Breaking: we got 1,200 tokens per second for Llama 2 7B on H100! 🔥🔥🔥 Check out our new project optimum-NVIDIA with fp8 inference support and first benchmarks for LLM acceleration: hf.co/blog/optimum-nvidia 🧵 👇

Transcribing 1 hour of audio for less than $0.01 🤯 The @huggingface team cooked with 8x faster Whisper speech recognition - @OpenAI whisper-large-v3-turbo transcribes at 100x real time on a $0.80/hr L4 GPU!

Document AI use cases: checkmate! ♟️👑 Introducing SmolDocling SmolDocling is ideal for enterprise use cases: 🤏 256M parameters - cheap and easy to run locally 🏆 Performs better than 20x larger models! 💨 Fast inference using VLLM – Avg of 0.35 secs per page on A100 GPU. ⚖️ Apache 2.0 license

Today @huggingface took a big step democratizing state of the art machine learning, by reducing the price of our $199/mo Lab and $599/mo Startup subscriptions to... $0️⃣ 🤯💸 Pay as you go, or contact sales for more. Let's put transformers to work!

Who's got two hands and should be publicly owned?

Excited to welcome @AMD to the @huggingface Hardware Partner Program!

🚨 Prediction: this will break the Internet 💥 Introducing LEDITS - Image concept editing. Make your dreams come true ✨ huggingface.co/spaces/editin…

Machine Learning these days (call it AI if you must) feels a lot like humans teaching machines to teach humans to teach machines… ad infinitum [1/14]🧵👇

Today @awscloud launched SageMaker Studio Lab to make ML more accessible, it's been fun collaborating toward the launch and lovely to see some Hugs Inside! 😍 studiolab.sagemaker.aws/

😍 Today Satya Nadella announced an expanded collaboration with Hugging Face - with over 11,000 of the most popular open models already available on Azure AI Foundry for easy deployment! 🚀

🥁 Today we announce a new collaboration with @nvidia to connect AI Researchers with GPU Clusters! 🤝 Introducing Training Cluster as a Service, powered by the new NVIDIA DGX Cloud Lepton. We hope this new service will help bridge the compute gap between the GPU rich and the GPU poor, and enable AI Research teams around the world to advance all scientific domains and benefit global communities. AI is too important to be centralized!

Excited to see @Google upgraded to Enterprise on @huggingface! With over 2,500 members and more than 18,000 followers, the Google organization is one of the most important for the AI community, with over 50 million monthly downloads of models, like Gemma 3n released just last week. With Enterprise, Google can secure its organization with SSO, implement access controls, enforce 2-factor authentication, get advanced download analytics and more. It's a big win for the 10 million AI builders on Hugging Face - knowing we can count on the Google models our apps rely on.

Today is a huge milestone for one of our latest libraries: Text Generation Inference - we released v1.0 and under a new license: HFOIL 1.0 github.com/huggingface/text-… This 🧵 explains what this new license means, and why the change!

Today at @Microsoft Build, @satyanadella announced a deepened partnership with @huggingface, with new experiences across cloud, hardware, open source and developers. Here's a round up of all the joint work we announced this week! huggingface.co/blog/microsof… - Azure AI Studio Model Catalog - Azure AMD GPU MI300X VMs - Phi-3 open models - WebGPU with transformers.js and optimum - Spaces Dev Mode

People ask me - so how does @huggingface make money exactly? To answer that, here's an overview of 🤗 Enterprise Hub - how companies build AI privately with added security, access controls and compute. This is the first of a series of short videos - the next ones will focus on a specific feature of Enterprise Hub. 🔍 To learn more and get started: hf.co/enterprise Also, for Hugging Face I'll do anything, even wear a tie 🤣

Hugging Face + Google = Open Source AI FTW! 🤗🤝🌎 All companies should be able to build their own AI and we will make that super easy for Google Cloud customers 🔥🔥🔥 👀 Learn more: huggingface.co/blog/gcp-part… 🍿Watch my interview with @thebradsmith finance.yahoo.com/video/goog…

🤯🤯 That's insane! @huggingface just shipped Data Studio on dataset pages... Ask a question about any dataset, the agent writes and runs a SQL query... then share the result or download as CSV... check it out! 👀 ⤵️ Wow GG @calebfahlgren 👏

👀 We show you how to use @deepseek_ai R1 on @awscloud Many options today, and more to come ⤵️

