What’s new in TensorFlow
TensorFlow is an open-source machine learning library developed by Google. It is widely used for building and training machine learning models. With a strong community of developers and researchers, TensorFlow continues to evolve and improve. Here are some of the latest updates in TensorFlow:
TensorFlow 2.7 Release
The latest version of TensorFlow, 2.7, brings several new features and improvements. This release includes updates to TensorFlow’s Keras API, improved support for distributed training, and enhanced performance for large models. TensorFlow 2.7 also introduces new experimental features such as support for sparse tensors and performance improvements for TensorFlow.js.
TensorFlow Extended (TFX)
TensorFlow Extended is a platform for deploying machine learning models in production. The latest updates to TFX include better integration with TensorFlow 2.x and improvements to the TFX pipeline components. TFX also introduces new tools for model validation and monitoring, making it easier to deploy and maintain machine learning models at scale.
TensorFlow Lite
TensorFlow Lite is a lightweight version of TensorFlow designed for mobile and edge devices. The latest updates to TensorFlow Lite include support for new hardware accelerators, improved performance for on-device inference, and new tools for model optimization. TensorFlow Lite continues to be a popular choice for deploying machine learning models on resource-constrained devices.
TensorFlow.js
TensorFlow.js is a JavaScript library for building and training machine learning models in the browser. The latest updates to TensorFlow.js bring support for new layers and activations, improved performance for training and inference, and better compatibility with TensorFlow 2.x models. TensorFlow.js remains a powerful tool for developing web-based machine learning applications.
Conclusion
With each new release, TensorFlow continues to push the boundaries of what is possible in machine learning. Whether you are building models for research, deploying them in production, or creating interactive applications, TensorFlow offers a comprehensive set of tools and resources to support your work. Stay tuned for more updates and advancements in the world of TensorFlow!
Watch more AI/ML Sessions → https://goo.gle/IO23_aiml_pin
fonts are small
No one cares about Tensorflow anymore, I started out with Tensorflow and didn't look back once after being introduced to PyTorch.
Pass.
I've already invested into learning Pytorch, cuz all the excitement had shifted there. Additionally I don't wanna be locked into Google hardware.
Learning a new library (TF), that gives the same benefits isn't worth overcoming the sunk cost. Besides, Pytorch supports so many other kinds of hardware including M1 MacOs.
Would like to see a 2023 detailed feature comparison of JAX2TF vs PyTorch 2 ecosystem – tech-skill development is a big time investment, so best to make the optimal strategic choice beforehand. At this point, I'm revising TF and learning JAX
the transformer_encoder in kerasNLP isn't even bi-directional like the ones in TF.NLP library. I'd rather use tf than keras. Also, I was wondering for a while, is the GPUDirect Storage implemented already in TF? or there are few extra steps I need to add to my code
4:40 Casual LM, I prefer my LMs to be formal
I love all this, I think the new Quantization API will be the thing I'll be waiting for the most
With the affordability of nvme and more and more people trying to fit pretrained LLM models on their local machines for inference I'd really like to see a non proto buff config/ibm solution (native) for spilling models to disk/virtual memory and out of VRAM with builder esque configuration in keras and Tensorflow graph. There's no respectable reason this doesn't exist. Hire me and I'll write it I'm not just complaining.
I anticipate the release of the TF Quantization API.
Having a tf implementation of quantization without requiring tflite conversion is going to be amazing.
❤😊