Circle-RoPE: Cone-like Decoupled Rotary Positional Embedding for Large Vision-Language Models

Re-assign community