Tools

After 6 years, BERT, the workhorse of encoder models, finally gets a replacement: 𝗪𝗲𝗹𝗰𝗼𝗺𝗲 𝗠𝗼𝗱𝗲𝗿𝗻𝗕𝗘𝗥𝗧! 🤗

We talk a lot about ✨Generative AI✨, meaning "Decoder version of the Transformers architecture", but this is only one of the ways to build LLMs: encoder models, that turn a sentence in a vector, are maybe even more widely used in industry than generative models.

The workhorse for this category has been BERT since its release in 2018 (that's prehistory for LLMs).

It's not a fancy 100B parameters supermodel (just a few hundred millions), but it's an excellent workhorse, kind of a Honda Civic for LLMs.

Many applications use BERT-family models - the top models in this category cumulate millions of downloads on the Hub.

➡️ Now a collaboration between Answer.AI and LightOn just introduced BERT's replacement: ModernBERT.

𝗧𝗟;𝗗𝗥:
🏛️ Architecture changes:
⇒ First, standard modernizations:
- Rotary positional embeddings (RoPE)
- Replace GeLU with GeGLU,
- Use Flash Attention 2
✨ The team also introduced innovative techniques like alternating attention instead of full attention, and sequence packing to get rid of padding overhead.

🥇 As a result, the model tops the game of encoder models:
It beats previous standard DeBERTaV3 for 1/5th the memory footprint, and runs 4x faster!

Read the blog post 👉 https://huggingface.co/blog/modernbert

𝐇𝐮𝐠𝐠𝐢𝐧𝐠 𝐅𝐚𝐜𝐞 𝐫𝐞𝐥𝐞𝐚𝐬𝐞𝐬 𝐏𝐢𝐜𝐨𝐭𝐫𝐨𝐧, 𝐚 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐢𝐜 𝐥𝐢𝐛 𝐭𝐡𝐚𝐭 𝐬𝐨𝐥𝐯𝐞𝐬 𝐋𝐋𝐌 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝟒𝐃 𝐩𝐚𝐫𝐚𝐥𝐥𝐞𝐥𝐢𝐳𝐚𝐭𝐢𝐨𝐧 🥳

🕰️ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.

👴🏻 If they had needed all this time, we would have GPU stories from the time of Pharaoh 𓂀: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons "

🛠️ But instead, they just parallelized the training on 24k H100s, which made it take just a few months.
This required parallelizing across 4 dimensions: data, tensor, context, pipeline.
And it is infamously hard to do, making for bloated code repos that hold together only by magic.

🤏 𝗕𝘂𝘁 𝗻𝗼𝘄 𝘄𝗲 𝗱𝗼𝗻'𝘁 𝗻𝗲𝗲𝗱 𝗵𝘂𝗴𝗲 𝗿𝗲𝗽𝗼𝘀 𝗮𝗻𝘆𝗺𝗼𝗿𝗲! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry.
And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening!

⚡ 𝗜𝘁'𝘀 𝘁𝗶𝗻𝘆, 𝘆𝗲𝘁 𝗽𝗼𝘄𝗲𝗿𝗳𝘂𝗹:
Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this)

Go take a look 👉 https://github.com/huggingface/picotron/tree/main/picotron

🐇 Tumble down the AI rabbit hole without any technical knowledge!

Explore AI models on the Hub by a simple and quick search

Demo: davidberenstein1957/transformers-pipeline-playground

Just created a Gradio space for playing with the new OAI realtime voice API!

freddyaboulton/openai-realtime-voice

Gemini can talk 🗣️

Check out the new multimodal API from Google on @akhaliq 's anychat or my space. It's very fast and smart 🍓

https://huggingface.co/spaces/freddyaboulton/gemini-voicehttps://huggingface.co/spaces/akhaliq/anychat

Introducing the Synthetic Data Generator, a user-friendly application that takes a no-code approach to creating custom datasets with Large Language Models (LLMs). The best part: A simple step-by-step process, making dataset creation a non-technical breeze, allowing anyone to create datasets and models in minutes and without any code.

Blog: https://huggingface.co/blog/synthetic-data-generator
Space: argilla/synthetic-data-generator

𝗣𝗼𝘁𝗲𝗻𝘁𝗶𝗮𝗹 𝗽𝗮𝗿𝗮𝗱𝗶𝗴𝗺 𝘀𝗵𝗶𝗳𝘁 𝗶𝗻 𝗟𝗟𝗠𝘀: 𝗻𝗲𝘄 𝗽𝗮𝗽𝗲𝗿 𝗯𝘆 𝗠𝗲𝘁𝗮 𝗰𝗹𝗮𝗶𝗺𝘀 𝘁𝗵𝗮𝘁 𝘄𝗲 𝗰𝗮𝗻 𝗴𝗲𝘁 𝗿𝗶𝗱 𝗼𝗳 𝘁𝗼𝗸𝗲𝗻𝗶𝘇𝗲𝗿𝘀! 🥳

Current LLMs process text by first splitting it into tokens. They use a module named "tokenizer", that -spl-it-s- th-e- te-xt- in-to- arbitrary tokens depending on a fixed dictionnary.
On the Hub you can find this dictionary in a model's files under tokenizer.json.

