Valohai Hybrid AWS

What do you like best about the product?

Valohai has a relativel y shallow learning curve which makes getting off the ground easy. From there, implementing our ideas has been straightforward with only minimal help needed. Speaking of help, a member of their staff has been with us every step of the way to help debug, implement new ideas, and communicate updates from their end. We use it exclusively for training models but there are several more features which we have not touched that should help expand even further.

What do you dislike about the product?

Nothing major - it works great for our use cases. There has been a hicup or two along the way, but nothing significant and the support we got from the staff helped immensely.

What problems is the product solving and how is that benefiting you?

Valohai is helping us solve the problem of not enough on-premises compute for model training

What do you like best about the product?

It is very easy to use and has a straightforward UI. Valohai makes building pipelines an easy and enjoyable process. Most importantly, the support from the Valohai team is amazing. They are responsive and friendly.

What do you dislike about the product?

Nothing encountered so far; it is very straightforward to use.

What problems is the product solving and how is that benefiting you?

building pipelines and making inferences according to schedule.

What do you like best about the product?

The Valohai platform truely enables collaboration by ensuring transparency and traceablilty of data and models and by being fully integrated into version control.
- The whole team can access and inspect experiments.
- Changes can be easily implemented and tested.
- Individual executions are highly customizable, allowing efficient and ECONOMICAL resource use.
- Ability to reuse "good" or unchanged steps in a pipeline; saves time!
- Comprehensive documentation making it very easy to get your first implementation up and running.
- Incredible flexibility and outstanding customer support... if I ever had trouble getting something to run, the solution was only a quick chat, personalized video, or one-on-one debug session away.

Valohai is now my daily go-to platform for ML projects.

What do you dislike about the product?

Sometimes debugging a Valohai-specific feature can inflate your git commit history... when hunting down the ever-elusive one character bug. But thanks to that I learned about git squash! So really no problem in the end :D

What problems is the product solving and how is that benefiting you?

Being able to collaborate transparently on an ML project with a team of Data Scientists. Several people have the knowledge to modify and contribute to the "nuts and bolts" of the project and Valohai enables us to trace and carefully inspect how and which changes affect the model outcome. It is a GREAT tool for collaboration!

What do you like best about the product?

- Easy to use, understand and setup either recurring to the UI or the command line tools
- Very good documentation
- Excellent customer support, always eager to improve in the smallest of details
- Flexible and easy to integrate with other solutions such as HF, W&B
- Experiment tracking and reproducibility at its finest

What do you dislike about the product?

- The tags of the experiments in one step are not directly ported to the downstream steps

What problems is the product solving and how is that benefiting you?

Valohai solves: experiment tracking and management, reproducibility of results, collaboration and version control, scaling, automation and pipeline management.

What do you like best about the product?

The platform is very straightforward and easy to use, and the UI is accessible to a wide range of users regardless of their technical expertise. It's easy to get started and learning its intricacies doesn't take long at all. Just write some yaml, store some environment variables, connect to your repo, and go to town on your projects.

In terms of collaboration features, it's not lacking, as a team we can work on shared workspaces meaning all the people involved in the same project can access and work on the same experiments. Due to how it integrates with Git, it also provides version control and traceability. It's incredibly easy to share setups with other team members as anyone can go and review, debug, or replicate previously set up tasks or pipelines. This also enables a collaborative workflow between data scientists and data engineers, where we can contribute to the different stages of the project at the same time, which streamlines the dev process.

It has an efficient hyperparameter tuning setup making it a useful tool for fine-tuning. No matter your flavor of framework, whether you're team PyTorch or team Tensorflow, the support for multiple frameworks ensures you don't have to make significant changes to your tech stack.
When you define the parameters for your tuning run, it immediately gives you a number of how many combinations your parameters result in, which is really handy as it enables users to be conscious about the number of runs and costs associated with them. In the cases where you need to do heavy grid searches, the auto-scaling queue handles all the runs, which is one less thing you have to worry about.

The team behind Valohai is incredibly lovely and the customer support is knowledgeable, friendly, and responsive. I really like that they encourage us to get in touch directly with them whenever we come up with any issues. They're great at troubleshooting the issues we encounter and are quick to offer solutions that work.

What do you dislike about the product?

Not necessarily a dislike, but I'd like to see more documentation or examples of how to run things in a notebook and how to capture the results of notebook runs.

What problems is the product solving and how is that benefiting you?

- Putting ML models into production
- Fine-tuning models/experimenting with different models and params in the same pipeline in a fast and seamless way
- Keeping track of model metrics