SwiftKV: Fast Prefill-Optimized Inference with Knowledge-Preserving Model Transformation Paper • 2410.03960 • Published 7 days ago • 1