WEBVTT 00:00.000 --> 00:01.560 -: Over the next couple of lessons, 00:01.560 --> 00:05.250 we'll be looking at a package in Python called Lang Chain. 00:05.250 --> 00:07.410 Lang Chain is a Python package 00:07.410 --> 00:09.600 and also they have a JavaScript package, 00:09.600 --> 00:12.060 and what it allows you to do is easily interface 00:12.060 --> 00:15.450 with large language models such as open AI's GPT-4. 00:15.450 --> 00:16.860 It has a range of different modules 00:16.860 --> 00:19.170 that we'll be exploring such as Model I/O. 00:19.170 --> 00:22.440 Model I/O in this case means the inputs and outputs 00:22.440 --> 00:23.490 of our model. 00:23.490 --> 00:25.710 For the language model, that means looking at specific 00:25.710 --> 00:29.250 things such as prompt templates, so how can we add templates 00:29.250 --> 00:32.670 to give a language model the examples that we need. 00:32.670 --> 00:35.400 Also, as well as that it looks at how can we use 00:35.400 --> 00:38.070 different types of models, so the ones that we'll be 00:38.070 --> 00:40.530 specifically using are chat models, 00:40.530 --> 00:43.830 and we'll go through a installation on how you can easily 00:43.830 --> 00:44.730 use chat models. 00:44.730 --> 00:47.430 Finally, though, we need to get some structured data out 00:47.430 --> 00:48.990 of our large language models. 00:48.990 --> 00:51.630 It's important that we are able to extract 00:51.630 --> 00:54.510 structured data from large language models 00:54.510 --> 00:56.850 rather than just being able to extract text. 00:56.850 --> 00:59.280 This is where output passes will come in handy. 00:59.280 --> 01:03.060 We'll use these to extract serializable and deserializable 01:03.060 --> 01:07.170 objects from within Python using a package called Pidantic. 01:07.170 --> 01:09.270 You'll see the output passes allow us to go 01:09.270 --> 01:11.610 beyond just simply extracting text. 01:11.610 --> 01:14.940 We'll be able to extract structured data from the LLM, 01:14.940 --> 01:17.490 which you can use in your applications. 01:17.490 --> 01:20.280 Chains are a first class citizen in Lang Chain 01:20.280 --> 01:22.410 and you'll be exploring how to use chains 01:22.410 --> 01:25.170 to create more complicated architectures 01:25.170 --> 01:27.180 when you're using a large language model. 01:27.180 --> 01:28.890 There's lots of different types of chains 01:28.890 --> 01:31.440 and the reason why is that we can string these chains 01:31.440 --> 01:34.050 together to create lots of clever components 01:34.050 --> 01:36.150 when using large language models. 01:36.150 --> 01:39.780 We'll also look at things such as the retriever section 01:39.780 --> 01:43.140 and looking at how we can load documents using text, 01:43.140 --> 01:47.190 for example, so loading CSV files or mark down a PDF 01:47.190 --> 01:50.160 and how we can split these files so that you can easily 01:50.160 --> 01:53.070 interact with them inside of Lang Chain. 01:53.070 --> 01:55.410 This is due to the fact that some of your documents 01:55.410 --> 01:57.360 might be quite long and therefore we might 01:57.360 --> 01:59.310 need to do some sort of document splitting 01:59.310 --> 02:01.230 and document transformation. 02:01.230 --> 02:03.750 We'll also touch on the retriever section, 02:03.750 --> 02:06.600 which is how you can use different types of retrievers 02:06.600 --> 02:11.250 once you have ingested your documents to retrieve and search 02:11.250 --> 02:13.200 for different types of results. 02:13.200 --> 02:15.690 You'll investigate memory, which is a way for you 02:15.690 --> 02:19.890 to store the chat bot history inside of your code. 02:19.890 --> 02:22.080 This will be very useful if you're building chat bot 02:22.080 --> 02:24.600 applications or any application where there is 02:24.600 --> 02:27.090 a required amount of memory that needs to happen 02:27.090 --> 02:28.770 across multiple components. 02:28.770 --> 02:30.510 There's also callbacks as well. 02:30.510 --> 02:32.520 So there's a range of different things that we can use 02:32.520 --> 02:34.710 Lang Chain for, all the way from prompting 02:34.710 --> 02:37.020 and also working with documents that are too large 02:37.020 --> 02:39.120 to fit into a single LLM request. 02:39.120 --> 02:41.790 We'll explore specifically chat models 02:41.790 --> 02:45.060 and work through prompt templates looking at chains, 02:45.060 --> 02:47.373 retrieval methods, and much more.