# March 2024


Read updates from the new release, run LLMs on-device with MediaPipe LLM Inference API (experimental), learn how ML can reduce operational costs, and more.

|---|
| |---| | |---| | [![](https://gstatic.com/growthlab/api/5bVWsgT8ttKvlVaE5gRkjgAiBv9mwnyIQkgpcp0q.png)](https://www.tensorflow.org/subscribe/2024/03#) | | | |---| | |---| | |---| | What's new in TensorFlow 2.16 | | | |---| | |---| | |---| | |---| | Updates include Clang as default compiler for building TensorFlow CPU wheels on Windows, Keras 3 as default version, support for Python 3.12, and more! | | | |---| | [Read the blog](https://blog.tensorflow.org/2024/03/whats-new-in-tensorflow-216.html?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | | | |---| |   | | | |---| | |---| | |---| | |---| | |---| | |---| | |---| | [![](https://services.google.com/fh/files/emails/io24_svd_twitter_banner_327.png)](https://io.google/2024/?utm_source=newsletter&utm_medium=email&utm_campaign=tensorflow-mar24) | | Google I/O is back! | | Join us online on May 14 to discover, innovate, and unlock new possibilities with AI and all of Google's latest developer tools. | | |---| | |---| | |---| | |---|---| | |---| | [Register now](https://io.google/2024/?utm_source=newsletter&utm_medium=email&utm_campaign=tensorflow-mar24) | | |---| | [![](https://www.gstatic.com/growthlab/api/Obmfy5bg9fJotE3zdZDXu4HiZx1oT3KnprcfW3ED.png)](https://io.google/2024/?utm_source=newsletter&utm_medium=email&utm_campaign=tensorflow-mar24) | | | | | | | | | | | |---| | |---| | |---| | |---| | [![](https://services.google.com/fh/files/emails/tokopedia.png)](https://web.dev/case-studies/tokopedia-ml#why_choose_ml_as_opposed_to_other_solutions?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | How Tokopedia reduced operational costs using ML for identity verification | | The team at Tokopedia decided to use MediaPipe and TensorFlow.js's Face Detection library to detect a user's face with six key points during the seller onboarding process. | | |---| | |---| | |---| | |---|---| | |---| | [Read the blog](https://web.dev/case-studies/tokopedia-ml#why_choose_ml_as_opposed_to_other_solutions?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | |---| | [![](https://www.gstatic.com/growthlab/api/Obmfy5bg9fJotE3zdZDXu4HiZx1oT3KnprcfW3ED.png)](https://web.dev/case-studies/tokopedia-ml#why_choose_ml_as_opposed_to_other_solutions?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | | | | | | | | |---| | |---| | |---| | |---| | [![](https://services.google.com/fh/files/emails/media-pipe.png)](https://developers.googleblog.com/2024/03/running-large-language-models-on-device-with-mediapipe-andtensorflow-lite.html?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | Run large language models on-device with MediaPipe and TensorFlow Lite | | The new experimental MediaPipe LLM Inference API includes performance optimizations such as quantization, caching, and weight sharing to run LLMs on Web, Android, and iOS with low latency. | | |---| | |---| | |---| | |---|---| | |---| | [Read the blog](https://developers.googleblog.com/2024/03/running-large-language-models-on-device-with-mediapipe-andtensorflow-lite.html?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | |---| | [![](https://www.gstatic.com/growthlab/api/Obmfy5bg9fJotE3zdZDXu4HiZx1oT3KnprcfW3ED.png)](https://developers.googleblog.com/2024/03/running-large-language-models-on-device-with-mediapipe-andtensorflow-lite.html?