🆕 Today we announce Train on DGX Cloud, a new service built with @nvidia to easily train LLMs and image generation models on NVIDIA H100 GPUs, without code, directly on the Hub. 🤯 Here's how it works: huggingface.co/blog/train-dg…

4 years ago we were on the brink of AI becoming proprietary and centralized, when OpenAI kept GPT3 closed and VCs started dumping money on researchers. From fully open science, to fully closed, in a matter of months. It was scary, and 1,000+ leading researchers and scientists banded together to show the world that it was possible to do the same work in the open, and build an ecosystem that benefits everyone. That was the @BigscienceW BLOOM project, and it put us back on track to open science, starting with forward-thinking organizations like @Meta releasing OPT. Look at us now. Open models have not only caught up, they're state of the art now. Not just LLMs, but models for document AI, speech to text, text to speech, generating images and more. We're closing in on 2 million open weight models on @huggingface. Thanks for the reminder @Thom_Wolf .

Can’t spill the beans just yet, but we’ve had early access to @AMD new MI355X GPUs, and wow, AMD really delivered! We’re currently running over8️⃣0️⃣,0️⃣0️⃣0️⃣ tests to make sure @huggingface models go brrr when you get your hands on these GPUs... but when?? 🤔 Find out at the Advancing AI event next Thursday, June 12, in San Jose. Dr. @LisaSu keynote will have all the answers 🍿, before hands-on testing in the afternoon! 🕹️ tix ➡️ web.cvent.com/event/c137f02f…

We're ready for you! Join me and team @huggingface at the #AISummit NYC Wed 12/8 Thu 12/9 and ask us anything! 👋🗽 I will speak about accelerating Transformers to millisecond latency and our work with @IntelAI ⏰ Wed 1:45pm EST newyork.theaisummit.com/

Breaking news! 🗞️📣 Hope you’re not allergic to robots — @huggingface has acquired @pollenrobotics 🐝🤖 Calling all educators and tinkerer bees: with Pollen and @LeRobotHF, we’re building open-source robots that are 100x cheaper, fun to use, fully hackable — and built to decentralize AI robotics so everyone can benefit. 🤗 (Article link in 🧵)

Wow. Dr. @LisaSu closing remarks resonated hard with me. I feel like @ClementDelangue could have used the exact same words. Proud to be an @AMD partner! "The future of AI is not going to be built by any one company, or in a closed ecosystem. It's going to be shaped by open collaboration [...]"

Breaking - Gemma 3 from @Google is live on @huggingface ! 💎🤗 Gemma 3 in a nutshell: 🖼️ Multimodal - understands images, videos and text, generates text 🌎 Multilingual - 140 languages! 📲 On-device ready - sizes of 1B, 4B, 12B, 27B 🧠 Hella useful - 27B scores on par with OpenAI o1 on Chatbot Arena (*) 📏 Mucho Capable - 128k tokens context window Resource links in 🧵!

GRPO is how DeepSeek AI created R1, the best open reasoning model. The latest TRL update makes GRPO 60x faster! 🏎️💥 This all comes in the v0.16 update of TRL that shipped today... 🐦‍ multi-step optimization for 6x speedup 🩺 global normalization from Dr GRPO 🤸‍♀️ domain-specific rewards 🍃 beta=0.0 memory savings 🕊️ padding-free batching for SFT Congrats @QGallouedec and team!

The @huggingface Hub is officially open for business with the ENTERPRISE 🏢🔒 Give your team the most advanced platform to build AI with enterprise-grade security (SOC2 Type2), access controls (SSO), and dedicated support. 👀 hf.co/pricing

Loving the new @huggingface Paris office! 😍

Announcing our new partnership with @awscloud ! 🤗 bloomberg.com/news/articles/… Tip @Techmeme

What does 2,000 tokens per second feel like? ⚡️ This video is not accelerated, I just asked Llama3 70B if pineapple should be legal on pizza 🍍🍕 This insane speed is powered by @cerebras, the latest Inference Provider integrated on @huggingface. Try it for yourself - link in 🧵. GG @andrewdfeldman and team!

Hello from 🤗 HQ!

