Inference Memory Analysis on NVIDIA H100 NVL (seq_len=1024)
744Model Weights SizeTernary Mamba
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ternary MambaConfiguration=Ternary (packed, proj.)2026.06 | 744 | 1,100 | |
| FP16Configuration=FP162026.06 | 2,599 | 2,932 | |
| Ternary MambaConfiguration=Ternary (sim.)2026.06 | 5,322 | 6,932 |