Bram Wallace retweeted
GPT-5.2 derived a new result in theoretical physics.
We’re releasing the result in a preprint with researchers from @the_IAS, @VanderbiltU, @Cambridge_Uni, and @Harvard. It shows that a gluon interaction many physicists expected would not occur can arise under specific conditions.
openai.com/index/new-result-…
Banger work by our imagegen team. Cool to have been involved to a small extent, I learn a lot from these folks
Bram Wallace retweeted
📢 Another #NeurIPS, another diffusion circle!
Join us to talk about diffusion models on Friday Dec 5 at 3:30PM in San Diego! Bayside terrace outside room 11 (upstairs) ☀️🚢🌊
Please help spread the word, tell your friends! No slides, no talks, we just sit down and chat 🗣️
Bram Wallace retweeted
That's a wrap for #ICLR2025! See you all next year in Brazil! Please all welcome @BharathHarihar3 as the new Senior Program Chair! (With @cvondrick continuing on as General Chair.)
Bram Wallace retweeted
GPT-4.5 is ready!
good news: it is the first model that feels like talking to a thoughtful person to me. i have had several moments where i've sat back in my chair and been astonished at getting actually good advice from an AI.
bad news: it is a giant, expensive model. we really wanted to launch it to plus and pro at the same time, but we've been growing a lot and are out of GPUs. we will add tens of thousands of GPUs next week and roll it out to the plus tier then. (hundreds of thousands coming soon, and i'm pretty sure y'all will use every one we can rack up.)
this isn't how we want to operate, but it's hard to perfectly predict growth surges that lead to GPU shortages.
a heads up: this isn’t a reasoning model and won’t crush benchmarks. it’s a different kind of intelligence and there’s a magic to it i haven’t felt before. really excited for people to try it!
Bram Wallace retweeted
sora.com signups are fully open
Bram Wallace retweeted
aditya (@model_mechanic) is a legend and visionary in the field, and runs a very special team.
openai.com/index/sora-is-her…
Stoked to finally be able to say what I've been working on this year! There's so much exciting post-training research ahead of us, we're just getting started over here
Today’s deployment is only possible with our new model, Sora Turbo. Huge shoutout to @bram_wallace and @JureZbontar who spearheaded Sora model acceleration research!
Bram Wallace retweeted
sora is launching today to all chatgpt pro and plus users!
it's been a big effort to make this possible + i think the product is really fun & intuitive.
my fav thing to do is generate fake historical found footage. video inpainting is also really strong.
have fun!
Bram Wallace retweeted
Today's the day!
sora.com
Included with ChatGPT Plus/Pro!
Bram Wallace retweeted
here's a clip of video set to my new song "sugarize" - every frame created from clips i made using Sora, by @OpenAI
Bram Wallace retweeted
Made with Sora. The Golden Record - from raw earth material to a time capsule of human life on Earth. Using 11 different generations cut together from Sora, I was able to explore what the odyssey of this record might look like. @OpenAI
Great work! Love the integration of dataset + self learning and the results look sweet. I think this style of method has tons of potential
Self-Play Fine-Tuning of Diffusion Models for Text-to-Image Generation
paper page: huggingface.co/papers/2402.1…
Fine-tuning Diffusion Models remains an underexplored frontier in generative artificial intelligence (GenAI), especially when compared with the remarkable progress made in fine-tuning Large Language Models (LLMs). While cutting-edge diffusion models such as Stable Diffusion (SD) and SDXL rely on supervised fine-tuning, their performance inevitably plateaus after seeing a certain volume of data. Recently, reinforcement learning (RL) has been employed to fine-tune diffusion models with human preference data, but it requires at least two images ("winner" and "loser" images) for each text prompt. In this paper, we introduce an innovative technique called self-play fine-tuning for diffusion models (SPIN-Diffusion), where the diffusion model engages in competition with its earlier versions, facilitating an iterative self-improvement process. Our approach offers an alternative to conventional supervised fine-tuning and RL strategies, significantly improving both model performance and alignment. Our experiments on the Pick-a-Pic dataset reveal that SPIN-Diffusion outperforms the existing supervised fine-tuning method in aspects of human preference alignment and visual appeal right from its first iteration. By the second iteration, it exceeds the performance of RLHF-based methods across all metrics, achieving these results with less data.
Our research codebase for Diffusion-DPO is now public!
github.com/SalesforceAIResea…
This is the exact (cleaned up) code we used for the primary results in our paper. It also has an SDXL-Turbo script which works well (see xitter.gq/bram_wallace/status/17…).
Enjoy!
Bram Wallace retweeted
Our blog for Diffusion-DPO is now live!🚀
In this project we brought the benefits of Reinforcement Learning from Human Feedback (RLHF) to text-to-image diffusion models at scale for the first time.
blog.salesforceairesearch.co…
Turboing into 2024! Why let SDXL-Base have all the fun? DPO-tuning 4-step SDXL-Turbo turns out to work pretty well. Original SDXL-Turbo on left, Turb(DP)o on right. The video shown is sped up 2x to stop typing being the bottleneck
In a Partiprompts evaluation, this Turb(DPO)o checkpoint already beats "base" Turbo 55% of the time. We're excited to see where this goes!
As with the recently released SD1.5/SDXL ckpts we expect public models soon.
Also, this training will soon be integrated into diffusers as well (github.com/huggingface/diffu…)