InfLLM-V2: Dense-Sparse Switchable Attention for Seamless Short-to-Long Adaptation

Open in new window