😍 Announcing @nvidia L40S GPUs in @huggingface Inference Endpoints Fit larger models, get better perfs with FP8! Also available for your Spaces. 🔥 Starts $1.8/hr 🤑 ➡️ huggingface.co/inference-end…

Let's make the new wave of AI robotics open and collaborative! 🦾🤖 Great introduction to @huggingface LeRobot by @VentureBeat - showing how @RemiCadene and team worked with @pollenrobotics to quickly teach their cute robot some new tricks. venturebeat.com/ai/hugging-f…

#WoodstockAI #OpenScience FTW!

Friday already? 🤯 Let's take a look at the 5️⃣ most popular apps this week on @huggingface 🔥 ⬇️🧵

Closing 2️⃣0️⃣2️⃣1️⃣ with a million thanks 🙏 to the 1M+ people who visited huggingface.co this month to discover the latest models, datasets and spaces shared by our community. Can't wait to make the latest ML open and accessible to even more people next year.

Excited to present how we industrialize 🤗 Transformers for production workloads tomorrow 10am PT at EMC2, the 7th Workshop on Energy Efficient Machine Learning. Register free and stream it live on YouTube 🍿emc2-ai.org/virtual-21

🤩 We got the GPUs! We got the Accelerators! 🤩 New H100, L4 and @awscloud Inferentia 2 instances available in @huggingface Inference Endpoints Each Hugging Face user or organization gets quota: - 2x A100 and H100 on GCP - 8x L4 on AWS And you can deploy @AIatMeta Llama 3 on dedicated AWS Inferentia 2 instances today. Build your own AI! Get started at ui.endpoints.huggingface.co/…

Deploy Mistral 7B today in 1 click with 🤗 Inference Endpoints 🔥🔥🔥 Mistral 7B by @MistralAI apparently outperforms all open LLMs up to 20B... but why take anyone's word for it when you can deploy it for yourself in 1 click for ~$1 per hour? Deploy 🔗 ui.endpoints.huggingface.co/…

Find a Dataset you like? 🤩 With a few clicks, AutoNLP turns it into SOTA Models! 😍 💡 New in the @huggingface Hub: start training AutoNLP models from a dataset page.

This is real time, no speedup 🤯🤯🤯 Welcome @GroqInc to @huggingface 🔥🔥🚀 Agents gonna go brrr 🏎️💨💨💨

Visiting our new 🤗 friends at @salesforce - thanks @Khushpatel919 !

Excited to talk about @BigCodeProject StarCoder and Generative AI for the Enterprise here at #Know23 - thanks @ServiceNow for inviting me!

👏 I'm proud of the work of our team at @huggingface with @Qualcomm to enable state of the art AI in the palms of billions with #Snapdragon8Gen1 🤖📲 🍿 Watch our demo of a smart notifications UX at the #TechSummit2021 keynote piped.video/watch?v=3H6tfcZL…

[Cmd]+[Shift]+[Enter] ➡️HuggingChat on Mac Game Changer! Kudos @cyrilzakka at @huggingface for building my favorite Mac app! Download link in thread below.

Last night #WoodstockAI was the largest AI meetup ever Thanks @huggingface community for showing up, and immense gratitude to volunteers, we couldn’t have done it without you! Dave Martin, Jerome, Howard, Alex, and so many more community heroes 🙏❤️🧡💛💚💙💜

🏜️🤗

Proud to connect the global community of AI researchers with the global network of @nvidia GPUs with Training Cluster as a Service announced today at #GTCParis! 🤗🌎 Thanks @AlexisBjorlin for building DGX Cloud Lepton, and thanks Jensen for the shoutout!

Welcome @Cloudflare to the 🤗 Hub!

It's Getting Hot In Here 🤗🔥🔥🔥 The @huggingface Heatmap by @calebfahlgren - see which orgs released the most models over the last year! Looks like @nvidia and @Alibaba_Qwen are quickly ramping up 🔝

Thanks @awscloud for the invite to talk about building Generative AI apps with open source! The AWS Startup Loft in SF is a super cool space. 🔥🔥

Just in time for @AMD Advancing AI event, @huggingface did the numbers on the latest gen of AMD EPYC CPU announced today... ...and we see a 2X speedup on all the LLM use cases we tested 🔥🚀 Check out the blog post to learn more: huggingface.co/blog/huggingf…

About to go on stage at #VMwareExplore with @cswolf to announce and demo SafeCoder, our new enterprise code assistant solution! Read our post for a sneak peak: huggingface.co/blog/safecode…

