| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 9.798719 ± 0.061509 | |
| Mean PPL(base) : 7.237090 ± 0.045539 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 94.51% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.303032 ± 0.002083 | |
| Mean PPL(Q)/PPL(base) : 1.353958 ± 0.002821 | |
| Mean PPL(Q)-PPL(base) : 2.561629 ± 0.023725 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.285340 ± 0.000933 | |
| Maximum KLD: 8.015823 | |
| 99.9% KLD: 3.395762 | |
| 99.0% KLD: 1.885579 | |
| 99.0% KLD: 1.885579 | |
| Median KLD: 0.207489 | |
| 10.0% KLD: 0.018084 | |
| 5.0% KLD: 0.005005 | |
| 1.0% KLD: 0.000535 | |
| Minimum KLD: 0.000000 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -7.606 ± 0.042 % | |
| Maximum Δp: 79.040% | |
| 99.9% Δp: 41.873% | |
| 99.0% Δp: 22.236% | |
| 95.0% Δp: 8.510% | |
| 90.0% Δp: 3.272% | |
| 75.0% Δp: 0.019% | |
| Median Δp: -1.956% | |
| 25.0% Δp: -11.831% | |
| 10.0% Δp: -26.778% | |
| 5.0% Δp: -39.549% | |
| 1.0% Δp: -72.463% | |
| 0.1% Δp: -89.273% | |
| Minimum Δp: -99.255% | |
| RMS Δp : 17.751 ± 0.067 % | |
| Same top p: 74.688 ± 0.115 % | |