| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 7.586838 ± 0.048693 | |
| Mean PPL(base) : 7.534124 ± 0.048206 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 99.91% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.006972 ± 0.000278 | |
| Mean PPL(Q)/PPL(base) : 1.006997 ± 0.000280 | |
| Mean PPL(Q)-PPL(base) : 0.052714 ± 0.002153 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.004726 ± 0.000029 | |
| Maximum KLD: 1.109797 | |
| 99.9% KLD: 0.120477 | |
| 99.0% KLD: 0.036395 | |
| 99.0% KLD: 0.036395 | |
| Median KLD: 0.002865 | |
| 10.0% KLD: 0.000173 | |
| 5.0% KLD: 0.000053 | |
| 1.0% KLD: 0.000006 | |
| Minimum KLD: -0.000141 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -0.024 ± 0.005 % | |
| Maximum Δp: 40.801% | |
| 99.9% Δp: 10.374% | |
| 99.0% Δp: 4.971% | |
| 95.0% Δp: 2.605% | |
| 90.0% Δp: 1.614% | |
| 75.0% Δp: 0.410% | |
| Median Δp: -0.000% | |
| 25.0% Δp: -0.359% | |
| 10.0% Δp: -1.547% | |
| 5.0% Δp: -2.679% | |
| 1.0% Δp: -6.014% | |
| 0.1% Δp: -15.549% | |
| Minimum Δp: -67.857% | |
| RMS Δp : 1.918 ± 0.020 % | |
| Same top p: 96.547 ± 0.048 % | |