Prompt Compression on L-Eval (test)
64.4Coursera QA AccuracyLLMLingua-2
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LLMLingua-2Tokens=2,110, 1/τ=4×, Compression Budget=2,0002026.03 | 64.4 | 55 | 66.4 | 73.2 | 45 | 23.7 | 54.6 | |
| BEAVERTokens=2,180, 1/τ=4×, Compression Budget=2,0002026.03 | 64.4 | 56.9 | 76.6 | 74 | 44.7 | 28.8 | 57.6 | |
| Simply-BEAVERTokens=2,123, 1/τ=4×, Compression Budget=2,0002026.03 | 63.1 | 52.9 | 74.2 | 72.5 | 39.7 | 27.2 | 54.9 | |
| LongLLMLinguaTokens=2,196, 1/τ=4×, Compression Budget=2,0002026.03 | 62.2 | 51 | 65.6 | 71 | 37.7 | 21.3 | 51.5 | |
| LLMLinguaTokens=2,122, 1/τ=4×, Compression Budget=2,0002026.03 | 58.9 | 50 | 60.9 | 65.1 | 34.3 | 21.9 | 48.5 |