52,000 So for that also you can actually go ahead and display it. 331 00:18:52,000 --> 00:18:57,000 So I'll go ahead and write Streamlit with a Streamlit uh expander. 332 00:18:57,000 --> 00:19:02,000 And here I'm just going to write with St Dot expander. 333 00:19:02,000 --> 00:19:02,000 the. 334 00:19:05,000 --> 00:19:10,000 Document similarity search. 335 00:19:12,000 --> 00:19:13,000 Okay. 336 00:19:13,000 --> 00:19:19,000 So I will go ahead and write for I comma doc in enumerate okay. 337 00:19:19,000 --> 00:19:23,000 And we basically take C along with this response of answer no. 338 00:19:23,000 --> 00:19:27,000 There will be another one variable which is called as response of context. 339 00:19:27,000 --> 00:19:32,000 That will actually give you all the relevant context information from that particular documents. 340 00:19:32,000 --> 00:19:32,000 Okay. 341 00:19:32,000 --> 00:19:34,000 We'll go ahead and write this also. 342 00:19:34,000 --> 00:19:42,000 So I'll go ahead and write SD dot write doc dot page underscore content okay. 343 00:19:42,000 --> 00:19:48,000 Now quickly let me just go ahead and uh make sure that right all the other information I'm just displaying 344 00:19:48,000 --> 00:19:51,000 what all other context is basically coming over here. 345 00:19:51,000 --> 00:19:52,000 I'll just give this. 346 00:19:52,000 --> 00:19:56,000 So one final change that I really wanted to do. 347 00:19:56,000 --> 00:19:56,000 Okay. 348 00:19:56,000 --> 00:20:00,000 So after writing this code, see we are using all of my embeddings. 349 00:20:00,000 --> 00:20:04,000 Uh, it is of uh, no worries, you can definitely use it. 350 00:20:04,000 --> 00:20:09,000 But sometimes, you know, since this is running in our local machine, you know, it may definitely 351 00:20:09,000 --> 00:20:17,000 take a lot of time to probably, you know, convert that into, you know, embeddings specifically convert 352 00:20:17,000 --> 00:20:18,000 all the text into vectors. 353 00:20:18,000 --> 00:20:23,000 So that is the reason what I will do is that over here, instead of using my embedding, I will just 354 00:20:23,000 --> 00:20:27,000 go ahead and use OpenAI embedding because OpenAI embedding will be in the form of APIs. 355 00:20:28,000 --> 00:20:33,000 Uh, the reason why I'm doing this is that because see, if I don't do this automatically, it is going 356 00:20:33,000 --> 00:20:37,000 to take a lot of time just to probably, um, uh, convert all our text into vectors. 357 00:20:37,000 --> 00:20:41,000 Along with this, we will go ahead and load our opening AI API key. 358 00:20:41,000 --> 00:20:42,000 Okay. 359 00:20:42,000 --> 00:20:44,000 So please make sure that you do this. 360 00:20:44,000 --> 00:20:46,000 You can go ahead and try with all of my embedding also. 361 00:20:46,000 --> 00:20:51,000 But as soon as you will probably run it, you will be seeing that it will take a lot of time, you know, 362 00:20:51,000 --> 00:20:53,000 and I have already done that. 363 00:20:53,000 --> 00:20:57,000 So that is the reason I'm saying you, uh, but I think with the help of OpenAI embedding, it should 364 00:20:57,000 --> 00:20:58,000 happen very much quickly. 365 00:20:58,000 --> 00:20:59,000 Right. 366 00:20:59,000 --> 00:21:01,000 So let's go ahead and execute now. 367 00:21:01,000 --> 00:21:03,000 So I will just go ahead and clear my screen. 368 00:21:03,000 --> 00:21:07,000 I'll say, hey Streamlit, run App.py. 369 00:21:07,000 --> 00:21:13,000 Okay, so now my entire website is over here. 370 00:21:13,000 --> 00:21:15,000 I will click on Document Embedding. 371 00:21:15,000 --> 00:21:16,000 Let's wait for some time. 372 00:21:16,000 --> 00:21:20,000 Now as soon as I click on Document embedding, see what is happening inside the hood. 373 00:21:20,000 --> 00:21:21,000 Right. 374 00:21:21,000 --> 00:21:22,000 You should really understand this. 375 00:21:22,000 --> 00:21:25,000 So here this function has got created. 376 00:21:25,000 --> 00:21:28,000 Now this session state variable will get created. 377 00:21:28,000 --> 00:21:30,000 We are going to create embeddings. 378 00:21:30,000 --> 00:21:31,000 Then it will become loader. 