Sentiment Analysis on SST (Clean/Trigger Evaluation)
100Accuracy (Clean, Eager)Llama-3.2-1B-Instruct
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Llama-3.2-1B-InstructExecution Backend=CUDAGraphs, Attack=Attack II2026.05 | 100 | 100 | 45 | 53.7 | |
| Llama-3.2-3B-InstructExecution Backend=CUDAGraphs, Attack=Attack II2026.05 | 100 | 100 | 37.5 | 63.7 | |
| Qwen2.5-3B-InstructExecution Backend=CUDAGraphs, Attack=Attack II2026.05 | 100 | 100 | 81.2 | 53.7 | |
| Qwen2.5-1.5B-InstructExecution Backend=CUDAGraphs, Attack=Attack II2026.05 | 95 | 96.3 | 58.8 | 88.7 |