Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality
Not assessedPlannedMeasurementclaim-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.
Source: source-paper:Section 9.3, Figure 10 (Left), page 29
Reported and observed measurements
speedup_factor_vs_mamba1
rm-ssd-speedup-lower
Reported 2 factor
Observed — factor
speedup_factor_vs_mamba1
rm-ssd-speedup-upper
Reported 8 factor
Observed — factor
Assessments (0)
No immutable Assessment has been published for this Claim yet.
Runs (0)
No execution has been linked to this Claim yet.