Communication Cost Analysis on LLaVA 1.5
97.268Total Latency (s)RD-FSQ
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RD-FSQBit=2-bit2025.11 | 97.268 | 133.33 | 0.558 | 782.8 | |
| QLoRABit=2-bit2025.11 | 131.872 | 177.01 | 0.756 | 1,039.17 | |
| RD-FSQBit=3-bit2025.11 | 177.558 | 197.98 | 1.018 | 1,162.25 | |
| QLoRABit=4-bit2025.11 | 218.56 | 306.26 | 1.254 | 1,798.07 | |
| QLoRABit=3-bit2025.11 | 235.404 | 241.63 | 1.35 | 1,418.62 | |
| RD-FSQBit=4-bit2025.11 | 241.993 | 262.6 | 1.387 | 1,541.71 | |
| Original ModelBit=16-bit2025.11 | 6,773.924 | 1,038.47 | 38.839 | 6,095.12 |