nepa-next-embedding-prediction

【效率与协作】Train vision transformers through autoregressive next-embedding prediction without pixel reconstruction, tokenizers, or contrastive losses. Apply causal masking and stop-gradient on target embeddings. Achieve 83.8% (ViT-B) and 85.3% (ViT-L) ImageNet-1K accuracy with strong transfer to downstream tasks.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准