Xiaomi trained a robot foundation model on 100K hours of real-world data. Success rates scale predictably with data and model size, no saturation yet. The LLM scaling laws playbook is becoming a clear trend across recent robotics work.
https://t.co/8wWa2wqkdH
Turns out learnings from low data scale and overfit scenarios don’t translate to the high data regime.
Last year we relied on inductive biases and extremely strict data collection procedures. The result was a capable but brittle policy.
This year we embraced chaos and allowed our dataset to span the real world state distribution. The result was a set of capabilities we would never have predicted would emerge.
A world model trained on a sanitized world is a contradiction. My deep learning career taught me: at low data, curate carefully. At scale, stop protecting models from reality. That is where unexpected capabilities emerge. https://t.co/1fO6faqNZV
31 yıl önce, insanlık tarihine kara bir leke olarak geçen Srebrenitsa Soykırımında şehit edilen Boşnak kardeşlerimizi rahmetle anıyoruz. Bu soykırımı unutmadık, unutturmayacağız.
Još jednom oštro osuđujemo genocid u Srebrenici na 31. godišnjici obilježavanja. S rahmetom spominjemo šehide, dijelimo bol njihovih porodica. Da se nikada ne ponovi i ne zaboravi!
@farazdotai@nvidia Wow! I was an early beta tester of TensorRT-LLM through a partnership, even before its GitHub release. It took my large multi-modal transformer from 25 to 130+ tokens/sec on A10s and 300+ with quantization. Happy to see it still going strong and hope it reclaims its market share.
If you're trying out FA4, you're likely to run into not being able to load cutlass.cute - the issue is that nvidia-cutlass-dsl wheel installs its python files into - wait what - site-packages/nvidia_cutlass_dsl/python_packages/cutlass
so of course python can't find these packages - so to overcome I had to use
export PYTHONPATH=$CONDA_PREFIX/lib/python3.12/site-packages/nvidia_cutlass_dsl/python_packages/
now I can load FA4.
to be honest I think Russia did give world a very clear message; if you disrespect me we will have a fight and I won't step back, I will be stubborn because I have resources to sustain.
So many countries depend on USA Ukraine was one of them and my heartfelt sympathy for them 💐
Sad to see what has been coming from Palestine; war is bad! but a good anology for this situation could be going out for dinner to a restaurant, you can eat as much as you can and you don't have to get anyone's permission, it's your free will. But, you gotta pay the bill later.
The new plan is apparently to make as many people as possible as miserable as they can be made without blowing them up. And this will be pursued with the same relentless ingenuity as blowing them up was.