The design of attention mechanisms in long-context large language models faces an ``impossible triangle'' among KV cache size, training and prefill FLOPs, an…
机构:百度
来源:arXiv 2610.17959 | AI4Papers 论文推荐平台