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | | | | | | | | |---| | |---| | |---| | |---| | [![](https://services.google.com/fh/files/emails/applied-ml.png)](https://www.youtube.com/watch?v=8aAghDxgXhY?utm_source%3Demail&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | Learn applied ML with KerasNLP and KerasCV | | Watch new videos explaining how to use the KerasNLP and KerasCV libraries for applications including: | | |---| | |---|---| | |---| | • | | |---| | [Object detection](https://www.youtube.com/watch?v=zl5l6s6eb3o?utm_source%3Demail&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | |---|---| | |---| | • | | |---| | [Image data augmentation](https://www.youtube.com/watch?v=URG92p1bBXA&t=17s?utm_source%3Demail&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | |---|---| | |---| | • | | |---| | [Text generation](https://www.youtube.com/watch?v=GgTQr6BoDAY?utm_source%3Demail&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | |---|---| | |---| | • | | |---| | [Text classification](https://www.youtube.com/watch?v=Wb8-1O8bW68?utm_source%3Demail&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | | |---| | |---| | |---| | |---|---| | |---| | [Watch the overview](https://www.youtube.com/watch?v=8aAghDxgXhY?utm_source%3Demail&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | |---| | [![](https://www.gstatic.com/growthlab/api/Obmfy5bg9fJotE3zdZDXu4HiZx1oT3KnprcfW3ED.png)](https://www.youtube.com/watch?v=8aAghDxgXhY?utm_source%3Demail&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | | | | | | | | |---| | |---| | |---| | |---| | [![](https://services.google.com/fh/files/emails/datasets.png)](https://blog.research.google/2024/03/croissant-metadata-format-for-ml-ready.html?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | Make datasets easier to find and use with Croissant | | Use TensorFlow Datasets [CrossaintBuilder](https://www.tensorflow.org/datasets/format_specific_dataset_builders#croissantbuilder?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) to define Croissant metadata, a format that builds upon [schema.org](https://www.tensorflow.org/subscribe/2024/03#) structure and combines resource file descriptions, data structure, and default ML semantics into a single file. | | |---| | |---| | |---| | |---|---| | |---| | [Read the blog](https://blog.research.google/2024/03/croissant-metadata-format-for-ml-ready.html?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | |---| | [![](https://www.gstatic.com/growthlab/api/Obmfy5bg9fJotE3zdZDXu4HiZx1oT3KnprcfW3ED.png)](https://blog.research.google/2024/03/croissant-metadata-format-for-ml-ready.html?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | | | | | | | | | | | |---| |   | | | |---| | ![](https://www.gstatic.com/gumdrop/files/spacer-11.gif) | | Stay Connected | | |---|---|---|---|---|---|---|---|---| | [![TensorFlow blog](https://www.gstatic.com/growthlab/api/e7WEFvQxwbiFjhuJkQsF5Fq09BSs8t0t7zjVc2ZZ.png "TensorFlow blog")](http://blog.tensorflow.org/?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) |   | [![GitHub](https://www.gstatic.com/growthlab/api/3oLbQMNQ8xdPTrZaFXgbpypikCrz1cifbZnJ7hYv.png "GitHub")](https://github.com/tensorflow?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) |   | [![X](https://gstatic.com/growthlab/api/qTLa5CDPfCaU74g6OzQPE546CJPaqLheI2Vw8qev.png "X")](https://twitter.com/tensorflow?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) |   | [![YouTube](https://www.gstatic.com/growthlab/api/Hmw6zspJ6zRcW9dalh8m4Yztyv80TMDVsycNLpCi.png "YouTube")](https://www.youtube.com/tensorflow?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) |   | [![TensorFlow forum](https://www.gstatic.com/growthlab/api/o2mjGQiE1CeSDhViQ9I47utHTpyx929ckrYzUOoR.png "TensorFlow forum")](http://discuss.tensorflow.org/?utm_source=email&utm_medium=newsletter&utm_campaign=tensorflow-mar24) | | | | | |

![](https://notifications.google.com/g/proof/img/ANiao5qlLy166lCIbMaYVFQZ8zytS94sapN5WtGn4jGCYxSqmDUBWeuN2duE8En0riRoQBd0-2N6A5lfgdKiHonCKxfOhYdV3BP-TqLsTSnFwlwmXX1O9SbzQjxh2Gz5TGciE7HEl5iYVNzd0uOAho5KLrSakQ.gif)