Average Per-Batch Runtime on Text Reconstruction from Gradients on CoLA
3Average per-batch runtime (seconds)DAGER
Evaluation results
| Method | Links | |
|---|---|---|
| DAGERModel=GPT-2, Hardware=RTX 5060, Batch Size (B)=12026.03 | 3 | |
| DAGERModel=GPT-2, Hardware=RTX 5060, Batch Size (B)=22026.03 | 6 | |
| SOMPBeam Width (Bw)=4, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=12026.03 | 7 | |
| SOMPBeam Width (Bw)=8, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=12026.03 | 7 | |
| SOMPBeam Width (Bw)=16, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=12026.03 | 8 | |
| DAGERModel=GPT-2, Hardware=RTX 5060, Batch Size (B)=42026.03 | 10 | |
| SOMPBeam Width (Bw)=4, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=22026.03 | 10 | |
| SOMPBeam Width (Bw)=8, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=22026.03 | 10 | |
| SOMPBeam Width (Bw)=4, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=42026.03 | 13 | |
| SOMPBeam Width (Bw)=16, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=22026.03 | 13 | |
| SOMPBeam Width (Bw)=8, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=42026.03 | 14 | |
| DAGERModel=GPT-2, Hardware=RTX 5060, Batch Size (B)=82026.03 | 19 | |
| SOMPBeam Width (Bw)=4, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=82026.03 | 21 | |
| SOMPBeam Width (Bw)=16, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=42026.03 | 21 | |
| SOMPBeam Width (Bw)=8, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=82026.03 | 24 | |
| SOMPBeam Width (Bw)=4, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=162026.03 | 26 | |
| SOMPBeam Width (Bw)=8, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=162026.03 | 31 | |
| SOMPBeam Width (Bw)=16, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=82026.03 | 33 | |
| DAGERModel=GPT-2, Hardware=RTX 5060, Batch Size (B)=162026.03 | 36 | |
| SOMPBeam Width (Bw)=16, Model=GPT-2, Hardware=RTX 5060, Batch Size (B)=162026.03 | 41 |