> We propose adopting diffusion language models for text embeddings, motivated by their inherent bidirectional architecture and recent success in matching or surpassing LLMs especially on reasoning task.
I didn't realize diffusion language models were at this point yet. But what's the catch? why aren't diffusion models (or some kind of hybrid) taking over?
Klathmon•1d ago
I didn't realize diffusion language models were at this point yet. But what's the catch? why aren't diffusion models (or some kind of hybrid) taking over?