Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality
尚未评估已列入计划指标结果claim-ssd-speed-speedup
On an NVIDIA A100-80GB PCIe GPU with state dimension N=64, the SSD layer implementation achieves 2x to 8x speedup over Mamba-1's optimized fused associative scan across sequence lengths from 512 to 512K, and outperforms FlashAttention-2 at sequence lengths 2K and above.
来源：source-paper:Section 9.3, Figure 10 (Left), page 29
报告指标与观测值
speedup_factor_vs_mamba1
rm-ssd-speedup-lower
论文报告 2 factor
实际观测 — factor
speedup_factor_vs_mamba1
rm-ssd-speedup-upper
论文报告 8 factor
实际观测 — factor
Assessment（0）
这条 Claim 暂无已发布的不可变 Assessment。
关联运行（0）
这条 Claim 暂无关联执行。