WEBVTT 00:00.000 --> 00:02.520 -: In this video, we're gonna cover a concept 00:02.520 --> 00:05.340 inside of LangSmith called LangChain Hub. 00:05.340 --> 00:08.550 LangChain Hub is a prompt resource 00:08.550 --> 00:11.220 which allows you to explore different types of prompts 00:11.220 --> 00:13.590 that other prompt engineers have made. 00:13.590 --> 00:15.240 Let's have a look at some of the different types 00:15.240 --> 00:17.850 of things you can do inside of LangChain Hub. 00:17.850 --> 00:18.900 One thing, for example, 00:18.900 --> 00:22.170 is to browse the hubs by a point of interest. 00:22.170 --> 00:23.550 Another thing that you might decide 00:23.550 --> 00:25.680 to do is try out specific prompts 00:25.680 --> 00:27.720 in the playground inside of LangSmith. 00:27.720 --> 00:29.070 Let's have a look in LangSmith 00:29.070 --> 00:30.810 at all these types of use cases. 00:30.810 --> 00:33.780 If you go into LangSmith, and then as well as that, 00:33.780 --> 00:36.390 what we're gonna do is click on the world symbol, 00:36.390 --> 00:38.730 which then takes us into LangChain Hub. 00:38.730 --> 00:40.530 Now inside LangChain Hub, there's a range 00:40.530 --> 00:42.410 of different types of prompts and use cases. 00:42.410 --> 00:44.550 So if you're interested in, for example, 00:44.550 --> 00:46.440 chatbots and code understanding, 00:46.440 --> 00:48.750 we can see that these are the types of prompts 00:48.750 --> 00:51.630 that have seen a lot of resurgence in those. 00:51.630 --> 00:53.280 Now, I can click on a prompt. 00:53.280 --> 00:55.740 Now, I can see this is the Readme of the prompt, 00:55.740 --> 00:58.380 and like how you would potentially use this. 00:58.380 --> 00:59.910 We can also have a look at 00:59.910 --> 01:02.280 how regularly they commit their prompt. 01:02.280 --> 01:04.890 Things you can do is if you go into the top right, 01:04.890 --> 01:06.540 we can say try it, 01:06.540 --> 01:09.030 and then we can also then do a run 01:09.030 --> 01:11.070 by entering in some variables. 01:11.070 --> 01:13.470 So you can, you know, put in some variables here 01:13.470 --> 01:15.660 and you could run specific types of prompts 01:15.660 --> 01:17.070 by clicking start. 01:17.070 --> 01:20.220 And you'll see it's asking for an OpenAI key. 01:20.220 --> 01:23.310 So obviously then you would need to set your secret key 01:23.310 --> 01:26.430 up in the OpenAI key in the top right. 01:26.430 --> 01:28.410 So this is a way where you can see the template, 01:28.410 --> 01:30.270 you can see what is being sent. 01:30.270 --> 01:32.010 You can also edit the template. 01:32.010 --> 01:33.270 So as we scroll down here, 01:33.270 --> 01:34.860 you can see these are the variables 01:34.860 --> 01:37.350 and we can also edit the template in real-time. 01:37.350 --> 01:38.880 You've got other things like being able 01:38.880 --> 01:41.760 to copy the clipboard, delete messages. 01:41.760 --> 01:43.590 We can also move messages around. 01:43.590 --> 01:45.660 So it's possible for us to really play around 01:45.660 --> 01:48.390 inside this little prompt playground. 01:48.390 --> 01:50.790 And we can also save our own prompts 01:50.790 --> 01:52.740 and give it a repository name. 01:52.740 --> 01:55.320 So you can play a lot around in the prompt playground. 01:55.320 --> 01:56.400 Let's actually switch gears 01:56.400 --> 01:58.440 and have a look inside of a Jupyter Notebook. 01:58.440 --> 02:02.130 So what we're gonna do is install Faiss and LangChain Hub. 02:02.130 --> 02:05.040 And then I'm just gonna go and connect to LangChain Hub. 02:05.040 --> 02:08.250 And I'm gonna pull down several different prompts. 02:08.250 --> 02:10.710 If you're looking specifically for the prompts, 02:10.710 --> 02:11.820 if you go on a prompt 02:11.820 --> 02:13.890 and you scroll right down to the bottom, 02:13.890 --> 02:15.810 you'll see this is how you import it. 02:15.810 --> 02:20.430 So it shows you exactly how to import that specific prompt. 02:20.430 --> 02:23.610 Now, we can have a look at some of these prompts. 02:23.610 --> 02:24.780 The one that I'm specifically 02:24.780 --> 02:26.760 interested in is this RAG prompt, 02:26.760 --> 02:30.210 which RAG stands for Retrieval Augmented Generation. 02:30.210 --> 02:31.980 So what we're gonna do is we're gonna follow along 02:31.980 --> 02:33.990 with the example that it's given us. 02:33.990 --> 02:35.550 And just to highlight this, 02:35.550 --> 02:37.650 depending upon the different types of prompts you get. 02:37.650 --> 02:42.120 So I'm gonna go into RAG prompt LangChain Hub. 02:42.120 --> 02:44.400 And then what I'm gonna do is look at this prompt. 02:44.400 --> 02:46.860 And this person's given a common usage 02:46.860 --> 02:48.450 for this specific type of prompt, 02:48.450 --> 02:49.410 which is really, really nice. 02:49.410 --> 02:51.330 So just straight from their repo, 02:51.330 --> 02:54.270 we can already start to get a lot of information. 02:54.270 --> 02:57.810 So we're gonna use a web-based loader, get that webpage, 02:57.810 --> 03:01.770 split the document into individual small documents, 03:01.770 --> 03:03.360 make embeddings from those, 03:03.360 --> 03:04.830 and store those in a vector store. 03:04.830 --> 03:07.050 Then we're gonna pull the RAG prompt. 03:07.050 --> 03:10.440 And then we're gonna set up a retrieval Q&A chain 03:10.440 --> 03:12.330 with a chat OpenAI. 03:12.330 --> 03:14.940 And then we're gonna pass at the LLM, the retriever, 03:14.940 --> 03:18.210 and the chain type args passing in the prompt. 03:18.210 --> 03:19.043 And then we're gonna say, 03:19.043 --> 03:21.420 what are the approaches to task decomposition? 03:21.420 --> 03:23.550 And then we're gonna pass that chain in with the query 03:23.550 --> 03:25.590 and the question and then return the result. 03:25.590 --> 03:29.160 So I think the important thing is LangChain Hub gives you 03:29.160 --> 03:30.540 a lot of flexibility in seeing 03:30.540 --> 03:32.790 what other prompt engineers have created 03:32.790 --> 03:34.620 to solve similar types of problems. 03:34.620 --> 03:36.450 You can easily pull down their prompts using 03:36.450 --> 03:38.010 the LangChain Hub package. 03:38.010 --> 03:39.990 Hopefully, this gives you a good insight 03:39.990 --> 03:42.090 and introduction into using LangChain Hub.