| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 7.841948 ± 0.049502 | |
| Mean PPL(base) : 7.237090 ± 0.045539 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 98.36% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.080268 ± 0.001143 | |
| Mean PPL(Q)/PPL(base) : 1.083578 ± 0.001238 | |
| Mean PPL(Q)-PPL(base) : 0.604858 ± 0.009476 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.081774 ± 0.000354 | |
| Maximum KLD: 7.690053 | |
| 99.9% KLD: 1.654508 | |
| 99.0% KLD: 0.555790 | |
| 99.0% KLD: 0.555790 | |
| Median KLD: 0.056256 | |
| 10.0% KLD: 0.003426 | |
| 5.0% KLD: 0.001063 | |
| 1.0% KLD: 0.000157 | |
| Minimum KLD: 0.000000 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -2.133 ± 0.021 % | |
| Maximum Δp: 73.495% | |
| 99.9% Δp: 32.336% | |
| 99.0% Δp: 17.093% | |
| 95.0% Δp: 7.846% | |
| 90.0% Δp: 4.045% | |
| 75.0% Δp: 0.372% | |
| Median Δp: -0.301% | |
| 25.0% Δp: -3.967% | |
| 10.0% Δp: -10.805% | |
| 5.0% Δp: -16.007% | |
| 1.0% Δp: -30.015% | |
| 0.1% Δp: -62.256% | |
| Minimum Δp: -96.763% | |
| RMS Δp : 8.316 ± 0.043 % | |
| Same top p: 85.224 ± 0.094 % | |