New model ready for testing!

[github-actions[bot]]

This model is ready for testing. If you are assigned to this issue, please try it out using the CLI, Google Colab and DockerHub and let us know if it works!

[GemmaTuron]

Hi @emmakodes and @ZakiaYahya

Please test the model, it is not yet on DockerHub but make sure it works on CLI and Colab meanwhile

[ZakiaYahya]

Hello @GemmaTuron and @HellenNamulinda I've tested this model on CLI and Colab, The model is working fine on CLI but it is failing on Colab at fetching time. Here's the output and log file eos9yui_ COLAB.txt eos9yui_CLI.csv

[HellenNamulinda]

@ZakiaYahya From the log file, I see the model is failing with this error ModuleNotFoundError: No module named 'ipykernel' I will trace it and find a way of fixing that.

[GemmaTuron]

The ipykernel is related to running jupyter notebooks, that is why it runs normally on CLI. Im unsure of why it is showing in Colab, does it happen for other models @ZakiaYahya ?

[ZakiaYahya]

Hello @GemmaTuron oh right, No i have tested model eos2r5a after that and it executed successfully.

[emmakodes]

Hello @HellenNamulinda @GemmaTuron

CLI

• Model fetch, serve and makes prediction successfully 👍 Model eos9yui fetched successfully! eos9yuilog.txt eos9yuiOutput.csv

Colab

I get this error on colab ModuleNotFoundError: No module named 'ipykernel' eos9yui_colab_log.txt

Does Ersilia need to plot graphs? Do we have to import seaborn? I think why this is failing might be because we import seaborn which is built on top of matplotlib.pyplot and the code is trying to switch_backend() to ipykernel since we are running on google colab(Jupyter notebook) so as to be able to make graph plots.

[HellenNamulinda]

@emmakodes, you're right. I traced it and realized it was connected to seaborn. And like you know we had been advised not to make changes to the model code, unless it is not working. I realized some of the models not only have code for making predictions, but also functions for training,. This is beacuse all code from the original repositories was included. The training functions have code for plotting the the training loss and accuracy. This is why we're getting this issue. @GemmaTuron, I'm commenting out this code for training, Ersilia doesn't need it. We only run inferences on the trained models.

[GemmaTuron]

Hi @HellenNamulinda,

Yes, that makes sense, we do not need any of the training code which only makes the models heavier and require more packages. Is this updated in the repository?

[HellenNamulinda]

Hello @GemmaTuron, I removed the code for plotting the graphs. I didn't remove all the code for training because there was an intersection with the prediction functions. However, going forward, I will do more of a re-incorporation, so that we only add code for prediction only, and not all the code from the original/authors' repositories.

[GemmaTuron]

So we can close this issue as the model is working? @ZakiaYahya or @emmakodes can you check if the ipykernel issue is solved?

[HellenNamulinda]

@ZakiaYahya and @emmakodes, You can try testing again on Colab, because I made the changes. Also, the model is now available on docker, you can try that too.

[emmakodes]

Okay @HellenNamulinda let me test again

[ZakiaYahya]

Hello @HellenNamulinda It is working on Colab, here's the output file eos9yui_output.csv

[emmakodes]

Yes @HellenNamulinda It's working on Colab now. Here is the output file and log: eos9yui_outputcsv eos9yui_colab_log.txt

[GemmaTuron]

I'll close this issue then!