AI Hardware Acceleration @huggingface Working on democratizing good machine learning without devouring resources. 🤗
Paris, France
Joined November 2021
- Tweets91
- Following252
- Followers1.1K
- Likes296
Jingya Huang retweeted
Reachy Mini has wheels now!
Our little open-source AI robot can finally explore beyond the desktop.
And that's not all:
📱 New mobile app: no computer needed
👑 NFC hats that unlock new personalities and apps
Physical AI just got a lot more fun.
Watch it👇
🤗
More reference control. Fewer denoising steps. More places to run H3.
The latest community contributions span 8-step reference-conditioned video + audio, 4-step distillation, sparse attention, and reference-token compression - alongside serving and workflow integrations across GPU servers, TPUs, and Apple Silicon.
What matters is what this enables: more affordable experimentation, finer creative control, and more choices about where and how video generation runs. Open models become more useful when the community can improve the methods, adapt the runtime, and build workflows around them.
Credit to @haoailab @yisun0618 @lanfeng491270 @guocheng_qian @Alissonerdx @ddalcu @tazmannner @HaochengXiUCB @huggingface @Jhuaplin, and the PTD, DMAD, SparkH3, TorchTPU and FlashML teams - including Zhengming Yu and zechengtang on GitHub.
11 additions to the H3 ecosystem index. Thank you to everyone building, testing, and sharing.
github.com/MiniMax-AI/awesom…
With native PyTorch support for AWS Trainium and Google Cloud TPUs, we’re bringing support for these accelerators to Diffusers and more libraries across the Hugging Face ecosystem. Stay tuned and help us build! 🚀
We secretly worked on bringing TorchTPU support to Diffusers with @Jhuaplin 🔥
It needed us to first ship TP support so that large models could fit in TPU chips. Then a bunch of small implementation-level changes.
Check out the post to know more:
huggingface.co/blog/hf-torch…
Jingya Huang retweeted
Tensor parallel loading in 🤗 Diffusers got a massive upgrade 🔥
On Flux.2-Dev DiT, with a TP degree of 4 (A10G):
• 30.4s → 12.5s loading
• 64.1 → 6.8 GB peak CPU memory/rank
~2.4× faster loading. ~89% less CPU memory.
Check it out:
huggingface.co/docs/diffuser…
Jingya Huang retweeted
BIG ANNOUNCEMENT FROM HUGGING FACE TODAY:
We're unveiling Microduck 🐥🤖
It's a tiny $399 open-source robot you can teach new tricks with reinforcement learning. It can walk, pick things up, get back up when it falls, and even roller-skate.
Welcome to the era of open-source affordable robots to democratize physical AI and world models!
🤗🤗🤗
Jingya Huang retweeted
Mitigating racial bias from LLMs is a lot easier than removing it from humans!
Can’t believe this happened at the best AI conference @NeurIPSConf
We have ethical reviews for authors, but missed it for invited speakers? 😡
Jingya Huang retweeted
BREAKING NEWS
The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”
👋I will present the 🤗@huggingface Optimum Neuron library at the AWS summit Zurich on September 4th.
🚀Optimum Neuron is a tool for easy training and inference on powerful AWS Neuron AI accelerator. Hope to see you next week, and if you haven't registered, 🔗 in the thread!
Jingya Huang retweeted
GenAI insights are set for #VivaTech with @Thom_Wolf, Co-founder & Chief Science Officer of @huggingface!
Don't miss his session on the future of large language models & where Generative AI will take us in the next decade 🚀
On stage this May.
Jingya Huang retweeted
The LLM-Perf Leaderboard 🏋️ is back with new features and benchmarks🔥
🆕New Stuff🆕
-Quantizations: AWQ with GEMM & GEMV kernels 🗜️
-Optimization: FlashAttentionV2 🚀
-Models: Th rise of MoEs 🤼
-DType: BFloat16 🤔
-Plots 📊
Space: huggingface.co/spaces/optimu…
🔽Key Takeaways🔽
Jingya Huang retweeted
AMD GPU support in 🤗 Hugging Face libraries just got vastly improved with Transformers 4.36 and TGI 1.2 releases! A 🧵 about the integration & how AMD ROCm software stack enables a seamless dev and user experience. huggingface.co/blog/huggingf… (1/13)
Jingya Huang retweeted
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
🧵 👇
🫤
Hey @Adweek, it's 2023 and you're still playing into sexist stereotypes of women having to "juggle" motherhood and a career? Yikes. @SashaMTL is one of the most brilliant AI scientists out there. Do better.
ALT AdWeek sexist headline of AI researcher Sasha Luccioni: "This AI ethics expert juggles motherhood and a tech career"
PRECIOUS AS GOLD 🥇
Ready for the *biggest update* of the Open LLM Leaderboard yet?
We just spent A YEAR of GPU time to make it more interesting and fairer! 🤯
How? With @nathanhabib1011, we added 3 new evals from the great EleutherAI harness 💥 and re-ran 2000+ models! 🚀
So, what changes? 🧵
Have a look and try!
Struggling to Access GPUs to deploy Generative AI models? Or want to optimize cost performance? We've got something for you! 🤗 We are happy to share support for Stable Diffusion XL on @awscloud Inferentia2 and Amazon SageMaker. 🧠
👉 philschmid.de/inferentia2-st…
🧶
Jingya Huang retweeted
Today 01.ai released their open source model Yi-34B on the Hub👏🚀
✨ The FIRST Chinese model to top the Open LLM Leaderboard
💪 Better performance than Falcon-180B and Llama2 70B on pre-training
🇨🇳 Supports both English and Chinese
huggingface.co/01-ai/Yi-34B