What's new in dstack: bugfixes, improvements, and roadmap
It's now almost a month since we've open-sourced dstack. Over this month, we've been overwhelmed with feedback and attention. It's always thrilling to know that so many people find what you're doing useful. The feedback greatly helped us understand what to do next. Here in this post, we'd like to share some of the news on the recent improvements and on the long-term roadmap.
Now, if you run a server for the first time or update its version, you'll see the progress of the loading dependencies:
Also, when you start a server, if possible, the server runs
dstack config with the proper username, token, and server URL automatically - so you don't have to do it yourself:
When the server is up, it now displays a few hints on what's next – with the links to the documentation, the Discord community chat, and the GitHub repo. If you haven't yet stared it, but interested about the tool, please go ahead and add a ⭐️. This little favor of yours will be very much appreciated by everyone on our team! 🤗
The user interface of the web application got improvements too. The sidebar got icons, a separator, and a link to the Discord community chat:
The page that shows the list of stacks changed its appearance. Now, instead of heavy stack previews, it shows little thumbnails. You also can now switch between the grid and list views. The new appearance is much faster and more responsive.
The gallery on dstack.ai got a new example – an application that tracks the downloads of Python packages:
A week or so ago, we published a long read on making such an application. If you haven't seen it yet, you're very welcome to do it!
The documentation also got a few little improvements. All the code snippets were updated to use the new code style:
import dstack as ds.
If you use dstack's Python package from Jupyter, you'll notice that the result of
push_frame commands is now an interactive link instead of a text with the URL:
The update also got a number of bugfixes (pushing multiple attachments, Bokeh, etc). If something didn't work the last time you've tried dstack, please check it again. It's very likely that the issue got fixed.
If you still experience any problems or miss any particular feature, it's very important that you let us know about it via the issue tracker.
Last but not least, we're currently working on some very cool new features. These features will make it possible to build complex interactive applications that manipulate data and ML models as easily as it was never before. Please stay tuned, spread the word!
To get all the updates, make sure to run the following code:
pip install dstack --upgrade dstack server update
If you use Anaconda, use this instead of
conda update dstack -c dstackai
The dstack.ai team