WandB is absolute heaven.
With next to no modification of my code, I'm able to visualize metrics in real time when training my models on multiple systems and can look at their performance relative to each other, and previous runs, anytime on any device. Sweeps also allows me to train my models with different parameter settings on multiple systems in unison via a Bayesian approach to optimally test a range of input arguments and find the best solution.
I can't imagine going back to how I did things before.
Wandb has become the critical tool for us in bringing together the work of remote researchers across several continents. On a recent paper, we started using a wandb report as the central hub for collaboration, where everyone could see the most recent experimental results and essentially the latest draft of the paper. The flexibility and clarity of wand reports have enabled us to collaborate in ways that used to only be possible in person.
Since machine learning is a very experimental process, meaning you try something, see if it works and if not you try something else. I plan on building a number of different models to see which one works best. To track the results of each different model, I set up Weights & Biases, a tool for tracking deep learning experiments.
My team enjoys using this helpful tool. ML is always experimental in nature, in industry and in research and wandb is like a diary to record the whole ML development journey from the baseline to the SOTA. And the best thing is that you can share the journey with others.
Between architecture, hyper parameters and general problem approach, I always found myself going in circles trying to keep track of results, configs and the code version I was using in an experiment. After looking around for a tool/dashboard to help with experiment management I landed on WandB. With one init line I could track many of the metrics I was interested in. Wandb one of the few tools that makes it so easy that it’s a no brainer to try for yourself.
I am impressed daily by wandb and the amazing visualization tools. Now I present all my works and experiments through wandb— no more slides. I also, send progress report to my supervisor using wandb. The synchronization with TensorBoard helps me to further use embedding projector. So far, I have explored all the wandb tools, but I can't wait to explore more.
TensorBoard can be a nightmare when training on multiple machines. I have to run TensorBoard locally on a master machine and sync logs between computers to visualize real-time results. With wandb, this is so easy. Wandb is going to change everything for me.
Great interface. Does what I want it to do without much effort.
10/10. It's really great software. Helps me a ton with my work. Also, sweeps handle a lot of my work.
Effortless python usage, and excellent visualization of my experiments. Minimal hassle, maximum benefit.
Your tool is very easy to set up (well documented!) and works great.
Awesome UI for logging. Cross-platform(library) support. Excellent customer service. Free for an academic use. Easy for beginners etc.
10/10. It is purely awesome!
WANDB has made experimentation just so much more easy! I cannot express how much I love WANDB and how dependent our team has become on this wonderful tool. Love from India!
I really like the visualizations and it's very easy to integrate with the pipeline.
Great tool! Still a lot to improve but looks like you have an entire team fully working on that!
It's just amazing, you guys are doing an amazing job over there! I'm telling about it to all my friends, everyone should use wandb!