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!

[ZakiaYahya]

Hello @GemmaTuron @febielin I've tested it on CLI and COLAB but it's giving me an error, here's the log file eos2v11_ CLI.txt

and for COLAB, the run prediction code cell is keep running for hours and didn't gave me any result neither it gave me an error because it keep running and running although i tried it 2 3 times, it works perfectly at fetch and serve time both in CLI and COLAB but giving an error at prediction time. Thanks

[ZakiaYahya]

Hi @GemmaTuron @febielin I just checked the repo and it seems like you didn't modify the src/service.py file. Kindly change predict api torun in the service.py file and it will works. You have to modify the file, push the changes again and then made PR on it again. If you need any help in this regard, kindly let me know. Thanks.

[simrantan]

Tested successfully on Docker: Log:

Results:

Colab: I had issues with fetching the model in colab as well, took 3 hours to fetch and when trying to predict my browser stopped responding, which is probably an issue on my end- will try again soon to see if it will run

CLI: Everytime I tried to fetch the model on CLI, it caused my terminal to stop responding and shut down - the size might be the issue, will

[GemmaTuron]

@febielin Indeed the API showing from docker is still predict, please modify to run. Thanks!

[febielin]

Hi all,

Thanks for your help! I have submitted the PR this morning.

[GemmaTuron]

@ZakiaYahya and @simrantan Please test the updated code, thanks!

[febielin]

All of my output is coming out null. I believe there is an issue with the code that I need to go back and fix.

[GemmaTuron]

@febielin can you try forcing the install from github (--from_github)? to check that the actual code is working and there might be an issue with DockerHub

[febielin]

Attempted forcing the install from github, and I believe it is an error with bentoML again?

[febielin]

When I force the fetch from github on codespaces, the fetch works and I am no longer getting null outputs!

I do find it strange, however, that on Docker, Codespaces, and CLI, when I do the standard 'ersilia fetch eos2v11,' it fetches the old model. I can tell because it still has predict as part of APIs. And the outputs are all null. Only when I force a fetch through GitHub on codespaces do I get the updated model.

[miquelduranfrigola]

Thank you @febielin , this is very useful.

In codespaces can you maybe try ersilia -v fetch eos2v11 --from_github ? This will download the model from the GitHub repository, i.e. the new version.

[GemmaTuron]

Hi @miquelduranfrigola

From what I understand this is what she did already? I think the problem is only that the model is not on DockerHub in its newest version. @febielin can you please look at the failed Action and try to understand why the model could not upload to DockerHub?

[febielin]

@miquelduranfrigola @GemmaTuron Yes, just to clarify, as Gemma mentioned, I did ersilia -v fetch eos2v11 --from_github on Codespaces and this worked! The latest version of the model was fetched, and it was no longer producing null outputs.

I will continue looking into the code to try to resolve the issue of DockerHub upload.

[GemmaTuron]

@miquelduranfrigola and @febielin

We will need to figure out why the model work on codespace but not locally and why is then failing the upload to docker, probably related. Let's mark this for further discussion!

[miquelduranfrigola]

Hi @GemmaTuron and @febielin,

• Indeed, the model works on Codespaces.
• I also managed to build the image with Docker.
• Just in case, I have uploaded the workflows for the model.

I notice that the failed action took 6h before failing. This is clearly too long: link

• I am re-running the actions again, so we can see what may be happening

[febielin]

Hi @miquelduranfrigola,

The action once again exceeded the 6 hr mark and failed: link. Is that to say that the model isn't working, or it is just exceptionally slow?

[febielin]

Hi @ZakiaYahya,

This is the model that I am working on that fails to upload to docker. Do you have any experience with slow docker builds, as Miquel mentioned? If so, how can I expedite this upload process?

[ZakiaYahya]

Hello @febielin Yes in your case, the rdkit is installing using conda-forge channel which took a lot of time in installation, could you please try installing rdkit specific version using pip or pip-pypi, it's way more faster and stable in installing dependancies as compared to conda-forge. Maybe in your case it resolves the problem. Thanks

[febielin]

@ZakiaYahya Thank you for your advice! I will try this out.

[GemmaTuron]

@ZakiaYahya and @febielin,

What is the status of this model testing? All the actions have passed so it should be ready!

[ZakiaYahya]

Hello @GemmaTuron I've tested model on CLI, COLAB and DockerHub and it is working fine on COLAB and DockerHub but it is continuously giving me "Connection Error" on CLI. I'm trying it again but for now here is the complete error log on CLI eos2v11_CLI_log.txt

Colab: eos2v11_ COLAB.csv

Docker: eos2v11_ DockerHub.txt Thanks.

[febielin]

Hi @GemmaTuron,

I have tested this model again and it is functional on my end. @ZakiaYahya Seeing that you are experiencing a connection error, is it possible that this a network issue?

Documentation:

• CLI: eos2v11_fetch_3.log, eos2v11_output_3.csv
• Colab: eos2v11_output
• Docker: eos2v11_docker.log, eos2v11_docker_output.csv

[GemmaTuron]

Hi @febielin and @ZakiaYahya Yes, the connection error on the CLI seems something puntual, so I think we can close this issue as completed!