➡️ This process is called BPE tokenization. It is suboptimal, everyone says it. It breaks text into predefined chunks that often fail to capture the nuance of language. But it has been a necessary evil in language models since their inception.

💥 In Byte Latent Transformer (BLT), Meta researchers propose an elegant solution by eliminating tokenization entirely, working directly with raw bytes while maintaining efficiency through dynamic "patches."

This had been tried before with different byte-level tokenizations, but it's the first time that an architecture of this type scales as well as BPE tokenization. And it could mean a real paradigm shift! 👏👏

🏗️ 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲:
Instead of a lightweight tokenizer, BLT has a lightweight encoder that process raw bytes into patches. Then the patches are processed by the main heavy-duty transformers as we do normally (but for patches of bytes instead of tokens), before converting back to bytes.

🧩 𝗗𝘆𝗻𝗮𝗺𝗶𝗰 𝗣𝗮𝘁𝗰𝗵𝗶𝗻𝗴:
Instead of fixed tokens, BLT groups bytes based on their predictability (measured by entropy) - using more compute for complex sequences and efficiently handling simple ones. This allows efficient processing while maintaining byte-level understanding.

I hope this breakthrough is confirmed and we can get rid of all the tokenizer stuff, it will make model handling easier!

Read their paper here 👉 https://dl.fbaipublicfiles.com/blt/BLT__Patches_Scale_Better_Than_Tokens.pdf

Version 0.0.21 of gradio-pdf now properly loads chinese characters!

Hello Llama 3.2! 🗣️🦙

Build a Siri-like coding assistant that responds to "Hello Llama" in 100 lines of python! All with Gradio, webRTC 😎

freddyaboulton/hey-llama-code-editor

💥 𝗚𝗼𝗼𝗴𝗹𝗲 𝗿𝗲𝗹𝗲𝗮𝘀𝗲𝘀 𝗚𝗲𝗺𝗶𝗻𝗶 𝟮.𝟬, 𝘀𝘁𝗮𝗿𝘁𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗮 𝗙𝗹𝗮𝘀𝗵 𝗺𝗼𝗱𝗲𝗹 𝘁𝗵𝗮𝘁 𝘀𝘁𝗲𝗮𝗺𝗿𝗼𝗹𝗹𝘀 𝗚𝗣𝗧-𝟰𝗼 𝗮𝗻𝗱 𝗖𝗹𝗮𝘂𝗱𝗲-𝟯.𝟲 𝗦𝗼𝗻𝗻𝗲𝘁! And they start a huge effort on agentic capabilities.

🚀 The performance improvements are crazy for such a fast model:
‣ Gemini 2.0 Flash outperforms the previous 1.5 Pro model at twice the speed
‣ Now supports both input AND output of images, video, audio and text
‣ Can natively use tools like Google Search and execute code

➡️ If the price is on par with previous Flash iteration ($0.30 / M tokens, to compare with GPT-4o's $1.25) the competition will have a big problem with this 4x cheaper model that gets better benchmarks 🤯

🤖 What about the agentic capabilities?

‣ Project Astra: A universal AI assistant that can use Google Search, Lens and Maps
‣ Project Mariner: A Chrome extension that can complete complex web tasks (83.5% success rate on WebVoyager benchmark, this is really impressive!)
‣ Jules: An AI coding agent that integrates with GitHub workflows

I'll be eagerly awaiting further news from Google!

Read their blogpost here 👉 https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/

𝐒𝐜𝐚𝐥𝐢𝐧𝐠 𝐥𝐚𝐰𝐬 𝐚𝐫𝐞 𝐧𝐨𝐭 𝐝𝐞𝐚𝐝 𝐲𝐞𝐭! New blog post suggests Anthropic might have an extremely strong Opus-3.5 already available, but is not releasing it to keep their edge over the competition. 🧐

❓Since the release of Opus-3.5 has been delayed indefinitely, there have been lots of rumors and articles about LLMs plateauing. Scaling laws, the main powering factor of the LLM competence increase, could have stopped, according to these rumors, being the cause of this stalling of progress.

These rumors were quickly denied by many people at the leading LLM labs, including OpenAI and Anthropic. But these people would be expected to hype the future of LLMs even if scaling laws really plateaued, so the jury is still out.

