Saving and Loading Models (Coding TensorFlow)

Рет қаралды 169,813

5 жыл бұрын

Training models can take a very long time, and you definitely don’t want to have to retrain everything over a single mishap. Make sure you listen to Magnus as he explains the importance of saving and loading models so you won’t end up losing your valuable training time. Follow along in this episode of Coding TensorFlow with the Colab provided below. And as always, let us know what you think in the comments below!
Colab Save and Restore Models → bit.ly/save-and-restore
For more information on SavedModel, check out this link → bit.ly/2yXmIQS
Want to watch more? → bit.ly/Coding-TensorFlow
Subscribe to the TensorFlow KZfaq channel → bit.ly/TensorFlow1

Пікірлер: 74

@moustafa_shomer 5 жыл бұрын

you really helped me , thank you very much , u saved us time and a lot of work going through all that searching part ,, deeply thankful .

@frankfluit302 2 жыл бұрын

Thank you very very much, really helpful during the final stages of my thesis writing

@jaffarbh 5 жыл бұрын

Great. I would love to see more coverage on TensorFlow Serving next time. Thanks?

@tallwaters9708 5 жыл бұрын

This is great thank you, this is one of the things that always confused me about TF

@JZX619 2 жыл бұрын

Amazing content, needed to learn this

@nikitadabizha4611 5 жыл бұрын

Love this guy

@IvanGoncharovAI 5 жыл бұрын

It's so awesome!

@bryarmustafa2294 4 жыл бұрын

Machine learning blippi! love you man.

@atharvaarya6281 4 жыл бұрын

I love this guy!

@alphacharith 5 жыл бұрын

Thank you Magnus.

@HeavyMetal56339 4 жыл бұрын

Very nice video!

@theRenjie 4 жыл бұрын

I have been wondering how to save the model when training.... it's really annoying when you don't know the Patience to set in the earlystopping.. I learned a lot on checkpoint and model saving skills in this video. Thank you!

@lokeshgehlot3107 2 жыл бұрын

Can u help me to save my model in google collab

@sartaj0 4 жыл бұрын

Thank you sir👍

@suraj-ram7488 5 жыл бұрын

Awesome thanks

@pallavijog912 3 жыл бұрын

Thanks for this video.. I am currently facing problem while storing weights of model under TPU processor of Google Colab. It does not accept local file storage and if specified Google Cloud address, the function does not understand this path starting with gs://. Please help.

@felipeguimaraes7565 5 жыл бұрын

I was hoping that we would learn to save the models directly using Tensorflow, not Keras.

@AkshayAradhya 5 жыл бұрын

Yup. Was pretty annoyed by this.

@QuintinMassey 4 жыл бұрын

Yes thank you! Every example I see uses the Keras API. I don’t want the complexity to be hidden from me because I want to understand what is going on behind the abstraction that the Keras API adds first and then use it later when I feel confident I understand what is going on.

@-abstrvct-4438 3 жыл бұрын

@@QuintinMassey you always can implement it from scratch if you want to see the bare-bones network

@user-px6vf3rm6i 5 жыл бұрын

How save model MobileNet in .pb or . pbtxt format?

@user-zr1ex8hq6c 5 жыл бұрын

Hi, Trained model loading is so slow. So, Case of One image inference, CPU is faster than CUDA. How can I fast inference with CNN?

@gayrattangriberganov8802 3 жыл бұрын

I mean that we can generate images by loading selected savedModels? Is there any video related to this. If yes, plz share

@ameytilve3927 5 жыл бұрын

Where does model get saved ??on google drive?

@amitnair92 4 жыл бұрын

I get “OSError: Unable to open file (File signature not found)” error when converting darknet weights using saveModels

@CalogeroZarbo 5 жыл бұрын

The two thumb down are PyTorch and Caffe. Great video by the way.

@emanuelebarberis2192 3 жыл бұрын

I'm finding a problem in the accuracy on the same testset between a model and the same model reloaded in a different python kernel. Someone know why?

@ennoza41 3 жыл бұрын

i love this guy

@aminuabdulsalami4325 2 жыл бұрын

The drama is quite interesting and informative

@gayalkuruppu9387 3 жыл бұрын

Link in the description does not work

@patinronin2991 5 жыл бұрын

hi i want to learn how to use a scripts.quantize_graph and tensorflow.python.tools.optimize_for_inference with keras to optimize my image classifier and put it in raspberry Pi, I would be very grateful if you could make a video about this. I was looking for information about this but I have not found much with tensorflow and keras.

