Language Modeling on PTB (LLaMA Model Evaluation)
37.91Perplexity (LLaMA-2 7/8B)Full Precision
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full PrecisionWeight Bits=162025.12 | 37.91 | 50.93 | 11.18 | |
| PB-LLMWeight Alignment (WA)=true, Block Size=128, Weight Bits=1.72025.12 | 657.24 | 816.31 | 106.25 | |
| ARB-XOutput Alignment (OA)=true, Block Size=128, Weight Bits=1.062025.12 | 681.24 | 182.1 | 53.86 | |
| ARB-RCWeight Alignment (WA)=true, Block Size=128, Weight Bits=1.062025.12 | 763.19 | 197.7 | 47.88 | |
| OursOutput Alignment (OA)=true, Block Size=128, Weight Bits=1.062025.12 | 3,166 | 196.64 | 45.66 | |
| BiLLMWeight Alignment (WA)=true, Block Size=128, Weight Bits=1.062025.12 | 5,243.01 | 309.12 | 87.25 |