Tencent releases EVIE-Preview-4.5B visual document retrieval model

tencent · hf · 2026-08-17

Tencent released EVIE-Preview-4.5B on Hugging Face, a vision-language model for the visual-document-retrieval pipeline. It uses ColPali/ColBERT-style late-interaction with multi-vector representations on a Qwen3.5 base, targeting document-image retrieval.

Related event: Tencent's EVIE Tops ViDoRe Leaderboard, Cutting Vector Storage 32x(2 posts)→

Original post →

More from Multimodal

Multimodal channel →