WEBVTT 1 00:00:00.660 --> 00:00:02.670 One other extremely useful 2 00:00:02.670 --> 00:00:05.310 thing you can do with Ollama, 3 00:00:05.310 --> 00:00:07.500 which you could also do with LM Studio, 4 00:00:07.500 --> 00:00:11.610 as you learned there, is request structured data, 5 00:00:11.610 --> 00:00:16.610 or request the response to be structured in a certain way. 6 00:00:16.830 --> 00:00:19.290 Specifically, you can ask for JSON data 7 00:00:19.290 --> 00:00:22.380 and define a json_schema to which Ollama 8 00:00:22.380 --> 00:00:25.293 and the underlying model then will adhere. 9 00:00:26.220 --> 00:00:27.900 So here, in this example script, 10 00:00:27.900 --> 00:00:30.840 I again, wanna talk to the gemma3 model, 11 00:00:30.840 --> 00:00:33.120 and I have a hard coded prompt here. 12 00:00:33.120 --> 00:00:35.190 And of course you could also build an application 13 00:00:35.190 --> 00:00:37.560 where you are dynamically fetching that prompt. 14 00:00:37.560 --> 00:00:40.623 Here, it's just hard coded for demo purposes. 15 00:00:41.460 --> 00:00:46.050 And I'm asking for a JSON list of three fictional users 16 00:00:46.050 --> 00:00:47.370 where every user should have 17 00:00:47.370 --> 00:00:50.580 a first name, last name, birth date, and country. 18 00:00:50.580 --> 00:00:52.920 And then I again, make it really clear 19 00:00:52.920 --> 00:00:57.510 that I want JSON data, and then I define a json_schema. 20 00:00:57.510 --> 00:01:00.480 And here in Python, that can be done as a dictionary 21 00:01:00.480 --> 00:01:04.380 where I set up the type of the root data object, 22 00:01:04.380 --> 00:01:07.110 so to say, that it should be an array full of items, 23 00:01:07.110 --> 00:01:10.890 and that every item is an object which has four properties, 24 00:01:10.890 --> 00:01:14.010 first name, last name, birth date, country, 25 00:01:14.010 --> 00:01:16.230 where every property is of "type": "string", 26 00:01:16.230 --> 00:01:19.890 so a text value, where I describe every property, 27 00:01:19.890 --> 00:01:22.230 which is important for the model to understand 28 00:01:22.230 --> 00:01:25.083 which kind of value should go into that property, 29 00:01:26.010 --> 00:01:29.670 and that all properties are required and must be set. 30 00:01:29.670 --> 00:01:33.030 Then, for the request that's about to be sent to the model, 31 00:01:33.030 --> 00:01:35.520 I insert the model name, my hard coded prompt, 32 00:01:35.520 --> 00:01:39.330 and this schema under this format parameter. 33 00:01:39.330 --> 00:01:41.580 And that is something you learn in the official docs. 34 00:01:41.580 --> 00:01:43.890 You can set such a format parameter 35 00:01:43.890 --> 00:01:46.350 on the request you're sending to the model 36 00:01:46.350 --> 00:01:51.090 to request data to be returned in a certain JSON schema. 37 00:01:51.090 --> 00:01:52.440 There also is a more detailed 38 00:01:52.440 --> 00:01:55.950 structured outputs example in those docs. 39 00:01:55.950 --> 00:01:59.760 Here, you see something similar as I'm setting it up here. 40 00:01:59.760 --> 00:02:01.260 Again, I don't wanna stream that data, 41 00:02:01.260 --> 00:02:03.123 I want to get it in one block. 42 00:02:04.350 --> 00:02:07.563 And then, I sent my request here and handle the response. 43 00:02:08.580 --> 00:02:12.270 And as a result, when running that script here, 44 00:02:12.270 --> 00:02:14.760 it will send that prompt to the model 45 00:02:14.760 --> 00:02:18.390 that is being executed, or that is being hosted by Ollama. 46 00:02:18.390 --> 00:02:20.310 And indeed then, I get back 47 00:02:20.310 --> 00:02:23.340 data formatted in that format I requested. 48 00:02:23.340 --> 00:02:24.780 And that's of course, super useful, 49 00:02:24.780 --> 00:02:26.460 because here it's just a demo, 50 00:02:26.460 --> 00:02:30.390 but you can use this feature for any use case you have. 51 00:02:30.390 --> 00:02:31.917 Whenever you are building some kind 52 00:02:31.917 --> 00:02:35.040 of automated AI-powered system, 53 00:02:35.040 --> 00:02:37.283 chances are high that you want the AI 54 00:02:37.283 --> 00:02:42.000 to generate data that adheres to a certain structure, 55 00:02:42.000 --> 00:02:44.760 and that's how you can get the model 56 00:02:44.760 --> 00:02:48.090 to return that data in that format you want. 57 00:02:48.090 --> 00:02:51.780 So that's another very useful feature offered by Ollama, 58 00:02:51.780 --> 00:02:54.063 which you can use programmatically.