📦 @cohere just dropped Command A Vision on @huggingface 😍 Designed for enterprise multimodal use cases: interpreting product manuals, analyzing photos, asking about charts... ❓👀🧠 A 112B dense vision-language model with SOTA performance - check out the benchmark metrics in the blog 📊👀

This Figma story shows why building AI with closed model APIs is so risky - you don't control your customers experience. What happened: 1️⃣ Figma shipped a beta "Make Designs" AI feature built on closed model APIs (GPT-4o + AWS Titan) + prompt engineering 2️⃣ When users reported Make Design would generate Apple design lookalikes 😱, Figma couldn't answer why, or fix it 3️⃣ Figma had to pull the feature, now is building their own AI feature with open models and open source (without training on customer data!) Read the whole story by @jaypeters at The @verge theverge.com/2024/7/2/241908… I love Figma! I made the image for this post in Figma. I know they will build amazing AI features and I'm looking forward to using them. But the lesson for us all is that if the AI features are important to you, you can't outsource the AI, you need to do the work. 💪

Great news for enterprises: @cohere models are available for inference on @huggingface starting today! 🔥🔥 Cohere models are designed specifically for Enterprise use cases, and excel at Agents (tool use) and multilingual use cases. 🤖🌎 The 10 Open and Enterprise models are available today for serverless inference in Hugging Face Inference Providers, so you can use them on the Hub and Playground, and build with them with our OpenAI API compatible Python and Javascript SDKs. Learn how in the blog post (in 🧵)

What do Transformers see? 👀 See for yourself with this amazing Vision Transformers self-attention visualization app by @xenovacom 😍 Try it here - huggingface.co/spaces/webml-… - it runs entirely in your browser using transformers.js 🔥 [🖼️] Click the 🐯 image to load your own image [⚙️] App loads locally the DINOv2 w/ Registers model (a @NielsRogge contribution to transformers) [🧭] Explore all the transformer layers and heads to see what features they focus on Give the app a Like if you think it's cool! If you're curious, source code is here: github.com/huggingface/trans…

So excited to attend the #AIHardwareSummit - my first in-person conf since you-know-what - with pockets full of exciting announcements with our hardware partners - stay tuned on @huggingface to find out! #EfficientAI #YayVaccines

Support open science, donate to @arxiv ! 💝 Today, @huggingface triples your donations 💸

We are releasing a course next week! 🔥🔥🔥 Learn Transformers, Hugging Face, at your own pace.

Oh, hello Reachy Mini! 🐣😍 A $299 desktop robot you can program, and power by open source models and apps by the Hugging Face community... Is it the ultimate learning platform for AI robotics? All I know is we'll ship them first come first serve, and orders are piling up fast - smash that order button if you want one! 👇🖲️ pollen-robotics.com/reachy-m… Also, when I left @GoPro 5 years ago to join @huggingface I didn't expect to come back to consumer hardware so soon... but that's how fast we're democratizing AI!

1️⃣5️⃣ things you probably didn't know about Hugging Face! That's a 15min talk I'll give today in SF - and 🚨 SPOILER ALERT 🚨 I have it here for you as a sneak preview 🧵 Want to see it live? Register 👇 eventbrite.com/e/future-tech…

Casual evening with @huggingface 🤗🤗🤗

Hugging Face - It’s Business Time! 🎶🎸

TGIF! Hugging Face community cooked this week - here are the highlights by the amazing @mervenoyann , with LLMs, VLMs, TTS and more from @JinaAI_ @kyutai_labs @BytedanceTalk @UnslothAI and more! 📚 huggingface.co/collections/m… Follow Merve on @huggingface to get her weekly updates!

"Our goal is to expand the universe of AI" Jayesh Govindarajan was asking me how we think about differentiation at @huggingface , at #Dreamforce a couple of weeks ago. Full video in 🧵

Excited to keynote at the #AISF Summit closing a 3-day hackathon where developers were able to experiment with Transformers Agents and Tools with 🤗 Hub credits! 🔥🔥 Thx @fdotinc for the invitation! 🙏

Kicking off NY Tech Week with @awscloud and @Lux_Capital - thx @NickHartmanNY for the great questions! 🤗

🙏 Preach, @Thom_Wolf ! The sermon is a call to action to make AI robots accessible to all, and AI robotics decentralized 🦾🤗 Thx @PrimeIntellect @madisenxtaylor for the invite, and go @LeRobotHF !

