by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    baaastijn
    @baaastijn

    Hello there !

    We are glad to announce the release of OVHcloud ML Serving Engine !
    Since yesterday, you can find this new service inside your Public Cloud control panel and use it for free during the early access phase.

    What’s this, what do we solve ?

    Well, when we talk about machine learning adoption in companies, most of them face the same challenge : perform a smooth transition from prototype to industrialization.
    Usually, a datascientist build a model then developers and data engineers push it in production, allowing the company to use it on a daily basis. This process is quite often manual and time consuming.

    OVHcloud ML Serving Engine solves this challenge and helps you to deploy models in production. Built by our data team, you can now deploy models with a few clicks.
    Models can be yours, since we are compliant with standards formats such as ONNX, PMML, Tensorflow HDF5/SaveModel, but you can also deploy pre-trained models from our catalog.

    Once your model is deployed, you will be provided an API endpoint, but also nice features like autoscaling, versioning and observability.

    Want to give a try for free ? Connect to OVHcloud control panel, select Serving Engine in the left menu : https://www.ovh.com/manager/public-cloud/#/

    For more information, you can find our documentation here : https://docs.ovh.com/gb/en/serving-engine/

    And of course don’t hesitate to share your feedbacks and questions here !

    Have a good day :)

    baaastijn
    @baaastijn
    Cc @/all :)
    XavBart
    @xavbart
    Hi there. I did upload a PMML model (generated at BigML) and indeed it seems to have correctly been set up (it does describe correctly parameters and all in the expected input / output on the describe page and on the direct URL) and it gets a 500 error. https://cl.ly/8386b2400022 (I thought I was making mistake in my json doing my own request, so I tried OVH-provided test on the URL page, same result) Any clue as to where I should be looking ?
    XavBart
    @xavbart
    @xavbart (and also, let me know if this is not exactly the right area to discuss things that might be caused by my own stupidity ;-) )
    baaastijn
    @baaastijn
    hello, this is the right place :)
    checking with the team
    Adrien Carreira
    @XciD
    Hello @xavbart checking
    could you provide us your namespace id ? the random char in the begining of your url
    XavBart
    @xavbart
    @XciD do you mean the namespace I decided ? As it is not random ;-) "thinking-elements-test1"
    Adrien Carreira
    @XciD
    We found the issue
    you have an output column that have the same name as an input column. we are fixing it
    XavBart
    @xavbart
    @XciD Oh ok ! Indeed I know which one, it was a shortcoming from the pmml-generating tool and my colleague who set it up (but that was in the training phase. In the evaluate part, it doesn't ask for the same column - which was the objective and the given result -). But I can't see a way to correct, because the process is set. So when you're saying you fixed it, it means it wouldn't be a problem once we try again with same PMM ?
    Adrien Carreira
    @XciD
    Yes, we are working to fix it in our code without any change on your side.
    XavBart
    @xavbart
    @XciD thanks guys
    Christophe Rannou
    @ChrisRannou_twitter
    @xavbart Hey, we fixed the problem, you can just upgrade the model and once the upgrade is done you should be good to go
    XavBart
    @xavbart
    @ChrisRannou_twitter Yes, indeed. Seems the black magic works, well done!
    XavBart
    @xavbart
    On a different topic, I can't seem to use my OVH ID for OVH AI Mktplace, which might be fro a reason, BUT I cannot create an account either (login gets registered, but chosen password is not accepted and there IS NO RESET PASSWORD option on login. Tried twice. What gives?) And I cannot seem to be able to leverage the models offered there (but for sentiment analysis) on prescience, so I wanted to try and use them directly through the exposed API, but no obvious way to do so without a key, which I guess might be linked to having an account ? Sorry if this is not the correct venue for those complaints
    baaastijn
    @baaastijn
    hello, the marketplace is fully autonomous and not linked to OVH credentials
    you need to create a free account (jsut an email required) in the marketplace itself :https://market-place.ai.ovh.net/#!/registration
    Once done, you will be able to create token (what you call a « key ») directly in the marketplace
    So far, we added Sentiment Analysis in OVHcloud serving
    if you have requests, don’t hesitate to explain your use cases here :)
    XavBart
    @xavbart
    Yeah all those points were indeed what I surmised (different credentials, getting a key from marketplace, and using those models this way). So am stuck at the registration which fails to accept my chosen password (and no reset option). Will try different options, thanks for taking the time to validate my questions. Re: use cases will do once i can see compared to what our exploratory process produces.
    XavBart
    @xavbart
    (and today password seems to be accepted, maybe a refresh backend issue)
    baaastijn
    @baaastijn
    ah ok, sorry for that. I misread your request
    we put some examples here if you want : https://github.com/baaastijn/ovh-ai-api-examples/
    ChrisHaus
    @ChrisHaus
    Hello guys! I wanted to use the serving engine with a tensorflow saved model (.pb) but it can't even deploy it. In the documentation it says it accepts "Tensorflow serialized through SavedModel or HDF5" but in the console it's written "Select the pathway for your model’s file in Object Storage. Your file must have the extension: h5 / pmml or onnx". Does it means that I have to convert my model to a hdf5, onnx or pmml format? Thanks in advance.
    Adrien Carreira
    @XciD
    Hello, to use .pb, you need to choose the directory option and not the single file option. In fact, you should have some other file among your .pb (the weights), so the single file won't work.
    Adrien Carreira
    @XciD
    Capture d’écran 2020-04-06 à 09.49.12.png
    ChrisHaus
    @ChrisHaus
    Ok thanks
    Adrien Carreira
    @XciD
    Don't hesitate if you need any help
    Nicocg70
    @NicoCG70_twitter
    poke @XciD ?
    Nicocg70
    @NicoCG70_twitter
    Hi. With different trys, the DeOldify AI seems to not respond. Anyone to confirm ?
    I use the very basic exemple curl -X POST "https://api-market-place.ai.ovh.net/image-colorization/process" -H "accept: image/jpg" -H "X-OVH-Api-Key: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX" -H "Content-Type: application/json" -d '{"source_url":"https://i.ibb.co/0Ghxy72/input.jpg",
    "render_factor": 35}' --output my_colorized_image.png with my API-Key
    The return is a corrupted file of 26 Ko.
    Nicocg70
    @NicoCG70_twitter
    that says "input error" (as txt)
    Nicocg70
    @NicoCG70_twitter
    The same problem even in the swagger page...
    baaastijn
    @baaastijn
    hello @NicoCG70_twitter you’re right, we check thart tomorrow. same issue here.
    Nicocg70
    @NicoCG70_twitter
    @baaastijn ok j'attends votre retour qd c'est bon. Thx
    baaastijn
    @baaastijn
    @NicoCG70_twitter it’s now working :)
    Nicocg70
    @NicoCG70_twitter
    Thanks... I will try soon. Commercial contact are here too ?
    baaastijn
    @baaastijn
    you mean on the channel ? i don’t think so
    in short, we first developed AI Marketplace with pre-trained models, FREE for all, but some limitations (amount of calls etc)
    since 1 month we released ML Serving in OVH control panel
    allowing you to deploy your own models
    Or OVH pretrained ones.
    So far we only pushed Sentiment Analysis but we are here to discuss if you need something else
    i’m the Product Manager for Data, and i will do the bridge with commercial team if required
    Nicocg70
    @NicoCG70_twitter
    DeOldify works perfectly, many thks @baaastijn
    🎙Jean-Louis Queguiner
    @JiliJeanlouis_twitter
    sorry @NicoCG70_twitter My mistake on the swagger.