🗞️ This new article by Semianalysis (generally a good source, specifically on hardware) provides a counter-rumor that I find more convincing:

➡️ Maybe scaling laws still work, Opus-3.5 is ready and as good as planned, but they just don't release it because the synthetic data it helps provide can bring cheaper/smaller models Claude and Haiku up in performance, without risking to leak this precious high-quality synthetic data to competitors.

Time will tell! I feel like we'll know more soon.

Read the article: https://semianalysis.com/2024/12/11/scaling-laws-o1-pro-architecture-reasoning-infrastructure-orion-and-claude-3-5-opus-failures/

Just created a cookbook of real time audio/video spaces created using Gradio and WebRTC ⚡️

Use this and the [docs](https://freddyaboulton.github.io/gradio-webrtc/) to get started building the next gen of AI apps!

freddyaboulton/gradio-webrtc-cookbook-6758ba7745aeca7b1be7de0f

Last week was crazy in OS AI, with important models and datasets releases every day.

Here are the most important ones I've pinned:

🌎 Cohere relased GLobal-MMLU, a multilingual version of MMLU, to evaluate AI models' world knowledge in many languages!

🦙 Meta released Llama-3.3-70B-Instruct, a 70B model that's on par with Llama-3.1-405B-Instruct, GPT-4o and Claude. Probably my new go-to for agentic workflows.

🔉 FishAudio released fish-speech-1.5, multilingual text to speech model

🎨 Microsoft Research released TRELLIS, an extremely impressive image-to-3D model, which you can try here: JeffreyXiang/TRELLIS

📚 Yesterday, Hugging Face release FineWeb 2, a new version that extends the previous FineWeb to over 1000 languages, including extended coverage in Russina, Mandarin, German, Japanese, Spanish, French, so a huge, high-quality dataset of > 3 trillion words! HuggingFaceFW/fineweb-2

Now let's go build to make this week as productive as last one!

Open Preference Dataset for Text-to-Image Generation by the 🤗 Community

Open Image Preferences is an Apache 2.0 licensed dataset for text-to-image generation. This dataset contains 10K text-to-image preference pairs across common image generation categories, while using different model families and varying prompt complexities.

https://huggingface.co/blog/image-preferences

This is amazing for cheap models fine-tunes without the hassle of actual deployment! TIL: LoRA fine-tunes for models on the Hub can directly be used for inference!

𝗦𝗵𝗼𝘄𝗨𝗜: 𝗮 𝘀𝗺𝗮𝗹𝗹 𝗲𝗻𝗱-𝘁𝗼-𝗲𝗻𝗱 𝗮𝗴𝗲𝗻𝘁 𝘁𝗵𝗮𝘁 𝗰𝗮𝗻 𝗻𝗮𝘃𝗶𝗴𝗮𝘁𝗲 𝗮𝗻𝘆 𝗨𝗜 𝗮𝗻𝗱 𝗼𝘂𝘁𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝘀 𝗺𝘂𝗰𝗵 𝗯𝗶𝗴𝗴𝗲𝗿 𝘀𝘆𝘀𝘁𝗲𝗺𝘀! 📲

A team from NUS and Microsoft just released an agent that can act on any UI (Desktop, Android, Web) without needing additional text information. It works extremely well : they applied their method on a tiny Qwen2-VL-2B, and they managed to beat methods that use either much more powerful vision models (like GPT-4V) without using any additional info (e.g. leveraging the DOM of a webpage) like previous methods did ! 👏👏

They started from the idea that most existing methods rely heavily on text, which makes them less generalizable, while letting aside rich UI structure that user actually rely on when navigating this interfaces.

⚙️ They put several good ideas to work:

💡 Simplify screenshots to the max:
They prune a lot the heavy visual content of UI screenshots, by removing cloned image patches (like any vast patch of the same color will be reduced to a small patch, while maintaining positional embeddings), then group patches from the same GUI elements together to simplify even further

💡 Build a truly generalist dataset:
To train a general UI agent, you need trajectories from each possible UI, and express them in a common language. Authors merge datasets like OmniAct for Desktop, Mind2Web for websites, AMEX for Android trajectories to create a high-quality and diverse dataset.

➡️ Nice results ensued:
They fine-tune a tiny Qwen-2-VL-2B on their method, and it reaches SOTA on several task (element identification, web navigation), even beating methods that either use additional info from the DOM or use much bigger VLMS like GPT-4v! 🏆

And performance could certainly jump with a slightly bigger vision model. Let's hope the community builds this soon! 🚀

Paper added to my "Agents" collection 👉 m-ric/agents-65ba776fbd9e29f771c07d4e

The Data Is Better Together community is set to release the first Apache 2 licensed image preference dataset!

Great work and let's give this a final push :)

@aashish1904 congrats on your month of HF pro. There is more to win during this sprint!

