Hugging Face's TensorFlow Philosophy

📰 Hugging Face Blog

Hugging Face outlines its TensorFlow philosophy, emphasizing Keras models, flexible loss functions, and pre-built data pipelines

intermediate Published 12 Aug 2022

Action Steps

Use Keras Model objects for all TensorFlow models
Use Keras Layer objects for all TensorFlow layers
Utilize pre-built loss functions and modify them as needed
Leverage pre-built data pipelines for common tasks
Explore XLA for optimized model performance

Who Needs to Know This

This philosophy benefits machine learning engineers and data scientists on a team by providing a standardized approach to building and deploying TensorFlow models, making it easier to collaborate and integrate models into larger projects

Key Insight

💡 Standardizing on Keras models and pre-built components can streamline TensorFlow development and improve collaboration

Key Takeaways

Hugging Face outlines its TensorFlow philosophy, emphasizing Keras models, flexible loss functions, and pre-built data pipelines

Full Article

Published Time: 2022-08-12T00:00:00.115Z

# Hugging Face's TensorFlow Philosophy

[![Image 1: Hugging Face's logo](https://huggingface.co/front/assets/huggingface_logo-noborder.svg)Hugging Face](https://huggingface.co/)

* [Models](https://huggingface.co/models)
* [Datasets](https://huggingface.co/datasets)
* [Spaces](https://huggingface.co/spaces)
* [Buckets new](https://huggingface.co/storage)
* [Docs](https://huggingface.co/docs)
* [Enterprise](https://huggingface.co/enterprise)
* [Pricing](https://huggingface.co/pricing)
*
*
* * *

* [Log In](https://huggingface.co/login)
* [Sign Up](https://huggingface.co/join)

[Back to Articles](https://huggingface.co/blog)

# [](https://huggingface.co/blog/tensorflow-philosophy#hugging-faces-tensorflow-philosophy) Hugging Face's TensorFlow Philosophy

Published August 12, 2022

[Update on GitHub](https://github.com/huggingface/blog/blob/main/tensorflow-philosophy.md)

[- [x] Upvote 7](https://huggingface.co/login?next=%2Fblog%2Ftensorflow-philosophy)
* [![Image 2](https://huggingface.co/avatars/86c6b09258797c1d0f3a45d8222ab1f2.svg)](https://huggingface.co/petnov "petnov")
* [![Image 3](https://cdn-avatars.huggingface.co/v1/production/uploads/64b695dcd3df8086e5ed7c89/ZULwnPE8Q5q7bxKJsJ7v5.jpeg)](https://huggingface.co/adamelliotfields "adamelliotfields")
* [![Image 4](https://huggingface.co/avatars/02260341aeecd8f604a3638c47c7797a.svg)](https://huggingface.co/Waras "Waras")
* [![Image 5](https://huggingface.co/avatars/8f15416a87a53b5a5a0710ea14fc4438.svg)](https://huggingface.co/MohammedSalem98 "MohammedSalem98")
* [![Image 6](https://cdn-avatars.huggingface.co/v1/production/uploads/66ed87cc518ccb4e62843f25/11gpPSxwKcDTv1sCeU2OQ.jpeg)](https://huggingface.co/tbitai "tbitai")
* [![Image 7](https://huggingface.co/avatars/53a9145b2b6045d12acdb1e606b20e6f.svg)](https://huggingface.co/mattsauer "mattsauer")
* +1

[![Image 8: Matthew Carrigan's avatar](https://cdn-avatars.huggingface.co/v1/production/uploads/1660312628256-60ba519750effef3a58beac3.png)](https://huggingface.co/Rocketknight1)

[Matthew Carrigan Rocketknight1 Follow](https://huggingface.co/Rocketknight1)

### [](https://huggingface.co/blog/tensorflow-philosophy#introduction) Introduction

* [Introduction](https://huggingface.co/blog/tensorflow-philosophy#introduction "Introduction")

* [Interlude: 30 Seconds to 🤗](https://huggingface.co/blog/tensorflow-philosophy#interlude-30-seconds-to-%F0%9F%A4%97 "Interlude: 30 Seconds to 🤗")
* [Philosophy #1: All TensorFlow models should be Keras Model objects, and all TensorFlow layers should be Keras Layer objects.](https://huggingface.co/blog/tensorflow-philosophy#philosophy-1-all-tensorflow-models-should-be-keras-model-objects-and-all-tensorflow-layers-should-be-keras-layer-objects "Philosophy #1: All TensorFlow models should be Keras Model objects, and all TensorFlow layers should be Keras Layer objects.")

* [Philosophy #2: Loss functions are provided by default, but can be easily changed.](https://huggingface.co/blog/tensorflow-philosophy#philosophy-2-loss-functions-are-provided-by-default-but-can-be-easily-changed "Philosophy #2: Loss functions are provided by default, but can be easily changed.")

* [~~Philosophy~~ Implementation Detail #3: Labels are flexible](https://huggingface.co/blog/tensorflow-philosophy#philosophy-implementation-detail-3-labels-are-flexible "Philosophy Implementation Detail #3: Labels are flexible")

* [Philosophy #4: You shouldn’t have to write your own data pipeline, especially for common tasks](https://huggingface.co/blog/tensorflow-philosophy#philosophy-4-you-shouldnt-have-to-write-your-own-data-pipeline-especially-for-common-tasks "Philosophy #4: You shouldn’t have to write your own data pipeline, especially for common tasks")

* [Philosophy #5: XLA is great!](https://huggingface.co/blog/tensorflow-philosophy#philosophy-5-xla-is-great "Philosophy #5: XLA is great!")

* [Philo

Read full article → ← Back to Reads

Hugging Face's TensorFlow Philosophy

Key Takeaways

Full Article

Related Videos