WEBVTT 00:00.120 --> 00:00.520 Okay. 00:00.560 --> 00:04.360 So here we are looking at our node creation. 00:04.760 --> 00:06.840 Remember a node is a function. 00:06.840 --> 00:08.440 So we're going to create a function. 00:08.440 --> 00:10.200 It's called our first node. 00:10.600 --> 00:14.440 It takes an old state and it returns a new state. 00:14.760 --> 00:16.640 Its states are immutable as we say. 00:16.680 --> 00:18.720 And so we are not going to do anything. 00:18.720 --> 00:20.520 We're not going to come and mutate old state. 00:20.560 --> 00:23.920 In fact, you may see from the way that it's grayed out here that we don't actually touch Old State, 00:23.920 --> 00:26.560 which is unusual, but just for this example. 00:26.800 --> 00:28.120 So what are we actually going to do? 00:28.160 --> 00:30.160 So we're going to make a string called reply. 00:30.440 --> 00:35.240 And that reply is going to be some random word choice of a noun. 00:35.520 --> 00:38.480 And then the word r and then a random choice of an adjective. 00:38.960 --> 00:45.640 And we're going to create one of these kinds of message structures and OpenAI familiar message structure 00:45.800 --> 00:47.360 and put that into messages. 00:47.920 --> 00:53.640 We're going to say that the new state is R another a new state class. 00:53.640 --> 00:56.800 We're going to instantiate it passing in this messages. 00:57.000 --> 00:59.680 And we are going to return that the new state. 00:59.680 --> 01:02.320 And that is the end of our first node. 01:02.400 --> 01:05.330 And we then call graph dot add node. 01:05.330 --> 01:08.650 This is how we officially add it to the graph that is being built. 01:08.690 --> 01:14.850 We give it a name, we call it first node and we pass in the function the function that represents the 01:14.850 --> 01:17.170 node called our first node. 01:17.610 --> 01:17.930 Okay. 01:17.970 --> 01:19.490 So we're going to run that cell. 01:19.490 --> 01:22.890 And now we're coming on to look at creating our edges. 01:23.250 --> 01:28.930 So you can see here that I now call Graph Builder add edge to add a couple of edges. 01:28.930 --> 01:32.330 And you'll see that these words start and end there. 01:32.410 --> 01:33.250 So what are they. 01:33.290 --> 01:37.330 Well these are things that we've imported at the top in from graph graph. 01:37.370 --> 01:39.490 They are constants start and end. 01:39.490 --> 01:43.050 And of course they signify the beginning of our workflow and the end of it. 01:43.090 --> 01:49.650 And so here what we say is we want an edge to take us from the start to our first node. 01:49.650 --> 01:53.610 And then we want another edge to go from the first node to the end. 01:54.050 --> 01:55.970 Okay, that sounds logical doesn't it? 01:56.010 --> 01:57.210 So we'll run that. 01:57.410 --> 02:02.090 And now with that we get to step five which you'll remember is compiling the graph. 02:02.130 --> 02:03.210 It's saying we're done. 02:03.210 --> 02:04.730 This is our workflow. 02:04.930 --> 02:06.530 And we can now display it. 02:06.570 --> 02:10.500 This is rather nice a quick way to show visually what it is we're talking about. 02:10.500 --> 02:11.900 And hopefully this will be no surprise. 02:11.900 --> 02:16.140 We've got a start going into our first node going into our end. 02:16.540 --> 02:17.660 What could be easier? 02:17.900 --> 02:18.580 That's lovely. 02:18.620 --> 02:25.460 Okay, so we've done the five steps, which is of course the what the first part of running a graph 02:25.460 --> 02:26.060 system. 02:26.060 --> 02:31.300 We've compiled our graph and we've done it by adding nodes and edges. 02:31.300 --> 02:33.420 It's now time to run it. 02:33.580 --> 02:34.140 Okay. 02:34.180 --> 02:36.060 And what we're going to do is to run it. 02:36.100 --> 02:39.540 We're going to create a gradio chat function. 02:39.660 --> 02:40.420 Why not. 02:40.540 --> 02:48.060 Remember gradio chat functions take the user's current input and the history of prior inputs, and it's 02:48.060 --> 02:51.300 meant to respond with the next output. 02:51.300 --> 02:56.260 That's just what you do with a gradio chat function that we're going to pass into gradients chat interface 02:56.260 --> 02:57.020 right there. 