Ziteng Gao

[Ziteng Gao @ziteng_v]

时来天地皆同力，运去英雄不自由😶

[Jiaming Song @baaadas]

Over the weekend, I was using codex to update my homepage and a paper I wrote a year ago on the topic of diffusion LLMs (should be updated on Monday).

https://t.co/qvqldZ9H1w

While I did not want to make it too explicit back then, I have argued that discrete diffusion LLMs were not the right thing to do and if diffusion ever works on LLMs continuous dLLMs are the way to go.

A year later, we are seeing a lot cool papers in this space, and I hope the community can push for something practical and scalable.

[Ziteng Gao @ziteng_v]

@_machc @kiamada That's really a cool schedule for pixel diffusion!

[Ziteng Gao @ziteng_v]

@HanshengCh Cool work, Hansheng! Have you ever tried some network-defined subspace? Btw, thanks for discussing with PixNerd!

[Ziteng Gao @ziteng_v]

@JiaweiYang118 Great work, Jiawei! Few steps + pixel generation is really a thing! Thanks for discussing with PixNerd!

[Ziteng Gao @ziteng_v]

@LodestoneRock is it about massive activations?

[Ziteng Gao @ziteng_v]

😢just a note to every researcher on pixel diffusion that PixNerd is the FIRST large patch diffusion transformer on pixels without vaes.. please do not ignore and give the proper credit to it

[Ziteng Gao @ziteng_v]

@iScienceLuvr this reminds me of RIN……

[Ziteng Gao @ziteng_v]

@LodestoneRock @isskoro Very cooool idea! Converting LDMs to pixel models is definitely a next big thing!!

[Ziteng Gao @ziteng_v]

The main idea behind D-AR is to **bridge** AR and diffusion via the proposed visual tokenizer. Therefore, we can perform diffusion on image space by **the simple next token prediction**!

[Ziteng Gao @ziteng_v]

❓What do you think about image diffusion process "simulated" by an autoregressive LLM?

+ 🎨 Diffusion quality
+ ⚡️ LLM-style inference with KV caching
+ 🔢 Discrete tokens in & discrete tokens out

That's D-AR! Diffusion via Autoregressive Models
Code: https://t.co/WEN8J5DLET

[Ziteng Gao @ziteng_v]

D-AR can perform layout control with different granularities by conditioning prefix AR tokens

[Ziteng Gao @ziteng_v]

❓What do you think about image diffusion process "simulated" by an autoregressive LLM?

+ 🎨 Diffusion quality
+ ⚡️ LLM-style inference with KV caching
+ 🔢 Discrete tokens in & discrete tokens out

That's D-AR! Diffusion via Autoregressive Models
Code: https://t.co/WEN8J5DLET

[Ziteng Gao @ziteng_v]

D-AR naturally enables the coarse-to-fine ordering for image generation, thanks to diffusion properties!

[Ziteng Gao @ziteng_v]

❓What do you think about image diffusion process "simulated" by an autoregressive LLM?

+ 🎨 Diffusion quality
+ ⚡️ LLM-style inference with KV caching
+ 🔢 Discrete tokens in & discrete tokens out

That's D-AR! Diffusion via Autoregressive Models
Code: https://t.co/WEN8J5DLET

[Ziteng Gao @ziteng_v]

❓What do you think about image diffusion process "simulated" by an autoregressive LLM?

+ 🎨 Diffusion quality
+ ⚡️ LLM-style inference with KV caching
+ 🔢 Discrete tokens in & discrete tokens out

That's D-AR! Diffusion via Autoregressive Models
Code: https://t.co/WEN8J5DLET

[Ziteng Gao @ziteng_v]

I’ll present my work SparseFormer 10:45 this morning on #ICLR2024. Welcome to my poster #12!!😄👏👏

[Adina Yakup @AdinaYakup]

Near 500 paper claimed in Space ICLR 2024🎉
https://t.co/vTDeyBkXfT
Here are some tips if you want to engage more with the community 🤗
💡 Join the discussion threads below each paper.
💡 Start a conversation with authors by clicking "@" next to the author's profile photo on their @huggingface account.
💡 Link your models, datasets, and demos to your papers.
#ICLR2024 @_akhaliq

[Zechen Bai @ZechenBai]

🚀 Our latest survey paper is now released, presenting a comprehensive analysis of hallucination phenomena in multimodal large language models (MLLMs), also known as Large Vision-Language Models (LVLMs).
arXiv: https://t.co/8L4FC1QHvT
GitHub: https://t.co/YkLJl7XtgY

[Ziteng Gao @ziteng_v]

🤔Is SparseFormer "yet anthoer" way to handle any resolution and aspect ratio besides FlexiViT or NaViT?

[Ziteng Gao @ziteng_v]

😁SparseFormer finally got accepted at ICLR 2024 @iclr_conf !! Big thanks to all co-authors!!
⚡️A sparse visual architecture that can understand an image with down to only 9 tokens in the latent space.
🤖️Code for images is released and please give a try! @_akhaliq

[YUCHAO GU @YuchaoGu]

Thank @_akhaliq previous featured our work. Videoswap is now accepted by CVPR 2024. The code is now available at https://t.co/hwwOhSHVsA. Welcome to try！

[Jay Wu @jayzhangjiewu]

Thanks @_akhaliq for sharing our work!

TL;DR: We introduce T2VScore to automatically evaluate text-conditioned generated videos.

Project page: https://t.co/DJYjLsumt8