Model Compression Latency on C4
342Latency (rho=0.8)Swift-SVD
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Swift-SVDSamples=162026.04 | 342 | 146 | 133 | 621 | 3.2 | |
| Swift-SVDSamples=642026.04 | 346 | 158 | 150 | 654 | 12.1 | |
| Swift-SVDSamples=2562026.04 | 453 | 152 | 148 | 753 | 42.1 | |
| Swift-SVDSamples=5122026.04 | 540 | 157 | 130 | 827 | 76.9 | |
| SVD-LLM (W)Samples=162026.04 | 542 | 489 | 503 | 1,534 | 1.3 | |
| SVD-LLM (W)Samples=642026.04 | 555 | 530 | 565 | 1,650 | 4.8 | |
| SVD-LLM (W)Samples=2562026.04 | 757 | 734 | 722 | 2,213 | 14.3 | |
| Dobi-SVD(w/o)Samples=162026.04 | 960 | 650 | 373 | 1,983 | 1 | |
| SVD-LLM (W)Samples=5122026.04 | 1,080 | 1,069 | 1,063 | 3,212 | 19.8 | |
| Dobi-SVD(w/o)Samples=642026.04 | 3,823 | 2,551 | 1,508 | 7,882 | 1 | |
| Dobi-SVD(w/o)Samples=2562026.04 | 15,269 | 10,468 | 5,966 | 31,703 | 1 | |
| Dobi-SVD(w/o)Samples=5122026.04 | 30,862 | 20,984 | 11,795 | 63,641 | 1 |