The way AI transformers work reminds me of the 1990's. Back in the 90s in Gnome-land, PNG was not in wide use yet, and we needed some sort of transparency for images. So we used a library that faked it by using an uncommon color to mean "transparent" 1/

What's risk of getting COVID outside? VERY LOW. Previous meta-analysis showed outside transmission 20x less likely; this study shows just 1/1000 cases (0.1%) traced to outdoor transmission. States (CA, low cases) could peel off outside mask mandates now irishtimes.com/news/ireland/i…

"Could Hugging Face Accelerate really be this easy?" I asked myself, and the result is this blog post where we take a deep-dive into the source code of the package. wandb.ai/wandb_fc/pytor… Thanks Sylvain Gugger - you've done it again!! A thread: 1/n

Thanks Mark Saroufim for this "no bullshit" morning read. The tone of this makes me feel both calm and excited. THIS IS THE PLACE. The Rise of HuggingFace 🤗, by Mark Saroufim marksaroufim.substack.com/p/huggingface?…

Hugging Face feels like DataRobot for NLP but also open-source and infinitely hackable. It's truly a magical tool for ML. Haven't even finished exploring all of it yet. Also, PRs are reviewed and triaged incredibly fast. Almost nothing is stale. crazy.

Happy to announce that today is my first day at Hugging Face. Looking forward to meet the new team. First project will be on better integration of the huggingface hub into SentenceTransformers - Sharing your own SBERT.net models will become super easy!

Kustomer Supports Trans Youth on Bonfire bonfire.com/allies-3/?utm_…

The wait is almost over!

Today we are delighted to introduce the SiFive Performance family, including our fastest RISC-V processor: The SiFive Performance P550. Read more in here: hubs.la/H0QKZTB0 #sifive #riscv #performance #innovation

🤗 Transformers v4.9.0: 🟠 Brand new TensorFlow Examples 🟠: Examples for many NLP tasks are now available using Keras only, thanks to Matthew Carrigan! 🚀CANINE: Tokenizer-free, character-based model 🚂 Train a tokenizer from another: same config, different dataset!

Big project, first public release! 📢 ✨ SQLModel ✨ ...the biggest thing I've built since FastAPI & Typer. 🤓 SQL DBs based on Python type hints. 😎 Each model is both a Pydantic and SQLAlchemy model. 🤯 Optimized for FastAPI 🚀 github.com/tiangolo/sqlmo…

Writing simple code is not about writing code that is compact and terse. It's about the simplicity of the underlying mental model and the ease with which you can recover this mental model from reading the code.

Trials and tribulations of OPT-175B Training! It's amazing that you managed to finish the training in fp16. There should be an Oscar-like prize for the brave engineers who managed to pull off such a feat. This is so similar to our pre-BLOOM battles. youtube.com/watch?v=p9IxoS…

As I was just watching the OPT-175 training trials and tribulations, I think there are so few of us who have been doing this incredibly difficult LLM training work. We need to create an open group and share our discoveries and help each other during these trainings. Yes?

If you're a founder/CEO and struggling today, feel free to reach out to me or other founders, my DMs are open, we need to support one an other more than ever on days like this!

OK, I'm calling it: Large language models are having their Stable Diffusion moment right now simonwillison.net/2023/Mar/11/ll…

clem 🤗 Sincerely thank you for putting out the message. Chaos is not random it is rhythmic Let's make sense of it collectively.

Excited for y'all to watch my interview with Jeremy Howard at our user conference on March 15! We chat about making LLM training more accessible to everyone, what it means that most ppl are consuming models rather than producing them & more! Register at fullyconnected.com

I just got off the Zoom with Tanishq Mathew Abraham, Ph.D. and Jonathan Whitaker where we completed recording the final fast.ai part 2 video -- there are 18 of them! I'm working on creating the website for the new course now and will let you all know as soon as it's ready! :D