This might explain why REPA works: aligning DiT representations to DINOv2 may just be helping the model form these morphogen-like spatial gradients earlier and cleaner. One injects good representations, the other reveals them — same coin, two sides. Going further — what's the principle behind RAE?🤔
The future AI engineer probably spends less time training models and more time working on
data pipelines
retrieval
evaluation
agents
inference optimization
Training may become the easy part.
The future AI engineer probably spends less time training models and more time working on
data pipelines
retrieval
evaluation
agents
inference optimization
Training may become the easy part.
I was going crazy because I could not replicate TurboQuant. Turns out the community also had issues. The community quickly made adjustments to "make it work", but what they did not realize is that they reimplemented (most of) HIGGS in the process (full HIGGS would be even better)
انا مين نلقى جزايريين هنا نديرلهم فولو حتى لو المحتوى تاعهم لا علاقة بيا
وين كنتو الوقت هذا كامل والله سيريو؟ موالف هنا نلقى غير تع الخليج ها المزية لي بنتو
فرصتك للفوز بـ 30 رحلة عمرة مجانية في تحدي نسك 🕋! استخدم الكود لدعمي، وسجّل الآن عبر الرابط وابدأ التحدي — قد تكون رحلتك القادمة إلى العمرة من نصيبك. #تطبيق_نسك
https://t.co/QYNdwh149r