@gayrattangriberganov8802 3 жыл бұрын

generating images by loading savedModel or checkpoints of GAN model?

@bhuvneshsaini93 5 жыл бұрын

How to use save weights to train new dataset?

@SantoshGupta-jn1wn 5 жыл бұрын

Is this the same Magnus from Hvass Laboratories ?

@tiagocabral8161 2 жыл бұрын

Ok, but what if I have to load the saved model on another notebook? I've tried this and both loss and accuracy from the trained model were not loaded.

@franciscomassucci 2 жыл бұрын

Great content! I'm struggling to save my custom tf.keras model to a .pb file in order to deploy it with opencv dnn. Is there an 'easy' way to do this?

@bernardoolisan1010 2 жыл бұрын

model.save("path") and then model = load_model("path") -> model.predict(x)

@franciscomassucci 2 жыл бұрын

@@bernardoolisan1010 but that way you're still using tensorflow to predict. I was able to use opencv dnn by converting .h5 file to .onnx and then using the method cv2.dnn.readNetFromONNX() The predictions are quite fast setting cuDNN as preferable backend and target

@mtalhakhalid1679 2 жыл бұрын

Which error chk didnu use DGE mae ?? Optimize your label add weights to it

@medinikb 2 жыл бұрын

SuperBBBBB!!!!

@ademkakhadze5270 4 жыл бұрын

'Sequential' object has no attribute '_in_multi_worker_mode' i get this error when i try to use fit function please help

@example.com. 5 жыл бұрын

how to load SavedModel on Edge TPU

@Acer3859 Жыл бұрын

thank you. I can't access your code for Save and Restore Model.

@rakeshl2578 3 жыл бұрын

Sir how to override the get_config in colab?

@xyxyxy9083 5 жыл бұрын

I like this guy

@magelauditore333 4 жыл бұрын

Are all google gurus so fun. I have seen few tutorials of laurence too. He is also fun

@sachinsm 5 жыл бұрын

I had a model from tf. I trained and saved in one system but when i used the model in another system the outputs are different from the model for identical inputs

@mehmetyucelsartas5447 2 жыл бұрын

Hello how can we use trained model in other system ? Can you help me ?

@camus6525 4 жыл бұрын

Ok, but how can i save and really use a model out of colab/jupyter ? In case, on my PC .

@klrshak776 4 жыл бұрын

use PyCharm

@Relative0 4 жыл бұрын

Love the enthusiasm! Wading through TF is tough, at least I can do it with a smile on my face. I'm a... Happy Camper!

@kothapallivijayakumar7421 4 жыл бұрын

am getting ValueError: You are trying to load a weight file containing 16 layers into a model with 0 layers while using load_model

@neuron8186 3 жыл бұрын

Check you model it may be because you didn't tune the layer for model but loading the pre-recorded weight which was trained with 16 layers

@JoshBugg 3 жыл бұрын

i saved my model but i can't find the file locally

@karlbooklover 2 жыл бұрын

getting this error when calling keras.models.load_model() ModuleNotFoundError: No module named 'keras.engine.input_spec'

@ZOBAER496 7 ай бұрын

Say we are going to save a trained CNN model. Can we save only the trained convolutional base part instead of saving the entire model including the fully connected part? The purpose is mainly to use the convolutional base as pretrained feature extractor and fine tune the fully-connected part. How can we do that?

@duu-dungg-tack9522 5 жыл бұрын

I came to hear this guy

@Manjuju123 5 жыл бұрын

The Human Caffine

@truetech4158 5 жыл бұрын

That could possibly be built into tensorflow without requiring additional procedures that tend to be overlooked. Call me simple, however, a dyslexia option could be selected and even gradients of that, from idiot level upto proficiency to allow people with the challenges of grasping formulas to have a list of potential next choices they may be wanting to make along the route. It seems entirely possible to simplify things in a lot of ways, even moreso if to be most effective from such time saving perspectives.

@abdulqadar9580 9 ай бұрын

Thats Amazing

@wahabkhan6492 Жыл бұрын

I have cjeckpoints i want savedmodel file

@miguelalba2106 4 жыл бұрын

My main problem has been training with tf data api with tf records and then make the inference for GANS

@8cherio 2 жыл бұрын

is anyone else getting an error trying to load it?

@nazrin-7515 2 жыл бұрын

me too but no solution rn

@moahaimen 2 жыл бұрын

didnt find the code

@felixmalmenbeck9768 Жыл бұрын

There's an updated version of the Colab here: colab.research.google.com/github/tensorflow/docs/blob/master/site/en/tutorials/keras/save_and_load.ipynb