Yingfa Chen

Aggressive KV cache compression can be achieved with near-zero accuracy loss by treating the cache as a trainable neural network optimized at test-time.

Larger sliding-window attention can paradoxically slow down the formation of critical retrieval mechanisms in language models, challenging conventional design assumptions.

CoT fine-tuning can slash long-range recall by over 57% in hybrid LLMs, but a simple parameter restoration method can reverse this trend without additional training.