Black Forest Labs stellt Self-Flow vor. Das Verfahren halbiert die Trainingszeit für generative KI.
Statt externer Modelle nutzt Self-Flow eine Exponential Moving Average Kopie des eigenen Modells als Lehrer. Der Fehlerwert wird zwischen aktueller Vorhersage und der internen EMA-Kopie berechnet. Das vereinfacht die Infrastruktur und senkt den FID-Wert. #BlackForestLabs #SelfFlow #GenerativeAI
https://www.all-ai.de/news/news26/blackforest-self-flow
Hila Chefer (@hila_chefer)
bfl_ml의 새로운 연구 'Self-Flow'는 이미지·오디오·비디오·월드 모델을 아우르는 자기지도 프레임워크입니다. 저자들은 생성 모델이 강한 표현 학습을 위해 DINO 같은 별도 기법을 반드시 필요로 하지 않다고 보고, 생성 모델을 공동 프레임워크로 직접 가르치는 접근을 제안합니다.
https://x.com/hila_chefer/status/2029212708797661327
#selfflow #selfsupervised #multimodal #generativemodels #bfl
New research from @bfl_ml 🥳 Meet Self-Flow: our self-supervised framework for image, audio, video & world models 🤖 https://t.co/AshY8IkSEe Do generative models really need DINO to learn strong representations? We propose teaching them directly via a joint framework instead 🧵
Black Forest Labs' new Self‑Flow framework cuts multimodal AI training time by 2.8× versus REPA, thanks to smarter feature alignment and better computational efficiency. Open‑source researchers can now train larger models faster. Dive into the details to see how this could reshape your ML pipelines. #SelfFlow #MultimodalAI #AITraining #ComputationalEfficiency
🔗 https://aidailypost.com/news/black-forest-labs-self-flow-speeds-multimodal-ai-training-28-faster
Andreas Becker
sayzard
AI Daily Post