TL;DR 381 | The Google Developer News Show
0:00 - Intro
0:08 - Signal inputs now in preview for Angular devs! → https://goo.gle/4a7Pdv5
0:39 - Gemma is now available in the KerasNLP collection → https://goo.gle/43gi8uA
1:02 - Croissant, a new metadata format for ML datasets → https://goo.gle/3PgtGbG
1:49 - MediaPipe now supports LLM inference and image generation → https://goo.gle/3v3y331
Here to bring you the latest developer news from across Google is Rody Davis.Tune in every week for a new episode, and let us know what you think of the latest announcements in the comments below.
Follow Google for Developers on Instagram → https://goo.gle/googledevs
Watch more The Developer Show → https://goo.gle/GDevShow Subscribe to Google for Developers → https://goo.gle/developers
#Google #Developers
Speaker: Rody Davis; Products Mentioned: Material Design - General, Angular - General;
0:00 · [MUSIC PLAYING] RODY DAVIS: Hi.
0:03 · I’m Rody for “The Developer Show”.
0:04 · And this is your weekly update on the coolest developer news from Google.
0:07 · Starting with Angular, signal inputs are now in developer preview.
0:11 · Signals provide a powerful reactivity model that enables you to efficiently monitor changes, derive values, while automatically notifying Angular when any specific part of your application needs to be rerendered.
0:23 · Signal inputs can also improve your application by automatically marking OnPush components dirty, being more type safe, and can be derived and used in other signals.
0:32 · This can also be easily monitored when using effects.
0:36 · You can learn more on the Angular blog.
0:39 · Next, let’s head over to Google AI.
0:41 · The Keras team is happy to announce that Gemma is now available in the KerasNLP collection.
0:46 · Thanks to Keras 3, Gemma runs on Jax, PyTorch, and TensorFlow.
0:50 · Keras is also introducing two new features, a new LoRa API, or low-rank adaptation, and a large-scale model parallel training capability.
0:58 · If you want to learn more, check out the Google for Developers blog.
1:02 · If you’re a machine learning practitioner looking to reuse existing data sets to train ML models, you’ll probably spend a lot of time understanding data, making sense of its organization, and figuring out which subsets to use as features.
1:14 · ML data sets cover a broad range of content types, including text and structured data to images and video.
1:21 · There are some general-purpose metadata formats for data sets.
1:24 · But these were mostly designed for discovery.
1:26 · Today, we’re introducing Croissant, which is a new metadata format ready for ML data sets.
1:32 · It was developed collaboratively by a community of industry and academia as part of the ML Commons effort.
1:37 · Starting off, it will support three popular collections of ML data sets— Kaggle, HuggingFace, and OpenML.
1:43 · To learn more about the 1.0 release of Croissant, check out the Google research blog.
1:48 · And finally, MediaPipe now supports two new generative AI tasks, including LLM inference and image generation.
1:54 · The LLM inference API supports Gemma 2B and can run on-device generative AI text-to-text generation.
2:01 · You can learn more about MediaPipe on developers.google.com.
2:05 · To learn more about all these week’s stories, make sure to check the description box below for all the links.
2:10 · Please remember to like, subscribe, share, and stay safe.
2:13 · I’m Rody for “The Developer Show.”
2:14 · Thanks for watching.
2:15 · And we’ll see you next week.
2:17 · [MUSIC NOTE]