YANAMI: 400K-dialog multimodal dataset shows models still fail at stateful multiturn interaction
Shengyu Wang · hf · 2026-11-04
Researchers from HKUST, Shanghai AI Lab and other institutions released YANAMI, a large-scale multimodal dialogue dataset and benchmark for multiturn stateful interaction.
- Training set (YANAMI-400K): 400K dialogues, 2M+ images, 2.6M turns
- Benchmark (YANAMI Bench): human-verified, with 400 dialogues, 1,883 images and 8,467 scoring points, requiring models to track and realize evolving dialogue states across turns of image editing/generation
- Finding: current unified multimodal models still struggle with this task
Project page: rowletqwq.github.io/YANAMI
More from Multimodal
- Kitten-TTS 2 offers free voice cloning on laptops, no GPU needed — solyarisoftware · 2026-11-04
- "We'll Take Your RAM": a fully local-hardware cyberpunk meme from 2030 San Francisco — Wonderful-Yak-3348 · 2026-11-04
- Openfilm: open-source AI video agent turns coding agents into video makers — solyarisoftware · 2026-11-04
- Microsoft reopens VibeVoice: 10-second voice cloning, 90-minute local TTS — solyarisoftware · 2026-11-04
- A recipe for flawless one-step distant faces in H3 without a SAM face-swap pass — robertwellesley · 2026-11-04
- FlowTranslator: training-free in-image translation via diffusion flow editing — Fudan-University · 2026-11-04