Valohai

Valohai allows us to adapt the best ML practices and scale

Vorteile:

Valohai is a reliable platform for doing ML that works, it is easy to start using it and it is now a key piece of our data stack. We use it to scale our ML models, track experiment versions, performance and models. I can now feel reassured that I can scale to a growing volume of data and on-board new team members easily. The customer support has been excellent!

Nachteile:

I still find it tricky to combine multiple steps and executions into a pipeline with dependencies. I currently rely on Airflow for doing that and thanks to their API I can still hack it myself. I wish there would be a more developed documentation but you can still rely on a fast customer support if you have any question.

Mandatory ML Ops platform for today's ML teams

Vorteile:

Valohai helps us streamline our Machine Learning model development. All the features are thoroughly tested and they are very fast in releasing new features requested by the customers. The customer support is stellar and they go above and beyond to get you up and running fast.

Nachteile:

None yet, still looking to find the first bug in their platform.

Great Interface for Scalable Machine Learning Jobs

Kommentare: Convenient way to deploy Tensorflow code for shared use between client and internal staff.

Vorteile:

Interface is simple and straightforward; its easy to define jobs using the yaml files and creating jobs is even easier using the we interface. Burning money on AWS has never been so easy :-)

Nachteile:

Sometimes error messages are confusing. Minor interface gaps.

Automated deep learning pipelines

Kommentare: Valohai is one of the most production ready DL versioning tools in the market.

Vorteile:

Valohai makes is super easy to build automated deep learning pipelines and handle automatically the version control of the experiments.

Nachteile:

Standalone app with desktop notifications would be nice.

Perfect for my remote team

Kommentare: This is a great product!
Keep up the great work Valohai Team.

Vorteile:

1) Training sessions are catalogued; allowing easy replication of experimental results and sharing across many team members, 2) AWS resources are managed for you. No extra time is needed to launch VM instance or 3) AWS pricing is shown immediately after executing and logged on a per-experiment basis. This makes it very simple to bill clients for model training. 4) Valohai support team worked aside us from the beginning to ensure we were running properly 5) Live customer support provided in Slack channel

Nachteile:

When you change between commits during job creation it erases all of your previous form entries. This is inconvenient for users who'd like to use the "Copy" feature to run the exact same parameters but on a more recent commit.

A Great Product for Collaboration, Traceability, and Control

Kommentare: The team behind Valohai is excellent! They have helped improve our throughput time, and the tool has helped with traceability for both optimizing our models and rolling-back production models in case of poor results in a deployment.

Vorteile:

Valohai allows us to collaborate with our machine learning code and data sources, as opposed to everything on individual computers with shared copies. In addition, there is a major benefit of traceability by storing each training we run. My favorite thing, though, is the control and monitoring from a web browser, meaning, we can stop/start/re-train executions from a mobile phone, allowing ease of use and flexibility (e.g. I can update a training execution while sitting in a train).

Nachteile:

While I do like that we get plots of our training, such as accuracy or loss, additional plotting functionality could be utilized so we can better decide when to update our hyper-parameters.

awesome to get started, extreme customizability

Kommentare: Deep learning easy training, easy scale up

Vorteile:

No hassle startup, very easy to get help and custom solutions from the staff.

Nachteile:

As a hacker, I would want to get access in the running processes, and files during the jobs run. but allowing that is complicated and has security issues..

Valohai makes it easier to keep track of all your model files and experiments

Vorteile:

I love how Valohai makes it possible to version control models, model configs, and training data. Their scriptable approach provides the flexibility I need to interface with any number of data platforms, so regardless of whether my training data is in AWS, or Hadoop, or anywhere else, Valohai enables me to capture snapshots of training data so I can always know how my models were derived.

Nachteile:

The primary view that users see in Valohai is basically just a table list of ML experiments. The first impression of Valohai is therefore a little underwhelming. If the primary view had a more sophisticated user experience with more color (like analytics of past experiments), then users would be more likely to have a better first impression.