👁️🎧 DuplexJev-32B-Vision — it hears, and now it sees

One forward pass over a voice and a picture. Ask multiple-choice questions that need both: does the face match the voice? is the red word the one they stress? does the caption describe the sound? Each answer is one constrained token — nothing is decoded. Research preview · model card · ⭐ GitHub · speech-only demo: DuplexJev · 🏢 commercial licence & hosted API: jiejin@adventists.ai

_{⏱️ The first run after the Space has been idle takes a few seconds while Hugging Face attaches a GPU. Free GPU time per visitor is limited by Hugging Face — log in for more runs. 首次运行需要几秒分配 GPU；登录 Hugging Face 可获得更多免费次数。}

_{Model: DuplexJev-32B-Vision (CC BY-NC 4.0, research preview); answers averaged over both option orders. Questions work best in English; speech can be English or Chinese. One picture per request. Real-photo clips: CREMA-D (ODbL / DbCL), actors not seen in training. Pictures and audio are processed for this request only and not stored. Do not use the outputs to make decisions about individuals.}