WEBVTT 00:00.160 --> 00:02.080 Congratulations, guys. 00:02.080 --> 00:07.080 We succeeded in plotting the results and evaluating the model. 00:07.080 --> 00:11.960 So our model is well trained and ready to export. 00:11.960 --> 00:17.320 So here we're going to save and export our model. 00:17.360 --> 00:25.680 We can now export the trained model to a TensorFlow Savedmodel format which is directory with model 00:25.680 --> 00:26.280 files. 00:26.560 --> 00:30.080 So the model will be saved in a directory. 00:30.120 --> 00:33.240 Linear regression export. 00:33.240 --> 00:37.160 And the name of the model is Savedmodel dot PB. 00:37.360 --> 00:42.760 To export the model we use model dot export function. 00:42.760 --> 00:46.280 And what is the name of our model? 00:46.280 --> 00:56.280 I'll name it as linear regression export and here print model exported for serving linear regression 00:56.320 --> 00:57.800 export okay. 00:58.160 --> 01:00.280 So this is very simple. 01:00.400 --> 01:06.160 We use the export function in order to export our model. 01:06.160 --> 01:08.040 Let's run and here we go. 01:08.280 --> 01:12.200 Save a artifact at linear regression export. 01:12.380 --> 01:15.220 Go to this folder and here we go. 01:15.420 --> 01:18.780 We have the newly created folder. 01:18.780 --> 01:24.180 We have assets, variables, fingerprint and savedmodel. 01:25.140 --> 01:27.500 I'll write those notes here. 01:27.740 --> 01:37.020 The assets folder stores external files needed by the model vocabulary files for NLP models, lookup 01:37.060 --> 01:40.620 tables, custom operators, any external resources. 01:40.620 --> 01:42.460 The model depends on. 01:42.580 --> 01:45.460 If your model uses text processing. 01:45.580 --> 01:49.700 Preprocessing the vocabulary file would be here. 01:49.900 --> 01:55.460 The variable folder stores the trained model weights. 01:55.660 --> 02:03.820 The the first one binary file with weight values and the variables dot index. 02:03.820 --> 02:07.220 Index file mapping variable names to position. 02:07.220 --> 02:09.060 Why split across files? 02:09.060 --> 02:14.340 Large models may be split across multiple data files. 02:14.580 --> 02:18.140 Enables parallel loading and checkpointing. 02:18.380 --> 02:21.540 These files contain your trained parameters. 02:21.540 --> 02:22.700 Weight metrics. 02:22.700 --> 02:24.060 Bias vectors. 02:24.330 --> 02:30.170 Batch normalization parameters and any other trainable variables. 02:30.290 --> 02:32.650 Savedmodel Bebe. 02:33.010 --> 02:34.730 The main model. 02:34.770 --> 02:40.050 The main model architecture and computation graph contains. 02:40.050 --> 02:43.010 Model structure the layers and connections. 02:43.130 --> 02:45.290 Input output specifications. 02:45.330 --> 02:47.690 Computation operations. 02:47.730 --> 02:49.130 Signature definitions. 02:49.130 --> 02:51.290 How to call the model and others. 02:51.370 --> 02:57.730 Fingerprint is the unique identifier for the Savedmodel. 02:57.730 --> 03:02.930 So as a quick recap, we created this model. 03:02.930 --> 03:09.450 We exported this model by using model dot export and we named the name of the folder Linear Regression 03:09.450 --> 03:10.170 export. 03:10.290 --> 03:18.170 We opened the folder the files directory here and we see a folder named linear regression export. 03:18.170 --> 03:25.770 If we click on it we have assets folder stores external files needed by the model. 03:26.090 --> 03:35.210 The variables folder stores the training model weights that saved model the main model architecture 03:35.210 --> 03:42.270 and Computation graph and the fingerprint file unique identifier for the saved model. 03:42.310 --> 03:52.190 As a quick recap model Savedmodel dot PB model architecture variables, folder model parameters, assets, 03:52.470 --> 03:59.190 external resources and model identity is the fingerprint of PB file. 03:59.230 --> 04:05.390 For our example, the savedmodel dot pd file. 04:05.590 --> 04:12.110 Here we have the two that the the relationship the regression. 04:12.150 --> 04:18.470 The variables folder contains the two parameters weight and bias. 04:18.670 --> 04:21.190 The assets folder empty. 04:21.230 --> 04:25.310 We don't need any external resources. 04:25.710 --> 04:30.870 Fingerprint dot PB unique hash for your specific model. 04:30.910 --> 04:34.430 Okay so this is how we export our model. 04:34.430 --> 04:44.350 In the next videos we're going to export the TF Lite which is TensorFlow Lite that is used for Android 04:44.350 --> 04:44.950 Studio.