| ====== Perplexity statistics ====== | |
| Mean PPL(Q) : 26.143038 ± 0.254443 | |
| Mean PPL(base) : 23.352232 ± 0.220841 | |
| Cor(ln(PPL(Q)), ln(PPL(base))): 98.95% | |
| Mean ln(PPL(Q)/PPL(base)) : 0.112890 ± 0.001420 | |
| Mean PPL(Q)/PPL(base) : 1.119509 ± 0.001590 | |
| Mean PPL(Q)-PPL(base) : 2.790806 ± 0.048102 | |
| ====== KL divergence statistics ====== | |
| Mean KLD: 0.082486 ± 0.000324 | |
| Maximum KLD: 8.223216 | |
| 99.9% KLD: 1.163796 | |
| 99.0% KLD: 0.554553 | |
| 99.0% KLD: 0.554553 | |
| Median KLD: 0.040744 | |
| 10.0% KLD: 0.000979 | |
| 5.0% KLD: 0.000241 | |
| 1.0% KLD: 0.000008 | |
| Minimum KLD: -0.000550 | |
| ====== Token probability statistics ====== | |
| Mean Δp: -0.290 ± 0.018 % | |
| Maximum Δp: 83.668% | |
| 99.9% Δp: 39.492% | |
| 99.0% Δp: 22.349% | |
| 95.0% Δp: 10.000% | |
| 90.0% Δp: 4.974% | |
| 75.0% Δp: 0.538% | |
| Median Δp: -0.001% | |
| 25.0% Δp: -0.805% | |
| 10.0% Δp: -5.962% | |
| 5.0% Δp: -11.458% | |
| 1.0% Δp: -25.491% | |
| 0.1% Δp: -47.606% | |
| Minimum Δp: -94.438% | |
| RMS Δp : 7.163 ± 0.036 % | |
| Same top p: 87.635 ± 0.085 % | |
| llama_perf_context_print: load time = 80439.51 ms | |
| llama_perf_context_print: prompt eval time = 1705180.19 ms / 304128 tokens ( 5.61 ms per token, 178.36 tokens per second) | |
| llama_perf_context_print: eval time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second) | |
| llama_perf_context_print: total time = 1852832.45 ms / 304129 tokens | |