Aim 2.2.0 featuring Hugging Face integration is out! Thanks to the incredible Aim community for the feedback and support on building democratized open-source AI dev tools.
Thanks to siddk, TommasoBendinelli and Khazhak for their contribution to this release.
Note on the Aim versioning: The previous two release posts: Aim 1.3.5 and Aim 1.3.8 had used the version number of AimUI as those contained only UI changes. From now on we are going to stick to the Aim versions only regardless of the type of changes to avoid any confusion. Check out the Aim CHANGELOG.
We have also added milestones for each version. As well as the Roadmap.
Check out the new features at play.aimstack.io.
Hugging Face integration
Hugging Face integration has been one of the most requested features so far and we are excited to finally ship it. Here is a code snippet of easy it is to integrate Aim and Hugging Face.
from aim.hugging_face import AimCallback
from transformers import Trainer
aim_callback = AimCallback(repo='/log/dir/path', experiment='your_experiment_name')
trainer = Trainer(
model=model,
args=training_args,
callbacks=[aim_callback]
)
Here is how it works: aim_callback
will automatically open an Aim Session
and close it when the trainer is done. When trainer.train()
, trainer.evaluate()
or trainer.predict()
is called, aim_callback
will automatically log the hyper-params and respective metrics.
Find a full example here.
Metric Visibility Control
When dealing with lots of experiments some metrics may need hide (while still being in the Search) to allow better visibility of the overall picture.
A toggle is now available both for each metric/row as well as for global on/off. Check out the demo below:
Column resize
With lots of params tracked, some of them are too wide for table and take over the whole screen real estate. Column resize will allow to fully control data width on the table. Resize is available both on Explore and Dashboard tables.Here is a quick demo:
Hide columns with similar values
More often than not params have lots of repetition. Especially when tracking 100s of them. In that case the explore table becomes super-noisy.
This feature allows showing only the relevant info. This has been certainly the missing piece of the column management that many had requested. Here is a quick demo:
Logscale
One of the must-have features that we hadn’t had a chance to work on. Finally it’s available!!
New methods for aggregation
Now you can select different types of aggregations as well as whether to see the whole area or just the aggregated lines.
Learn More
Aim is on the mission to democratize AI dev tools.
We have been incredibly lucky to get help and contributions from the amazing Aim community. It’s humbling and inspiring.
Try out Aim, join the Aim community, share your feedback.
And don’t forget to leave Aim a star on GitHub for support 🙌.