| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 17.644226 ± 0.148490 | |
| Mean PPL(base) : 14.009216 ± 0.118474 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 92.27% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.230693 ± 0.003317 | |
| Mean PPL(Q)/PPL(base) : 1.259473 ± 0.004177 | |
| Mean PPL(Q)-PPL(base) : 3.635010 ± 0.060162 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.506686 ± 0.001786 | |
| Maximum KLD: 24.616240 | |
| 99.9% KLD: 6.167073 | |
| 99.0% KLD: 3.207879 | |
| 99.0% KLD: 3.207879 | |
| Median KLD: 0.314625 | |
| 10.0% KLD: 0.011043 | |
| 5.0% KLD: 0.002415 | |
| 1.0% KLD: 0.000202 | |
| Minimum KLD: 0.000000 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -5.331 ± 0.047 % | |
| Maximum Δp: 99.838% | |
| 99.9% Δp: 74.946% | |
| 99.0% Δp: 39.024% | |
| 95.0% Δp: 15.816% | |
| 90.0% Δp: 7.011% | |
| 75.0% Δp: 0.276% | |
| Median Δp: -0.392% | |
| 25.0% Δp: -8.480% | |
| 10.0% Δp: -26.744% | |
| 5.0% Δp: -40.760% | |
| 1.0% Δp: -69.996% | |
| 0.1% Δp: -93.958% | |
| Minimum Δp: -99.872% | |
| RMS Δp : 18.534 ± 0.068 % | |
| Same top p: 67.981 ± 0.123 % | |