I am developing a ML android app, i know i can use Keras and then turn it to a custom tensorflow model, but i don’t want to depend on an API, so what will i miss if i just train my model with Python and then use the weights i get in my application?
But with the weights, how are you going to run this model on your app? You’ll need more then the weights but also the operations your model needs.
Do you have any specific reason for not wanting to use the dependency?
TFLite takes care of everything regarding loading the model and executing inference using GPU or other accelerators when available.
1 Like
No, the dependency is great, i always love to depend on my self, so when i have to use a dependency i know what i am doing, and i am already using tensorflow in the python script so i do all the operations in a juptyer notebook , get the weights and put them in a formula in android studio
I guess I understood.
For small models that might work but when things gets a little more complex, than the price (in terms of time) might be too high.
If you’re using TF, going for a converted model and using with TFLite is the easiest path and of course, you get the GPU usage for free.
What kind of problems are you solving?
I actually started training because i intend to develop an android application to work like a cooking simulator so i may use variant of models, and as you mentioned “GPU” i have another question if you let me ask, is there a way to get tensorflow to work with openCL as i have amd gpu 
Yes, check:
1 Like
Thanks alot i’ll check it
I checked it and it’s only for android, but i train the model on my lap
On what GPU type? AMD?
No, unfortunately i can’t get tensorflow work on my GPU, it doesn’t detect it, so the operations are done on CPU
You need to use Tensorflow ROCM on your laptop. See:
2 Likes
yes, but i have been struggling since yesterday to install it, i can’t get it work
Hey after nearly 8 hours of struggling i have finally found out that my kernel isn’t supported -_-, i have kernel 5.11.0-34-generic, while ROCm supports only 5.8, do you know a way that i can downgrade my kernel?
What distro do you have?
it’s Ubuntu 20.04.3 LTS
It is supported but you don’t have the default kernel I think.
You can try if it works with your current kernel:
https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html#supported-operating-systems
Or you can ask to the Ubuntu community forum how to reinstall the standard kernel:
No it doesn’t work with my current kernel, they say that they support Ubuntu 20.4 which is mine, but support kernel 5.0-48 which isn’t mine
I think you need to ask to the Ubuntu forum how to revert It to the the official one
Thanks alot for your support, i appreciate it
i used ```
sudo apt install linux-generic
but it installed 5.4.0-48 while ROCm needs 5.8.0-48!! do they challenge me?