We’re excited to release the weights of our Time Series Foundation Model (TimesFM) on Hugging Face!

To access, visit our HuggingFace (huggingface.co/google/timesfm…) & GitHub (github.com/google-researc…) repositories.

Learn more ↓
#TimesFM #TimeSeries #Forecasting #FoundationModels

I was trying to figure out what was occupying so much memory in my computer for months because I didn't want to wipe it
today I removed 58 GB of a computer vision dataset that I downloaded a while ago
always check your datasets cache fam

public appreciation for Zach Mueller for maintaining accelerate which saves forever amount of time

So literally the day after I made this tweet I saw this great poster ICLR 2024 by Kim-Celine Kahl, Paul F. Jäger and others. It’s fantastic work on uncertainty estimation for segmentation methods.
Paper: arxiv.org/abs/2401.08501

merve you might find this interesting :)

I hate it when I try to find download links to datasets in dataset websites or where the weights are in a github repository through the code
I likely saved around 100 hrs thanks to Hugging Face repos (unbiased review)

If you're on Hugging Face Spaces and click on the Running button, you now get:

- Hardware information
- Live CPU and RAM usage
- Live usage of other Spaces

I've updated the guide for Hugging Face object detection transformers!

What's New?
⚡ Enhanced Quality of fine-tuned models
🎯 Evaluation in Trainer API with mAP
🏞 Transforms with Albumentations

--
📖 Guide twtr.to/YCKLv
👀 Examples twtr.to/izgGZ

We're very excited about this feature -- you can now publish your KerasNLP & KerasCV models to Kaggle Models and Hugging Face Hub in just one line of code! Try it out and let us know what you think.
developers.googleblog.com/en/publish-you…

Confidence calibration is one of the biggest problems of our time :)

It’s a bad word for an ill posed problem. It makes you think of probability but it’s not. Score might be better, but I’m still on the fence 🤷‍♀️

Have you checked out our latest short course with @HuggingFace, Quantization in Depth?

You will get hands-on and build a quantizer in PyTorch that can compress the dense layers of any open source model, from 32 bits to 8 bits.

Explore the course now: hubs.la/Q02wpZDs0

📰 Exciting news: you can now publish your fine-tuned models straight from the Keras API with @Kaggle or Hugging Face!

Here's your guide to getting started on either platform → goo.gle/3JRKD9h

I'm trying to distil foundation models for vision and the biggest problem is how the confidence threshold sometimes requires a change, sometimes the model acts weirdly too
tbh except for dynamically changing the threshold I can't think of anything else 🥲