| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 7.582235 ± 0.048626 | |
| Mean PPL(base) : 7.534124 ± 0.048206 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 99.90% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.006365 ± 0.000282 | |
| Mean PPL(Q)/PPL(base) : 1.006386 ± 0.000284 | |
| Mean PPL(Q)-PPL(base) : 0.048111 ± 0.002168 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.004976 ± 0.000032 | |
| Maximum KLD: 1.635007 | |
| 99.9% KLD: 0.127847 | |
| 99.0% KLD: 0.038030 | |
| 99.0% KLD: 0.038030 | |
| Median KLD: 0.003028 | |
| 10.0% KLD: 0.000180 | |
| 5.0% KLD: 0.000057 | |
| 1.0% KLD: 0.000006 | |
| Minimum KLD: -0.000153 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -0.061 ± 0.005 % | |
| Maximum Δp: 42.055% | |
| 99.9% Δp: 10.474% | |
| 99.0% Δp: 5.063% | |
| 95.0% Δp: 2.612% | |
| 90.0% Δp: 1.612% | |
| 75.0% Δp: 0.375% | |
| Median Δp: -0.001% | |
| 25.0% Δp: -0.410% | |
| 10.0% Δp: -1.671% | |
| 5.0% Δp: -2.791% | |
| 1.0% Δp: -6.049% | |
| 0.1% Δp: -16.358% | |
| Minimum Δp: -71.814% | |
| RMS Δp : 1.956 ± 0.020 % | |
| Same top p: 96.417 ± 0.049 % | |