1 00:00:03,410 --> 00:00:06,470 Let's talk about models in land change. 2 00:00:06,470 --> 00:00:13,340 So when land change talks about models in the land change documentation, what are they talking about? 3 00:00:14,180 --> 00:00:20,990 They are talking about the foundational models upon which LM applications are built. 4 00:00:20,990 --> 00:00:30,860 So what we call foundation models, foundation LM models so long chain connects with two types of models. 5 00:00:30,860 --> 00:00:37,970 So in the long chain world a long chain connects with two kinds of models. 6 00:00:37,970 --> 00:00:45,200 First what they call LMS and second, what they call chat models. 7 00:00:46,100 --> 00:00:54,620 LMS for long chain are models that receive text and respond with text and chat models. 8 00:00:54,620 --> 00:01:02,120 For long chain a receive a chat message and respond with a chat message. 9 00:01:02,120 --> 00:01:12,950 So in the code on the right side of the screen, we are going to try an example a connecting long chain 10 00:01:12,950 --> 00:01:21,950 with the OpenAI LM model, and also connecting long chain with llama to an open source model. 11 00:01:22,550 --> 00:01:33,590 So the first thing we do, as always, is to, uh, connect with the env file in order to load the OpenAI 12 00:01:33,590 --> 00:01:34,190 key. 13 00:01:34,580 --> 00:01:44,870 And after that, what we do is we load the OpenAI module, we create an instance of our LM, and we 14 00:01:44,870 --> 00:01:52,220 just send any question to that instance, and we will have the response from a chat. 15 00:01:52,220 --> 00:01:57,710 GPT in this case we can do the same thing for llama too. 16 00:01:57,710 --> 00:02:01,040 In the case of llama too, we have two options. 17 00:02:01,040 --> 00:02:07,430 We can run llama to locally, or we can use llama to in the cloud. 18 00:02:08,270 --> 00:02:09,199 It's best. 19 00:02:09,199 --> 00:02:10,340 The second option. 20 00:02:10,340 --> 00:02:11,360 The first option. 21 00:02:11,360 --> 00:02:15,950 Here you have a link to do that using a llama I. 22 00:02:15,980 --> 00:02:22,490 But it's too complex and uh, I would not recommend recommend that way. 23 00:02:22,490 --> 00:02:26,570 Right now this is much easier and simpler. 24 00:02:26,570 --> 00:02:32,600 You know, just use an external API, uh, that you can do with replicate. 25 00:02:32,600 --> 00:02:38,870 And here we provide you a link where you can get your free, uh, API token. 26 00:02:39,050 --> 00:02:50,630 So once you have that, you load the replicate module, you, uh, create the, the LM, uh, llama 27 00:02:50,630 --> 00:02:58,880 to lm an instance of that llama to LM, and then you can use it in the same way we have been using the 28 00:02:58,880 --> 00:02:59,660 OpenAI. 29 00:02:59,660 --> 00:03:01,550 What is going to be the difference? 30 00:03:01,550 --> 00:03:06,680 The difference is going to be the quality of the response and the speed of the response. 31 00:03:06,680 --> 00:03:12,560 Llama two is a very good model in in some cases is similar in quality than OpenAI. 32 00:03:12,560 --> 00:03:16,040 In some other cases is a little bit lower in quality. 33 00:03:17,180 --> 00:03:18,740 Second case. 34 00:03:19,730 --> 00:03:20,930 We can use. 35 00:03:20,930 --> 00:03:33,920 Well, here you have a just a quick note saying that you can achieve this same functionality with two 36 00:03:33,920 --> 00:03:36,620 other alternative ways. 37 00:03:36,620 --> 00:03:44,660 First, we can use instead of using this my Elm question, we can use my LM dot predict question or 38 00:03:44,660 --> 00:03:46,340 dot invoke question. 39 00:03:46,340 --> 00:03:50,120 This is going to generate exactly the same output. 40 00:03:50,120 --> 00:03:53,870 These are different ways that long chain provide us. 41 00:03:56,400 --> 00:04:02,970 In order to connect with a chat model a see that we are loading here. 42 00:04:02,970 --> 00:04:10,800 The chat OpenAI module is not the OpenAI module, now is the chat OpenAI module. 43 00:04:10,800 --> 00:04:17,370 And we are also need to load human message and system message. 44 00:04:18,550 --> 00:04:22,240 In this case to create an instance of a. 45 00:04:22,900 --> 00:04:25,600 The OpenAI chat is very easy. 46 00:04:25,600 --> 00:04:30,580 And to create the content, the question is a little bit more elaborate than before. 47 00:04:30,580 --> 00:04:36,460 In this case, we have the option to tell a OpenAI the chat. 48 00:04:37,060 --> 00:04:43,960 We have the option to instruct about the role that he's going to play. 49 00:04:43,960 --> 00:04:51,790 In this case, we are using System message to tell ChatGPT you are a helpful and concise assistant. 50 00:04:52,390 --> 00:04:57,190 And here in Human Messages where we include our question. 51 00:04:57,190 --> 00:05:03,310 So in this case we say name the most popular for us first ladies. 52 00:05:03,520 --> 00:05:07,510 The way to get the response is very similar in this case. 53 00:05:07,510 --> 00:05:11,290 But you see that the response is printed in a different way. 54 00:05:11,290 --> 00:05:13,150 See the content here. 55 00:05:13,150 --> 00:05:13,840 Right. 56 00:05:14,080 --> 00:05:21,640 So if we print the type of the response it's very interesting because now we are getting a class. 57 00:05:21,640 --> 00:05:31,060 And before if you bring the type of the response that an LM is getting is is sending is a string, okay, 58 00:05:31,060 --> 00:05:31,990 this is important. 59 00:05:31,990 --> 00:05:42,730 And you will see then that in this uh, scenario is where we may think about using what we call output 60 00:05:42,730 --> 00:05:43,570 parsers. 61 00:05:43,570 --> 00:05:52,900 We will talk about them later, but in short, output parsers allow us to format the response we get 62 00:05:52,900 --> 00:05:56,320 from the LM model in a different way. 63 00:05:56,320 --> 00:06:02,260 We can say, okay, instead of strings, I want to have the output in a Json dictionary, or I want 64 00:06:02,260 --> 00:06:05,350 to have the output in Python list or whatever. 65 00:06:06,040 --> 00:06:07,150 So. 66 00:06:08,360 --> 00:06:13,910 This is how we connect with a chat model. 67 00:06:15,750 --> 00:06:25,800 And here you have a link to the LMS with native integration with a long chain, and also the chat models 68 00:06:25,800 --> 00:06:28,920 with native integration with long chain.