Jiaqi-hkust/Robust-U1-SFT
Any-to-Any • 15B • Updated • 4 • 1
Jiaqi Tang PRO
Jiaqi-hkust
AI & ML interests
MLLM, LLM, Agentic AI
Recent Activity
liked a model 10 days ago
tencent/Hy4-preview reacted to theirpost with 🤗 27 days ago
🧠Remember-R1: Our fix for MLLMs forgetting the image during long reasoning
We noticed a frustrating problem: when multimodal models reason over long chains, they gradually stop looking at the image—and start hallucinating based on their own text.
So we built Remember‑R1, a simple RL framework that directly supervises visual attention on the original reasoning trajectory—no inference overhead, no proxy tasks.
We use three complementary rewards: coverage, persistence, and focus. They encourage the model to keep attending to relevant visual evidence even in later reasoning steps.
Results across 7 benchmarks and 2 model sizes: better reasoning, and—more importantly—visual attention decays much more slowly during generation.
No extra cost at inference, just cleaner supervision where it counts.
đź“„ Paper: https://arxiv.org/abs/2608.01314
đź’» Code: https://github.com/Ch921-cell/Remember-R1
Happy to answer any questions and receive feedback!
#MultimodalAI #RL #MLLM #CoT #VisualReasoning reacted to theirpost with âž• 28 days ago
🧠Remember-R1: Our fix for MLLMs forgetting the image during long reasoning
We noticed a frustrating problem: when multimodal models reason over long chains, they gradually stop looking at the image—and start hallucinating based on their own text.
So we built Remember‑R1, a simple RL framework that directly supervises visual attention on the original reasoning trajectory—no inference overhead, no proxy tasks.
We use three complementary rewards: coverage, persistence, and focus. They encourage the model to keep attending to relevant visual evidence even in later reasoning steps.
Results across 7 benchmarks and 2 model sizes: better reasoning, and—more importantly—visual attention decays much more slowly during generation.
No extra cost at inference, just cleaner supervision where it counts.
đź“„ Paper: https://arxiv.org/abs/2608.01314
đź’» Code: https://github.com/Ch921-cell/Remember-R1
Happy to answer any questions and receive feedback!
#MultimodalAI #RL #MLLM #CoT #VisualReasoningOrganizations
