WEBVTT 1 00:00:00.000 --> 00:00:01.590 [Maximilian Schwarzmuller] So as you learned, 2 00:00:01.590 --> 00:00:03.870 you can save and load sessions. 3 00:00:03.870 --> 00:00:06.390 And what happens technically is that a copy 4 00:00:06.390 --> 00:00:07.920 of the model gets created 5 00:00:07.920 --> 00:00:11.460 and for that copied model, all of your settings, 6 00:00:11.460 --> 00:00:14.490 your system message, your parameter settings, 7 00:00:14.490 --> 00:00:17.970 and your existing chat history is saved 8 00:00:17.970 --> 00:00:21.570 and baked into that model, so to say. 9 00:00:21.570 --> 00:00:25.980 And you can see that model if you run Ollama list 10 00:00:25.980 --> 00:00:28.110 because that lists all the models 11 00:00:28.110 --> 00:00:30.659 that have been downloaded onto your system. 12 00:00:30.659 --> 00:00:34.520 You also have Ollama ps, which you can run 13 00:00:34.520 --> 00:00:38.970 to list all currently running models. 14 00:00:38.970 --> 00:00:41.430 So here, Ollama ps does indeed show me 15 00:00:41.430 --> 00:00:44.490 the Gemma 3 model I started because by default, 16 00:00:44.490 --> 00:00:47.370 Ollama keeps them running for five minutes 17 00:00:47.370 --> 00:00:49.890 and after five minutes without action, 18 00:00:49.890 --> 00:00:52.260 it will remove them from memory so 19 00:00:52.260 --> 00:00:55.380 that they no longer take up memory space. 20 00:00:55.380 --> 00:00:58.310 But this one, since I just used it is still running. 21 00:00:58.310 --> 00:01:02.580 And here, you can also see the size it's occupying in memory 22 00:01:02.580 --> 00:01:06.553 and where it's running, if it's on the GPU or a CPU. 23 00:01:06.553 --> 00:01:09.481 GPU, of course, is better as you learned. 24 00:01:09.481 --> 00:01:13.410 But Ollama list shows you all the models you downloaded 25 00:01:13.410 --> 00:01:16.921 onto your system or all the models that are available here. 26 00:01:16.921 --> 00:01:18.690 And that does include 27 00:01:18.690 --> 00:01:23.280 that Gemma 3 quantized 12 billion parameters model 28 00:01:23.280 --> 00:01:24.800 I downloaded earlier, 29 00:01:24.800 --> 00:01:29.800 but it also includes this s1 model I created, this 30 00:01:30.150 --> 00:01:33.750 copy I created, which is that Gemma 3 model 31 00:01:33.750 --> 00:01:35.870 because that is the model I saved, 32 00:01:35.870 --> 00:01:39.300 but it's that model with my system message, 33 00:01:39.300 --> 00:01:41.160 with my parameter settings, 34 00:01:41.160 --> 00:01:44.510 and with my chat history baked-in. 35 00:01:44.510 --> 00:01:48.270 And therefore, if I knew that I want to use this model 36 00:01:48.270 --> 00:01:50.670 that I want to continue with that chat history 37 00:01:50.670 --> 00:01:51.531 and use all the settings 38 00:01:51.531 --> 00:01:55.440 as I saved them when using Ollama Run, instead 39 00:01:55.440 --> 00:01:57.921 of using ollama run gemma3 40 00:01:57.921 --> 00:02:00.720 and so on, I could ollama run s1 41 00:02:00.720 --> 00:02:02.820 since that's the name I chose for saving it. 42 00:02:02.820 --> 00:02:05.790 And that will now restore that session 43 00:02:05.790 --> 00:02:10.500 and run that model with all the settings I did save. 44 00:02:10.500 --> 00:02:13.268 Now, of course, over time, as you work with Ollama, 45 00:02:13.268 --> 00:02:16.830 as you experiment with it and as you add more 46 00:02:16.830 --> 00:02:18.780 and more models, you might end up 47 00:02:18.780 --> 00:02:20.640 with a long list of models. 48 00:02:20.640 --> 00:02:22.470 And, of course, these models 49 00:02:22.470 --> 00:02:24.673 are taking up space on your system. 50 00:02:24.673 --> 00:02:29.040 Now, technically, despite it saying here 51 00:02:29.040 --> 00:02:32.203 that each model has a size of roughly nine gigabytes, 52 00:02:32.203 --> 00:02:37.203 Ollama is smart and it won't just copy models 53 00:02:37.620 --> 00:02:41.040 if you save the derivatives that are based on that model 54 00:02:41.040 --> 00:02:42.450 like that s1 model. 