As AI races into consumer apps, it's more important than ever to provide access to the technology & tools to understand it, improve it and to build with it in a cost effective way. Our new partnership with @awscloud, announced today, is material to these goals. 4️⃣ whys 🧵👇

Me: Spring break in Japan 🌸🏯🇯🇵 Open source AI 📈🚀 - FineWeb 15T - Llama 3 - Idefics 2 - Mixtral 8x22B - Cohere Command R+ - TGI 2.0 back to Apache 2.0 Excited to be back!

Live from the @huggingface hackathon - 1 hour to go!

Proud of our team helping Rocket Money classify a billion transactions a month with the @huggingface Inference API 👏 "The uptime and confidence we have in the 🤗Inference API has allowed us to focus our energy on the value generated by the models" 😍 huggingface.co/blog/rocketmo…

llama weights is all we need

🏴‍☠️ Pirates, or Robin Hoods? 🏹 $10M of GPU compute for the HF community to build AI in the open Read about it in this great piece from @kyliebytes : theverge.com/2024/5/16/2415 Watch ZeroGPU Spaces going brrr live huggingface.co/spaces/enzost…

📰@huggingface acquired @argilla_io ! What does this mean? Argilla CEO @dvilasuero explains huggingface.co/posts/dvilasu… open data collaboration + open model collaboration = open source AI 🚀 #opensourceFTW

Today at the Humanoids Summit. We are at the dawn of robotic applications to everything. Change is coming. Like with AI, @huggingface will make sure the robotics revolution is decentralized, open source and community driven. Step 1 is to make robotics cheaper and more accessible. Thanks @RemiCadene for driving the costs down and the community forward! Join the movement at @LeRobotHF

vLLM, TensorRT-LLM, Llama.cpp... how do you switch without breaking prod? 🤔 Today, we are announcing multi-backend support in Text Generation Inference a.k.a. TGI - our production inference solution. It provides a unified frontend to build your own AI inference services using transformers, vLLM, TensorRT-LLM, llama.cpp for execution. This means a consistent developer experience whether you want optimized performance fo NVIDIA GPU, AMD GPU, Intel or Arm CPU, AWS Trainium & Inferentia or Google TPU. The @huggingface team is excited to contribute to and collaborate with the teams that build @vllm_project , llama.cpp, TensorRT-LLM, and the teams at @awscloud , @Google , @nvidia , @AMD and @intel to offer a robust and consistent user experience for TGI users whichever backend and hardware they want to use. I hope this new chapter for TGI will simplify production inference for everyone and make the best hardware performance more accessible. Blog post in comments - we are just getting started and will publish more info as we release new TGI backends in the weeks to come. 🚀

Wow! Enterprise organizations on @huggingface can now enable serverless Inference Providers for all their employees! 🔥🔥 Including the latest models from @deepseek_ai @Alibaba_Qwen @Meta @bfl_ml @StabilityAI - check out: - @togethercompute models: huggingface.co/models?infere… - @FAL models: huggingface.co/models?infere… - @novita_labs models: huggingface.co/models?infere… - @cerebras models: huggingface.co/models?infere…

At Next, who’s here?

How to use @togethercompute models with your @huggingface account 👍 PRO accounts get $2 free credits each month 🤑

💪 Small but mighty! Phi-3 mini from @Microsoft is already on Hugging Chat - try it now for free 💬 huggingface.co/chat/models/m… Also supported in TGI (pull main until release)

If you’re at GTC, join my session at 3pm PT to learn about Train on DGX Cloud and more! S63149 Building Accelerated AI with Hugging Face and NVIDIA

3rd year at the AI Hardware Summit - lots of old friends, and many new friends! 🤗🤗🤗

We simplified @huggingface plans and pricing! 🧑‍🍳 People upgrade to PRO for $9/mo 👨‍👩‍👦‍👦 Small orgs upgrade to TEAMS for $20/mo/user via credit card 🏢 Corporations upgrade to ENTERPRISE starting $50/mo/user via sales Why upgrade? To enable AI builders with better collaboration and compute options, and AI building companies with enterprise security, compliance and governance. hf.co/pricing What do you think?

