HeadInfer: Memory-Efficient LLM Inference by Head-wise Offloading

Re-assign community