| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 15.513099 ± 0.129202 | |
| Mean PPL(base) : 14.009216 ± 0.118474 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 94.85% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.101969 ± 0.002698 | |
| Mean PPL(Q)/PPL(base) : 1.107350 ± 0.002987 | |
| Mean PPL(Q)-PPL(base) : 1.503883 ± 0.041148 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.320092 ± 0.001344 | |
| Maximum KLD: 17.145338 | |
| 99.9% KLD: 5.248430 | |
| 99.0% KLD: 2.399956 | |
| 99.0% KLD: 2.399956 | |
| Median KLD: 0.171489 | |
| 10.0% KLD: 0.005972 | |
| 5.0% KLD: 0.001335 | |
| 1.0% KLD: 0.000109 | |
| Minimum KLD: -0.000000 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -3.246 ± 0.037 % | |
| Maximum Δp: 99.963% | |
| 99.9% Δp: 71.232% | |
| 99.0% Δp: 34.565% | |
| 95.0% Δp: 13.472% | |
| 90.0% Δp: 6.248% | |
| 75.0% Δp: 0.411% | |
| Median Δp: -0.166% | |
| 25.0% Δp: -5.212% | |
| 10.0% Δp: -17.956% | |
| 5.0% Δp: -28.544% | |
| 1.0% Δp: -57.576% | |
| 0.1% Δp: -87.811% | |
| Minimum Δp: -99.472% | |
| RMS Δp : 14.572 ± 0.064 % | |
| Same top p: 75.385 ± 0.114 % | |