Sentence Transformers v5.0 is out! Introducing Sparse Embeddings models 🔥🔥 Embeddings are the lifeblood of RAG, semantic search, web search agents and more. Sentence Transformers is the reference toolkit to train and run these models, now with 3 new Sparse architectures natively implemented across the library, and compatible with vector databases like @qdrant_engine @elastic , @OpenSearchProj and more.

Thanks SF community for showing up to our Generative AI on Inferentia2 workshop! @awscloud Inf2 went brrrr - together we trained more than 100 models and created hundreds of images! 🔥🔥

The App Store of AI (but, open)

You got AI models, and then you got sock models 🤗

I am so proud of my colleagues @mervenoyann and @alaradirik contributing to a grass roots volunteer efforts started by @acikyazilimagi to locate earthquake survivors in Turkey 👏💜 This post tells the story of the mad dash data science collaboration 👀 huggingface.co/blog/using-ml…

Thanks @StanfordLaw for inviting me to the Stanford AI Symposium to show how companies can build their own AI responsibly with open source!

Today @GoogleDeepMind released the Gemma commercially permissible open LLMs 🔥🔥 Below all the resources you need to start using Gemma with @huggingface without writing any code 🧵⤵️

Want to learn how to deploy LLMs on @AMD GPUs and Ryzen AI PCs? 🎙 Tune in to Hugging Cast this Thursday June 6 - register here: streamyard.com/watch/iMZUvJn… This episode will be all about using @huggingface on AMD, from cloud to laptop: 🚀 How we collaborate with AMD to make open models go brrr everywhere ☁️ Deploy LLMs on Azure with the new ND MI300X v5 VMs and TGI 💻 Unlock on-device inference with optimum-amd on the new Ryzen AI PCs Join @MorganFunto @efxmarty @mohit21sharmams and I and bring your questions!

🤗📣 New feature for @huggingface Enterprise! 📊 Manage Inference Providers usage and costs for your organization - that's 40k models available for serverless inference from @togethercompute @FireworksAI_HQ @replicate @cohere @FAL @cerebras @SambaNovaAI @nebiusai @hyperbolic_labs @novita_labs and more

Can 7 people please star a repo 🤣

I used to airdrop mobile app screenshots to edit on photoshop desktop, now I airdrop desktop screenshots to edit on my phone on @photoroom_app and @mojo_video_app

Today, @huggingface announces smolagents, a new open source library to turn LLMs into agentic systems! 🤗 Simple: no clunky JSON tools with CodeAgent 🤝 Open: use any open or closed LLMs with @LiteLLM 📚 Extensible: push/pull tools from HF Hub Learn more: huggingface.co/blog/smolagen… Install & star it: github.com/huggingface/smola…

▶️🗣️ My #GTC2025 talk is online - link in 🧵 - I showed how AI builders work with open models on @huggingface - and gave my personal tips to keep up with it all, including: ❓ why companies should build their own AI with open models 🧭 how to find the right model among 1.5M on the Hub 🧐 how to make the most out of model cards 💼 how to collaborate privately with organizations 🔒 how to secure organizations with Enterprise Hub 🛠️ how to prototype with serverless Inference Providers 🏭 how to build for production with Inference Endpoints 🤝 how to deploy on @awscloud @Azure or @DellTech 📲 how Spaces is the new AI App Store 🔬 how to keep up with AI research with Daily Papers by @_akhaliq 📈 how to keep up with models with Trending repos (and following @mervenoyann ) 🏪 how to keep up with model providers by following organizations like @nvidia 💬 how to keep up with the field with Posts - the AI Newsfeed 👷 how to build your own LLMs from scratch 📊 how to make GPU Clusters go brrr with Ultra-Scale Playbook 📋 how to learn distilling models with Open-R1 🥷 how to build your own agents with smolagents 🧰 how to find LLMs and tools for your agents on the Hub 🎙️ how to build Voice Agents with FastRTC 💻 how to build Browser Agents with smolagents 🔎 how to build Research Agents with Open DeepResearch 🧑‍🎓 how to learn to make agents with the free Agents Course (and the new Reasoning Course!) I don't have the timestamps but that's roughly n order. Sorry the audio is terrible ; there should be a model & Space for that on HF if someone at NVIDIA wants to fix it 🤣

Jensen Huang just introduced @nvidia Project Digits, a game changing AI PC to run LLMs locally! With 128GB of unified memory and a stackable design, no model will be too big to run at home. $3k a pop, avail later this year!