379 00:21:31,000 --> 00:21:33,000 It will read all the research papers inside this. 380 00:21:33,000 --> 00:21:35,000 Then you have this documents. 381 00:21:35,000 --> 00:21:36,000 Then you have this text splitter. 382 00:21:37,000 --> 00:21:37,000 Similarly. 383 00:21:37,000 --> 00:21:40,000 Finally you will be able to create your vector database. 384 00:21:40,000 --> 00:21:40,000 Right. 385 00:21:40,000 --> 00:21:42,000 And all these things are there. 386 00:21:42,000 --> 00:21:43,000 Perfect. 387 00:21:43,000 --> 00:21:44,000 It looks good enough. 388 00:21:44,000 --> 00:21:47,000 Let's see whether it has got executed or not. 389 00:21:47,000 --> 00:21:51,000 So yes here you can see the vector database is ready. 390 00:21:51,000 --> 00:21:59,000 But if you try to just use uh uh, you know, embedding it is going to take forever because that entire 391 00:21:59,000 --> 00:22:01,000 embedding is going to just happen in a local machine. 392 00:22:01,000 --> 00:22:06,000 So let me just go ahead and ask, what is Transformers? 393 00:22:06,000 --> 00:22:09,000 I'll ask a question, what is a Transformers? 394 00:22:09,000 --> 00:22:11,000 And I will go ahead and press enter. 395 00:22:12,000 --> 00:22:16,000 So here you can see based on the provided context, transformer is a model architecture that relies 396 00:22:16,000 --> 00:22:21,000 entirely on attention mechanism to draw global dependency between input and output. 397 00:22:21,000 --> 00:22:22,000 Right. 398 00:22:22,000 --> 00:22:24,000 And here is your document similarity search. 399 00:22:24,000 --> 00:22:25,000 Right. 400 00:22:25,000 --> 00:22:32,000 So from all this context, it was able to retrieve from the uh research paper, which is amazing right. 401 00:22:32,000 --> 00:22:33,000 This looks quite good. 402 00:22:33,000 --> 00:22:34,000 Right. 403 00:22:34,000 --> 00:22:41,000 Uh, let's say uh what I'll go ahead and write what is a large language model. 404 00:22:41,000 --> 00:22:41,000 Okay. 405 00:22:42,000 --> 00:22:44,000 And I'll go ahead and press enter. 406 00:22:44,000 --> 00:22:46,000 Or based on the context. 407 00:22:46,000 --> 00:22:47,000 So much information is there. 408 00:22:47,000 --> 00:22:49,000 And here you can also see your entire context. 409 00:22:49,000 --> 00:22:51,000 Information is also visible. 410 00:22:51,000 --> 00:22:51,000 Right. 411 00:22:52,000 --> 00:22:53,000 Pretty much nice. 412 00:22:53,000 --> 00:22:56,000 So you're able to see this entire information. 413 00:22:56,000 --> 00:23:01,000 See uh if you want to get more larger context information, uh, more larger information with respect 414 00:23:01,000 --> 00:23:07,000 to this research paper, what you can do is that you can go to the code and you can change, uh, this 415 00:23:07,000 --> 00:23:12,000 things the chunk size, you can keep the chunk size to 10,000, and you can probably keep the overlap 416 00:23:12,000 --> 00:23:14,000 to 500 and keep on doing this. 417 00:23:14,000 --> 00:23:14,000 Right. 418 00:23:14,000 --> 00:23:17,000 But here, uh, in short, we have actually got this. 419 00:23:17,000 --> 00:23:20,000 Uh, again, I'll tell you why I have used OpenAI document embedding. 420 00:23:20,000 --> 00:23:25,000 The open source LM model is specifically used, but other than that, I think everything looks fine. 421 00:23:25,000 --> 00:23:25,000 Okay. 422 00:23:26,000 --> 00:23:28,000 Um, uh, let me just go ahead and write. 423 00:23:28,000 --> 00:23:33,000 What is attention is all you need, okay. 424 00:23:33,000 --> 00:23:37,000 If I want to ask this question, I will just go ahead and press enter. 425 00:23:37,000 --> 00:23:40,000 And this is your entire information, right. 426 00:23:40,000 --> 00:23:43,000 And all the related context information is over here. 427 00:23:43,000 --> 00:23:44,000 See? 428 00:23:45,000 --> 00:23:46,000 Everything is visible. 429 00:23:47,000 --> 00:23:48,000 This is good enough over here. 430 00:23:48,000 --> 00:23:52,000 I think you are able to understand this specific project. 431 00:23:52,000 --> 00:23:54,000 Um, this was it for my side. 432 00:23:55,000 --> 00:23:57,000 I'll see you all in the next video. 433 00:23:57,000 --> 00:23:57,000 Thank you.