David W. Romero
@davidwromero
Followers
2K
Following
1K
Media
61
Statuses
638
Research Scientist @NVIDIA. PhD in Efficient Deep Learning @VUAmsterdam. Prev: @GoogleAI, @Qualcomm, @Merl_news. Opinions my own.
San Francisco, CA
Joined October 2019
RT @krandiash: I’ll be at ICML this week, reach out if you’d like to chat about. 👉 research at @cartesia_ai.👉 alternate architectures and t….
0
1
0
RT @_albertgu: Tokenization is just a special case of "chunking" - building low-level data into high-level abstractions - which is in turn….
0
172
0
HMAR models and code are finally public! 🥳.
Happy to share that our HMAR code and pre-trained models are now publicly available. Please try them out here:. code: checkpoints:
0
1
4
RT @_albertgu: I converted one of my favorite talks I've given over the past year into a blog post. "On the Tradeoffs of SSMs and Transfor….
0
112
0
RT @mli0603: Cosmos-Reason1 has exciting updates 💡.Now it understands physical reality — judging videos as real or fake! Check out the reso….
0
32
0
RT @krandiash: Today we shipped a new real time API for streaming speech to text (a new family of models called Ink), that’s extremely fast….
0
13
0
We really liked VAR's formulation towards image generation. During @KumbongHermann's internship, we noticed there were a few aspects to improve. The result: A better & faster multi-scale autoregressive image generation framework. Come to our poster at #CVPR2025 this week! 🥳.
Excited to be presenting our new work–HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation– at #CVPR2025 this week. VAR (Visual Autoregressive Modelling) introduced a very nice way to formulate autoregressive image generation as a next-scale prediction task (from
0
1
10
RT @ratankaliani: Over the last 2 weeks, I took a deep dive into Evo 2, Arc's Genomic Foundation model. But, I couldn't find a crisp primer….
0
40
0
Long convs go geometric!. "Our model processes geometric context of 30k tokens 20× faster than equivariant transformer and allows 72× longer context with the same budget." . There simply isn't anything better than long convs for fast long context on multidim data. Beautiful! 😍.
ICML Spotlight 🚨 Equivariance is too slow and expensive, especially when you need global context. It makes us wonder if it even worths the cost, especially in high-dimensional problems? We present Geometric Hyena Networks — a simple equivariant model orders of magnitude more
0
1
20
RT @ArnabMondal96: Check out our work on RL finetuning for SVG generation by one and only @joanrod_ai.
0
1
0
RT @realDanFu: An entire model. in a single kernel!. The H100 number is crazy - at 1000 toks/s on 1xH100, the Llama-1B is running at 72%….
0
7
0
RT @_onionesque: Finally a comprehensive paper on neural networks with non-linear equivariant maps
0
21
0
RT @mli0603: To help with the development of physical #EmbodiedAI, enhance how #Robot and #AutonomousVehicle understand the physical world,….
0
3
0
We are happy to introduce Cosmos-Reason1, our reasoning model specifically trained for Physical AI. Really proud of being part of this immense effort 💪.
DeepSeek R1 demonstrates AI mastering math through reinforcement learning. We introduce Cosmos-Reason1, which learns with Physical AI rewards and excels in physical AI tasks.
2
5
32
RT @pdhsu: Awesome to see our Evo 2 model highlighted in Jensen's keynote at GTC. @arcinstitute 🤝 @nvidia
0
11
0
It was a true pleasure to be part of this amazing collaboration! Super happy to see Evo2 out there! 😍. Interested in how we trained a 40B conv-attn Hybrid model with a 1M context length? Check out our technical report -- see @MichaelPoli6’s thread here:
AI provides a universal framework that leverages data and compute at scale to uncover higher-order patterns. Today, @arcinstitute in collaboration with @nvidia releases Evo 2—a fully open source biological foundation model trained on genomes spanning the entire tree of life 🧵
3
3
31
RT @artemmoskalev: AI/ML Internships in Drug Discovery🚨. Our team is hiring PhD research interns for summer 2025 in the US. Come work with….
0
19
0
Our team is actively recruiting at all levels. Come join us in this exciting journey! 🚀.
Our team is actively recruiting at various seniority levels. We’re looking for candidates with deep expertise in video generative models, LLMs, VLMs, large-scale model training, or data processing. Join us in shaping the next generation of Cosmos models for Physical AI!.
0
7
72
I am very happy to be involved in this collaboration. Stay tuned to hear about some cool updates soon! 🔥.
Today, Arc announced it is joining forces with @NVIDIAHealth to develop powerful computational tools that will help researchers everywhere explore and understand living systems in ways previously impossible.
0
0
16