55 00:02:42.450 --> 00:02:45.285 So it will actually not take up 18 gigabytes 56 00:02:45.285 --> 00:02:47.344 of disc space here. 57 00:02:47.344 --> 00:02:51.450 Instead, it will just save the differences, 58 00:02:51.450 --> 00:02:54.706 the additional configuration you applied to the save model 59 00:02:54.706 --> 00:02:58.306 and then use that saved metadata in conjunction 60 00:02:58.306 --> 00:03:02.925 with the original model files when you run the s1 model. 61 00:03:02.925 --> 00:03:05.616 So you can easily create derivatives 62 00:03:05.616 --> 00:03:07.807 and customized versions of models 63 00:03:07.807 --> 00:03:11.376 without quickly cluttering up your disc. 64 00:03:11.376 --> 00:03:13.110 That's the good news. 65 00:03:13.110 --> 00:03:16.260 But still, of course, over time, you might end up 66 00:03:16.260 --> 00:03:17.776 with a long list of models 67 00:03:17.776 --> 00:03:21.990 where you might simply not need all the models anymore. 68 00:03:21.990 --> 00:03:25.560 That's why Ollama also gives you the rm command, 69 00:03:25.560 --> 00:03:27.510 which you can use to remove a model. 70 00:03:27.510 --> 00:03:31.848 And remove means remove it, also removes it from the disc. 71 00:03:31.848 --> 00:03:36.240 So therefore here, I can run ollama rm s1 72 00:03:36.240 --> 00:03:39.133 and that will remove that s1 model. 73 00:03:39.133 --> 00:03:42.570 Again, in case of this model, since it was based on 74 00:03:42.570 --> 00:03:43.800 that Gemma free model, 75 00:03:43.800 --> 00:03:46.373 it didn't actually take up any significant amount 76 00:03:46.373 --> 00:03:47.565 of disc space. 77 00:03:47.565 --> 00:03:50.954 But still, of course, there is no reason to keep old models 78 00:03:50.954 --> 00:03:54.450 around if you're just not using them anymore. 79 00:03:54.450 --> 00:03:57.432 What you can also do with the Ollama command 80 00:03:57.432 --> 00:04:00.771 when managing your locally downloaded models 81 00:04:00.771 --> 00:04:05.074 is use the show command to learn more about a model. 82 00:04:05.074 --> 00:04:09.125 For example, here, I can use Ollama show gemma3, 83 00:04:09.125 --> 00:04:13.304 and then I'll just repeat that, identify here 84 00:04:13.304 --> 00:04:15.570 to learn more about this model. 85 00:04:15.570 --> 00:04:18.450 And essentially, that's the same kind of information we got 86 00:04:18.450 --> 00:04:22.500 with the slash show command when running that model. 87 00:04:22.500 --> 00:04:24.780 Here, you can get it without starting that model 88 00:04:24.780 --> 00:04:27.000 just to learn more about it. 89 00:04:27.000 --> 00:04:29.713 And that can, of course, be particularly helpful 90 00:04:29.713 --> 00:04:34.713 if you don't know what exactly a model was about anymore. 91 00:04:34.856 --> 00:04:38.070 Because, of course, here with the gemma3 model, 92 00:04:38.070 --> 00:04:41.130 it's pretty clear that it's, well, the gemma3 model, 93 00:04:41.130 --> 00:04:43.465 the 12 billion parameter version of it. 94 00:04:43.465 --> 00:04:47.766 But what if you did run that model 95 00:04:47.766 --> 00:04:49.832 and you then set a system message 96 00:04:49.832 --> 00:04:54.832 like You are a friendly assistant and you then saved this, 97 00:04:56.283 --> 00:05:00.723 custom-m1, whatever, you can choose any identifier you want, 98 00:05:00.723 --> 00:05:04.980 and then you quit and you went on vacation, and 99 00:05:04.980 --> 00:05:06.210 after a time, you came back 100 00:05:06.210 --> 00:05:09.024 and you're wondering, what's this model about? 101 00:05:09.024 --> 00:05:13.672 Well, for situations like this, you can use Ollama show 102 00:05:13.672 --> 00:05:17.220 and you will learn that this was indeed based 103 00:05:17.220 --> 00:05:19.410 on the gemma3 model family, 104 00:05:19.410 --> 00:05:23.400 the 12-billion parameter model flavor specifically. 105 00:05:23.400 --> 00:05:26.130 And then you might remember that this was indeed 106 00:05:26.130 --> 00:05:29.431 that custom version of that gemma3 model you saved 107 00:05:29.431 --> 00:05:32.201 that had that system message baked-in. 108 00:05:32.201 --> 00:05:35.670 So that's why the Ollama show command can be useful 109 00:05:35.670 --> 00:05:38.490 for inspecting and understanding your models, 110 00:05:38.490 --> 00:05:40.473 especially your custom models.