02:57.220 --> 02:58.100 So that's what we have to do. 02:58.100 --> 02:58.940 So what do we want to do. 02:58.940 --> 03:06.780 Well, we're going to turn the message into a standard OpenAI format and put that into a message. 03:07.140 --> 03:11.710 We're then going to create a state object with that as the is the message. 03:12.270 --> 03:14.950 We are then going to invoke our graph. 03:15.110 --> 03:17.870 And this is the key long chain word invoke. 03:18.030 --> 03:18.790 You may be familiar. 03:18.830 --> 03:20.430 Sorry I said long chain. 03:20.430 --> 03:22.310 It's the graph word invoke. 03:22.310 --> 03:25.590 But you may be familiar with it because it's the word in long chain as well. 03:26.150 --> 03:32.470 So you invoke a graph in land graph with the state in order to get the result. 03:32.710 --> 03:35.510 And that's what's going to execute our graph. 03:35.510 --> 03:37.950 And what will come out will be the result. 03:37.950 --> 03:41.390 And we will print it and we will also return it. 03:41.630 --> 03:45.430 And that will come out of our chat function. 03:45.590 --> 03:49.310 And so with that let's run this and see what happens. 03:49.870 --> 03:51.150 Well we have a greater UI. 03:51.190 --> 03:51.790 That's good. 03:51.790 --> 03:53.070 Let's say hi there. 03:55.590 --> 03:57.070 Muffins are haunted. 03:58.750 --> 04:00.390 Is that so? 04:02.430 --> 04:04.390 Penguins are sparkly. 04:05.470 --> 04:09.710 You don't say penguins are outrageous. 04:11.110 --> 04:11.750 For real. 04:13.830 --> 04:16.520 Pickles are untrustworthy. 04:17.640 --> 04:18.400 Ha ha ha. 04:18.880 --> 04:19.800 You get the idea. 04:20.160 --> 04:27.400 Uh, so, uh, this is the results of our small language model that is picking a random noun and adjective. 04:27.400 --> 04:28.640 And I show you this. 04:28.760 --> 04:30.680 If you're wondering what on earth are you doing? 04:30.800 --> 04:36.480 I'm doing it to show that the graph setup has nothing to do with Llms necessarily. 04:36.600 --> 04:42.080 The node is just a function, in this case a silly function, but it's a function there and it's taking 04:42.080 --> 04:46.320 in a state and it's returning a state, and it doesn't need to have anything to do with llms. 04:46.560 --> 04:50.200 Now down here, I'm printing the result of this. 04:50.240 --> 04:52.480 And let me just show you that print statement. 04:52.720 --> 04:57.640 Uh, we're printing what's coming back from calling invoke on the graph. 04:57.960 --> 05:02.560 And I want to point out that that it may be something a bit different to what you're expecting because 05:02.560 --> 05:07.320 it's not it's not messages with just a list of, of strings. 05:07.480 --> 05:13.040 It's got a list of these things called human message, which you may know from, from Lang chain work. 05:13.040 --> 05:16.680 It's like a, it's a construct to package things up. 05:16.680 --> 05:20.010 And this is of course the result of the reducer running. 05:20.010 --> 05:24.450 And when I said before that the reducer simply concatenates things into a list, I wasn't telling the 05:24.450 --> 05:29.490 full story because it also does some of this packaging up that comes with Landgraf. 05:29.490 --> 05:34.890 So if it just takes the text that's come back, it knows how to package it into a human message. 05:34.890 --> 05:41.090 In terms of the things that I was saying and an AI message in terms of muffins or haunted and penguins 05:41.090 --> 05:43.370 are sparkly and the like. 05:43.530 --> 05:49.810 So that's this is some of the of the stuff hanging happening behind the scenes as a result of Landgraf, 05:50.210 --> 05:51.730 which doesn't really matter to us. 05:51.770 --> 05:53.290 We're taking advantage of that. 05:53.290 --> 05:57.130 But we just wrote a simple, a super simple state. 05:57.370 --> 06:00.570 Uh, super simple node. 06:00.570 --> 06:02.810 We made a state that contained messages. 06:03.010 --> 06:09.530 We made a node we take in an old state, we return a new state which is using this random sentence, 06:09.730 --> 06:10.850 and it works. 06:10.850 --> 06:17.650 We can invoke our graph and get a response, and we can have a conversation of rather one sided conversation 06:17.650 --> 06:19.810 with a silly language model. 06:20.570 --> 06:22.530 All right, let's do something more sensible.