Inside a computer
Joined December 2022
- Tweets967
- Following217
- Followers3.2K
- Likes46.9K
Kunvar Thaman retweeted
1st place with 1.77x speedup over SGLang on the prefill layer leaderboard 🚀🚀
We're launching a $100,000 challenge to see who can make Muse Spark run fastest: openkernelchallenge.dev/
Kunvar Thaman retweeted
We’ve spent the past year developing new primitives for GPUs. We believe this enables new algorithms to change how machines learn, open new markets, and dramatically reshape AI’s cost structure.
We have raised a $1 billion Series A, co-led by Index and Lux.
Kunvar Thaman retweeted
Good morning friends. Flappy's quest for extreme learning efficiency has led us to build the 'Kittens Virtual Machine' (KVM) framework, which launches a megakernel that acts as a persistent runtime and exposes its own instruction-based programming model.
The KVM (a) provides unprecedented control and visibility over the hardware (something we've been extensively utilizing to accelerate our internal research), exposing a completely new kind of systems–training co-design space, (b) reaches vastly SoTA speeds without hand-tuned kernels, and (c) tremendously increases GPU utilization by co-packing multiple independent workloads into the same kernel.
Once upon a time, some unreasonably talented people decided that machines could learn a lot more from a lot less...
Kunvar Thaman retweeted
We recently published the eight-week process behind scaling our UMI dataset to 1M+ tasks and 45k+ new tasks a day.
Now, we’re open-sourcing DiverseUMI-100, a collection of 12,000+ tasks from a 100-hour sample of our dataset, complete with dense annotations by Argus, our advanced robotics annotation pipeline.
How do you build the most diverse UMI dataset in robotics?
In the last 8 weeks, we scaled an international UMI data collection operation from 5 to 90 operators, built out a dataset of over 1M+ unique tasks, and continue to capture over 45k+ new tasks each day.
Robotic World Models require much higher diversity of tasks than imitation learning. Here's the why and how behind our operations. 🧵
robotics is still underhyped the way llms were underhyped in 2023
Kunvar Thaman retweeted
pantheon back at it
How do you build the most diverse UMI dataset in robotics?
In the last 8 weeks, we scaled an international UMI data collection operation from 5 to 90 operators, built out a dataset of over 1M+ unique tasks, and continue to capture over 45k+ new tasks each day.
Robotic World Models require much higher diversity of tasks than imitation learning. Here's the why and how behind our operations. 🧵
UMI sits at the pareto frontier of cost/utility, if done right.
@PantheonInc's managed to make scaling complex manufacturing data a sub linear function of capital.
Soon we'll see 1 million hour UMI datasets (and some insanely good world models built on top of them)
How do you build the most diverse UMI dataset in robotics?
In the last 8 weeks, we scaled an international UMI data collection operation from 5 to 90 operators, built out a dataset of over 1M+ unique tasks, and continue to capture over 45k+ new tasks each day.
Robotic World Models require much higher diversity of tasks than imitation learning. Here's the why and how behind our operations. 🧵
Kunvar Thaman retweeted
this is incredibly scaling-pilled.
pantheon is ratcheting a machine that can turn human hand time into high quality robot data increasingly sharply each week, and taking audacious bets on the ability of future models to be superhuman labelers for increasingly freeform data!
How do you build the most diverse UMI dataset in robotics?
In the last 8 weeks, we scaled an international UMI data collection operation from 5 to 90 operators, built out a dataset of over 1M+ unique tasks, and continue to capture over 45k+ new tasks each day.
Robotic World Models require much higher diversity of tasks than imitation learning. Here's the why and how behind our operations. 🧵
Kunvar Thaman retweeted
We're taking a different approach to robot foundation models, which has different requirements.
We're growing one of the most diverse robotics datasets out there, with one of the most talented teams running it behind-the-scenes.
How do you build the most diverse UMI dataset in robotics?
In the last 8 weeks, we scaled an international UMI data collection operation from 5 to 90 operators, built out a dataset of over 1M+ unique tasks, and continue to capture over 45k+ new tasks each day.
Robotic World Models require much higher diversity of tasks than imitation learning. Here's the why and how behind our operations. 🧵
Kunvar Thaman retweeted
How do you build the most diverse UMI dataset in robotics?
In the last 8 weeks, we scaled an international UMI data collection operation from 5 to 90 operators, built out a dataset of over 1M+ unique tasks, and continue to capture over 45k+ new tasks each day.
Robotic World Models require much higher diversity of tasks than imitation learning. Here's the why and how behind our operations. 🧵
Kunvar Thaman retweeted
There is a relatively neglected opportunity to understand and train models to have better behavior, potentially optimizing less for frontier capability. This “good behavior” encapsulates many traits - truthfulness, the ability to understand/adapt to a user in various ways, not being overly reward-seeking, etc.
Kunvar Thaman retweeted
Argus: an open-source robotics data annotation and quality pipeline.
With new frontier VLMs like GPT-6 Astra, we can generate rich, high-quality annotations for robotics data for both training and dataset analysis. Argus delivers detailed, timestamp-level annotations while catching issues like mislabeled instructions, sped-up recordings, swapped camera streams, and unflagged operator mistakes.
To support better data quality for the robotics community, we’re open-sourcing Argus for anyone to use 🧵
Argus increases the effective value of a training sample 20-30x.
We're moving from robotics as a field being data constrained to one which is compute constrained
Argus: an open-source robotics data annotation and quality pipeline.
With new frontier VLMs like GPT-6 Astra, we can generate rich, high-quality annotations for robotics data for both training and dataset analysis. Argus delivers detailed, timestamp-level annotations while catching issues like mislabeled instructions, sped-up recordings, swapped camera streams, and unflagged operator mistakes.
To support better data quality for the robotics community, we’re open-sourcing Argus for anyone to use 🧵
Kunvar Thaman retweeted
Read the full analysis at pantheon.inc/research/argus, and run Argus on your data at data.pantheon.inc/. For collaborations or questions, contact us at [email protected].
Credit to @ericli_ for undertaking this project.
