nanovdr/NanoVDR-D-Fast-Qwen3VL8B-4096

Cite arXiv:2608.10636

Baseline comparison in Results; link the write-up

Ship the tiling image processor; load as a SentenceTransformer too

Ship the tiling image processor; load as a SentenceTransformer too

Ship the tiling image processor; load as a SentenceTransformer too

Ship the tiling image processor; load as a SentenceTransformer too

Ship the tiling image processor; load as a SentenceTransformer too

Ship the tiling image processor; load as a SentenceTransformer too

Ship the tiling image processor; load as a SentenceTransformer too

query tower carries -ML: it is trained on the multilingual mixture

point the banner at the current repo name

matrix: state the deployment gain rather than a GPU latency

rename note, pairing rule, and the 2x2 usage matrix

NanoVDR-D-Fast: 2-tile document tower distilled from Qwen3-VL-Embedding-8B

initial commit