| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 16.919080 ± 0.146005 | |
| Mean PPL(base) : 14.009216 ± 0.118474 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 94.48% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.188727 ± 0.002845 | |
| Mean PPL(Q)/PPL(base) : 1.207711 ± 0.003435 | |
| Mean PPL(Q)-PPL(base) : 2.909864 ± 0.051660 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.367114 ± 0.001473 | |
| Maximum KLD: 14.687531 | |
| 99.9% KLD: 5.463440 | |
| 99.0% KLD: 2.657840 | |
| 99.0% KLD: 2.657840 | |
| Median KLD: 0.200170 | |
| 10.0% KLD: 0.005165 | |
| 5.0% KLD: 0.001071 | |
| 1.0% KLD: 0.000090 | |
| Minimum KLD: -0.000003 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -3.211 ± 0.040 % | |
| Maximum Δp: 99.154% | |
| 99.9% Δp: 69.973% | |
| 99.0% Δp: 37.230% | |
| 95.0% Δp: 16.254% | |
| 90.0% Δp: 7.806% | |
| 75.0% Δp: 0.528% | |
| Median Δp: -0.114% | |
| 25.0% Δp: -5.124% | |
| 10.0% Δp: -19.489% | |
| 5.0% Δp: -31.482% | |
| 1.0% Δp: -60.025% | |
| 0.1% Δp: -89.899% | |
| Minimum Δp: -99.661% | |
| RMS Δp : 15.568 ± 0.064 % | |
| Same top p: 72.541 ± 0.118 % | |