C3W4_Assignment: Slow model training

[Sebastiano_Rossi]

Link: Coursera | Online Courses & Credentials From Top Educators. Join for Free | Coursera

Q: The training of the model is slow: 30 mins ETA per epoch using T4 runtime. Is this expected or an hint of an error?

[Deepti_Prasad]

Hello @Sebastiano_Rossi

Did you try the fallback runtime?

TF-AT: Using the Fallback Runtime to Avoid Slow Training in Colab

NOTE (02/08/2024): The fallback runtime is now unavailable. You can only access it for a few weeks after Colab updates its runtime. Please create a new topic if you’re having training issues with the labs. Thank you. Google Colab updates its runtime periodically to use newer versions of packages like Tensorflow. This can sometimes affect the use of GPUs for notebooks that require older versions. You will notice that when you reach the training part of the exercise. The lack of GPU will result i…

Let me know if your training is still slow

Regards
DP

[StexX]

Hi DP,

running on GPU is (again?) not possible. Currently, training an epoch takes around 20min. Unfortunately, if I search for “fallback” in the command palette, I do not get any results. Do you have any suggestions on how to run on GPU/train faster?

Thanks and regards,
Stephanie

[Deepti_Prasad]

hi @StexX

there can be other issues too for your training to run slow, kindly use search tool and check if your codes are not incorrect as per previous thread discussions.
i am sharing a post link where you can check if you have any similar error. there are similar threads which can contain other codes to be checked, you can use search tool

C3W4 low structural similarity index

Hello @MGriff Issues with your notebook in creating preprocessing function, where you def augment images(image, label), while you normalize the pixel values, you used true division // but you are suppose to / Next in the same grader cell you need to define function to take label as well using tf.one_hot, also remember instead of using num_classes use the value of num_classes Create a function to generate the saliency map def do_salience(image, model, label, prefix): img_file=image (TH…

the fallback runtime was removed as the assignment are now updated and requires one to run the cells till final last cells which gives your instruction on tf version matching and restarting the kernel.

if doing all this still didn’t resolve your issue, kindly ping me here with the correct tag name as I can miss out on your response as you have responded on an old post.

Regards
DP

[StexX]

Thanks for your help. Turns out that I had to use Colab Pro to run the code on GPU.

[Deepti_Prasad]

Yes but you really don’t require to purchase it, we all have done Colab environment provided by the assignment. Although GPU usage limit for free user has a 24-hour cycle being the only restriction but for an epoch to run to slow there could be other reasons too. Do check the tf version compatibility required for this assignment which is mentioned in the final cell at the end of the assignment.

[javvadi111]

have you completed c3w2 if you completed please reply me i am trying that but it showing me error i have only 1 day left