7
8 Comments

Poll: What's your biggest challenge in getting ML in production?

What do you struggle with the most getting ML in production?
  1. Reproducing training results
  2. Keeping ML models running in production
  3. Tracking of parameters
  4. Scaling ML training/inference to large datasets
  5. Retaining code quality alongside development velocity (tech debt)
  6. Reusing code + data (reducing waste)
  7. Maintaining comparability between ML models
Vote
  1. 2

    The company I work for is almost a century old so it was originally not a software company. Within our software department there was no A/B testing until recently. My team builds and deploys ML models to serve to various different front facing products. I would say our biggest difficulty has been getting front end teams to work with us so we can A/B test our models.

    1. 1

      That's very interesting. I've just launched my site and I'm looking for people to talk about this exact topic - https://embracing.tech/

      If you're up for it, we can connect on Linkedin?

      1. 1

        Just sent you a connection request. Unfortunately I'm pretty time constrained at the moment but if you give me some details I might be able to help in some way or in the future.

  2. 2

    Server costs, we had to close our startup with @hmert squarefalcon.ml due to high gpu usage costs on servers.
    We were detect and count how many customers in shops in real time but detecting and tracking real time consume so much GPU :(

    1. 1

      @fakturk, @hmert, yeah, totally see that! That should be covered in "scaling ML"! ZenML is enabling preemptible instances for that. You can save quite a lot of money -> check it out: https://docs.zenml.io/tutorials/running-a-pipeline-on-a-google-cloud-vm.html (We're currently working on a blog post on that) Looking forward to your feedback, Adam

  3. 2

    Let's open another group and discuss it there extensively -> https://www.indiehackers.com/group/mlops

    1. 1

      Joined the group, hope you reach the member level for the launch. 😀

      I've just launched a site and my primary interest is AI/ML - https://embracing.tech/

      So if a quick chat is an option for you, we can connect on Linkedin?

  4. 1

    Disclaimer! We just launched ZenML, an extensible and open-source MLOps framework, as we struggled with ALL of the above points. Go check it out and let us know what you think. Thanks, Adam

    https://github.com/maiot-io/zenml

Trending on Indie Hackers
How I grew a side project to 100k Unique Visitors in 7 days with 0 audience 49 comments Competing with Product Hunt: a month later 33 comments Why do you hate marketing? 28 comments My Top 20 Free Tools That I Use Everyday as an Indie Hacker 15 comments $15k revenues in <4 months as a solopreneur 14 comments Use Your Product 13 comments