@aashish1904 @AnyaDesdein @davidberenstein1957 @Malalatiana @beta3 @fffiloni @munish0838 @Reza2kn @bbunzeck @Creazycreator @andrei-saceleanu @jafhaponiuk @rca-etl @kf120 @burtenshaw @mmhamdy @grib0ed0v @Doopus @AnyaDes @ttkap @Xceron @Lewox @davanstrien @Azazelle @adirik @Ashish08 @AntonVic @kenantang @sdiazlor @g-ronimo @dennis-rall @prithivMLmods @girtss3 @flozi00 @WaveCut @Taylor658 @Wildminder @Sara9999 @phaelishall @sararob @dvilasuero @pgabrys @plaguss @CDS899 @timajwilliams @rudzinskimaciej @pavel-ai @aggr8 @ignacioct @MouseAI @Leeps @MaksKul @NicolasDmln @Muinez @kusht55 @caiolang @Jakub-Brand24 @loamy @Demijan @eliab96 @Viewegger @JosephCatrambone @p1atdev @mrshu @o639 @Targezed @Aviv-anthonnyolime @thliang01 @Ahmed-Amine @glards @pranaykoppula @nataliaElv @MaPirlet @alvarobartt @gabrielmbmb @zlicastro @Jaydip @Chouettecheveche @lilcheaty @ruyrdiaz @robintema @fdaudens @ggcristian @a-r-r-o-w @pates @joheras @stopsatgreen @bezo97 @chachi902 @iamyann @liamcripwell @dmb23 @korbih @anonymous7743 @akbdx18 @OVAWARE @severo @akontra @lichorosario @lhoestq @SebastianBodza @Vishnou @ameerazam08 @appoose @Mukei @mearco @joaquincabezas @Fizzarolli @thomastraum @igortopolski @OxxoCodes @patrickfleith @asoria @bn22 @sitammeur @Krodolf @bergr7f @Sbxxn @wietsevenema @sugatoray @Iamladi @MikeTrizna @feveromo @mokady @Bolero @prath @Dowwie @kfahn @decodingchris @alili2050 @RahulRaman @yzimmermann @Ameeeee @ecyht2 @MattMC001 @hemanthkumarak @Thegorgibus @akos2 @LawRun @ramithuh @SuperMuel @sjans @peterizsak @mosama @Eyel @mtr3 @cfahlgren1 @legentil @clem @Citaman @Aurelien-Morgan @AntoineBourgois @TotoB12 @Stanmey @osanseviero @multimodalart @maxiw @ariG23498 @ngk89 @femboysLover @dvs @tacohiddink @blanchon @DavidJimenez

Need a measurement for traction of a GitHub repo, a more reliable one than Github star history? (which is a bit too hype-driven) 📈

➡️ I've made a Space to visualize PyPI downloads.

Try it here 👉 m-ric/package-download-history

🤖 𝗔𝗱𝗼𝗯𝗲'𝘀 𝗰𝗼𝗱𝗲-𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁 𝗿𝗲𝗮𝗰𝗵𝗲𝘀 𝘁𝗵𝗲 𝘁𝗼𝗽 𝗼𝗳 𝗚𝗔𝗜𝗔 𝗹𝗲𝗮𝗱𝗲𝗿𝗯𝗼𝗮𝗿𝗱 - and their paper cites my work!

💡 Reminder: In short, Agentic systems are a vehicle in which you put your LLM to allow it access to the outside world.

➡️ The team of researchers at Adobe started from the idea that current agentic systems lack the ability to define their own tools. So they decided to make an agent that writes actions as code, thus allowing it to write python functions that can be re-used later as tools!

Here's what the LLM generations can look like with the proper prompt:

Thought: I need to access the excel file using a different method.
Action:


Then your system executes this and appends the observation to the agent's memory.

Why is this code formulation better than classical tool use formulation as JSON? The paper explains:

"Most existing work uses text or JSON as the representation of actions, which significantly lacks the two criteria mentioned earlier: generality and composability. In contrast, DynaSaur can utilize available actions or create new ones if necessary, using code as a unified representation. In principle, acting with code enables agents to solve any Turing-complete problem."

The idea of using code is not new: in fact, we do it in transformers.agents (thus the citation that I got). They implementation adds further refinements, like using RAG to retrieve relevant functions before generating an action, which increases performance further.

And they observe that code agents perform much better, reaching the top of GAIA leaderboard! 🥇

Go take a look, it's really clear and informative!

Paper added to my agents collection 👉 m-ric/agents-65ba776fbd9e29f771c07d4e

Single most important thing to do today: 𝗴𝗼 𝘁𝗿𝘆 𝗤𝘄𝗤 𝗼𝗻 𝗛𝘂𝗴𝗴𝗶𝗻𝗴 𝗖𝗵𝗮𝘁!

👉 https://huggingface.co/chat/models/Qwen/QwQ-32B-Preview