Hongcan Guo

Numquam ponenda est pluralitas sine necessitate

[Yian Zhao @zhao_yian94154]

Training a video world model from scratch shouldn't need a cluster.

MiniWorld is a minimal, reproducible recipe for action- and pose-conditioned streaming world models — trained end-to-end in a few days on one 8-GPU server.

253-frame rollouts from a single frame 👇

[Hongcan Guo @HongcanGuo]

@taskinfatih thanks for your attention

[Hongcan Guo @HongcanGuo]

Cola DLM is now open-sourced, including model weights and code!🎉

Thanks for the community support — we look forward to seeing your explorations and follow-up works.

Code: https://t.co/hR9q4XfcyD
Model: https://t.co/gbD60uZ1Kk

[Hongcan Guo @HongcanGuo]

@giladturok Thanks🥹

[Hongcan Guo @HongcanGuo]

@yechan_ai Thanks again for your amazing repo! It definitely couldn't have been easy to do such a great job in such a short amount of time. Also, the good news is that our official repo is now open-source. Feel free to check it out and use it here: https://t.co/hR9q4XfcyD

[Hongcan Guo @HongcanGuo]

@scaling01 Thanks to everyone! We want your comment about Cola DLM. It will help us to improve it in the next generation.

[Hongcan Guo @HongcanGuo]

@HuggingPapers Thanks! We truly look forward to seeing your creative explorations and follow up works😍

[Hongcan Guo @HongcanGuo]

@rasbt @yechan_ai we will open weights this weekend. Looking forward to your support for our DLM-related work.😍

[Hongcan Guo @HongcanGuo]

Glad to share that we have updated some thoughts and analysis on Cola DLM in our technical blog, making it easier for you to quickly grasp our core ideas. 💡

https://t.co/5e6P93Ofhs

#diffusion #coladlm #llm

[Hongcan Guo @HongcanGuo]

Gotta love the open-source community! ❤️ Ridiculously fast reproduction. Official release coming next week—stay tuned! 👀

[Yechan Do @doyc_1]

https://t.co/rxiNHTzg1T

Thanks for great research @HongcanGuo and ByteDance team.

[Hongcan Guo @HongcanGuo]

@yechan_ai Thank you for the interest! Really impressed by your speed in reproducing the code haha We're trying our best to drop our code and weights next week, and we'd totally welcome your future improvements. Just wanted to add: yours is a fantastic codebase. Stellar work doing this so⚡️

[Hongcan Guo @HongcanGuo]

@GeZhang86038849 Thank you very much for your interest. Your work on Dynamic Large Concept Models is truly outstanding, and I have learned a great deal from it!

[Hongcan Guo @HongcanGuo]

@rugbist_ @_akhaliq Thank you for your interest! You are also welcome to check out some of my thoughts, and I would greatly appreciate any feedback or suggestions you might have!

[Hongcan Guo @HongcanGuo]

A brand-new approach to text modeling

[Hongcan Guo @HongcanGuo]

Thanks again for the discussion! To reiterate our main point: VAE and diffusion can be replaced by any representation and matching methods. Our true focus is on achieving more efficient text representation. We welcome your feedback and critiques!

[Hongcan Guo @HongcanGuo]

A brand-new approach to text modeling

[Hongcan Guo @HongcanGuo]

@_akhaliq Thank you so much for spreading the word! We welcome everyone to join the discussion.🎉👍

[Hongcan Guo @HongcanGuo]

@YinjieW2024 thanks🥹

[Jiayi Weng @Trinkle23897]

Codex grew programmatic policies with no neural nets: max score on Breakout, and SOTA-level scores on MuJoCo.

Maybe heuristics were not too weak. Maybe they were just too expensive to maintain. Maybe it's the next paradigm.

https://t.co/1ZaIneleuW

[Hongcan Guo @HongcanGuo]

@festive_helix I plan to join Professor Hengshuang Zhao's research group at the University of Hong Kong to pursue a PhD in September this year, and will continue my internship with the seedance team. Welcome to follow my subsequent work.😊

[Hongcan Guo @HongcanGuo]

@JialeZhao5 Thank you very much for your recognition!