I WAS SO FRUSTRATED with the way GitHub renders Jupyter Notebooks, esp on mobile (You cannot horizontal scroll code, even with nbviewer)

So I made a thing: nbsanity.com

Source code is here github.com/hamelsmu/nbsan…

Now 24 hours post release, the model is bad according to every single person who tried it. Turns out the 'secret sauce' was overfiting the benchmarks.

I battled against GPT4 for quite a while trying to encourage it to solve another version of this problem in this video: youtu.be/jkrNMKz9pWU?si…

'Do stuff and then blog about it' remains one of the most underrated pieces of career advice

The PyTorch team is developing a library for large model training called torchtitan 👀

They have scripts to train Llama-3 from scratch

The library went public today on GitHub but it is still in pre-release state & active development

Check it out → github.com/pytorch/torcht…

This take on the FineWeb release is one of the most interesting feedback and also a reason FineWeb is very different from even larger datasets like RedPajama-V2 (which is double its size!)

Surprisingly, the size of the dataset of 15T tokens is not very important, what is much…

Phi-3 Mini (3.8B) from Microsoft was released today, claiming to match Llama 3 8B's performance! But how does it handle RAG, Routing, Query Planning, Text2SQL, Pydantic Program, and Agentic tasks?

Thanks to Ravi Theja, our benchmark cookbook offers an initial analysis:
✅ RAG…

Turns out if you do a cute little hack, you can make musicgen-songstarter-v0.2 work on vocal inputs 👀

🎤 Sing an idea ➡️ AI music sample 🎶

🔊Sound on 🔊

My goal was to make something useful for music producers 🔥

Demo, API, code + detailed info in thread below ⤵️

LLM Evaluators Recognize and Favor Their Own Generations

abs: arxiv.org/abs/2404.13076

1. Frontier LLMs exhibit self-preference in self-evaluation.
2. LLMs have non-trivial self-recognition capability out of the box.
3. Fine-tuning leads to near-perfect self-recognition.
4.…

Exciting update -- Llama-3 full result is out, now reaching top-5 on the Arena leaderboard🔥

We've got stable enough CIs with over 12K votes. No question now Llama-3 70B is the new king of open model. Its powerful 8B variant has also surpassed many larger-size models. What an…

🎉 Great news for model merging!

Charles Goddard implemented an evolutionary technique à la Sakana AI to MergeKit (cc hardmaru)

He also released an excellent tutorial on how to use it with lm-evaluation-harness and vllm.

📝 Article: blog.arcee.ai/tutorial-tutor…

very nice blog post on training a VLM purely in pytorch huggingface.co/blog/AviSoori1… 🔖

A big reason this works so well is because Benjamin Warner has put in many *many* hours over the last few weeks on debugging and fixing complex performance issues deep inside the Transformers library.

It's the kind of work that's rarely appreciated -- so let's do so now! :D

Today at Answer.AI we've got something new for you: FSDP/QDoRA. We've tested it with AI at Meta Llama3 and the results blow away anything we've seen before.

I believe that this combination is likely to create better task-specific models than anything else at any cost. 🧵

This is the best tutorial I've seen for fully understanding and implementing Transformers models.

It includes a complete working implementation from scratch of all the key pieces, written in a way to make learning and understanding as easy as possible.

New paper from OpenAI on prompt injection - it's the most detailed evaluation of the problem I've seen from them so far, and has some very interesting details

Posted some of my notes on the paper on my log here: simonwillison.net/2024/Apr/23/th…

Do yourself a favor and try Llama3 70B with Groq. GPT-4 level answers provided instant. Insane.

FB/Meta senior mgt has cared a lot about AI for many years. By way of example, here's a true story I haven't told before...

Do you remember when Zuck was giving testimony to congress in April 2018? I was watching it live, when the phone rang. It was the CTO of Facebook.

Models dropped on Hugging Face!

huggingface.co/meta-llama/Met…
huggingface.co/meta-llama/Met…

we just shipped HuggingChat on iOS 💬

The app is super polished and gives you access to the community's best open AI models, on the go.

Give it a try!

link to Appstore below ⤵️