| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 6.611877 ± 0.040700 | |
| Mean PPL(base) : 6.554978 ± 0.040159 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 99.79% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.008643 ± 0.000403 | |
| Mean PPL(Q)/PPL(base) : 1.008680 ± 0.000406 | |
| Mean PPL(Q)-PPL(base) : 0.056898 ± 0.002699 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.009900 ± 0.000071 | |
| Maximum KLD: 2.807195 | |
| 99.9% KLD: 0.340907 | |
| 99.0% KLD: 0.088291 | |
| 99.0% KLD: 0.088291 | |
| Median KLD: 0.005181 | |
| 10.0% KLD: 0.000230 | |
| 5.0% KLD: 0.000062 | |
| 1.0% KLD: 0.000007 | |
| Minimum KLD: -0.000068 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -0.112 ± 0.008 % | |
| Maximum Δp: 59.985% | |
| 99.9% Δp: 17.692% | |
| 99.0% Δp: 7.839% | |
| 95.0% Δp: 3.690% | |
| 90.0% Δp: 2.194% | |
| 75.0% Δp: 0.496% | |
| Median Δp: -0.001% | |
| 25.0% Δp: -0.613% | |
| 10.0% Δp: -2.427% | |
| 5.0% Δp: -4.015% | |
| 1.0% Δp: -9.158% | |
| 0.1% Δp: -26.484% | |
| Minimum Δp: -87.765% | |
| RMS Δp : 2.969 ± 0.028 % | |
| Same top p: 94.988 ± 0.058 % | |