DeepSeek-R1-Distill-Qwen-7B-GGUF / scores /deepseek-r1-distill-qwen-7b-q8_0.log
eaddario's picture
Generate perplexity and kld scores
75442e7 unverified
====== Perplexity statistics ======
Mean PPL(Q) : 25.118649 ± 0.245079
Mean PPL(base) : 24.931431 ± 0.241228
Cor(ln(PPL(Q)), ln(PPL(base))): 99.91%
Mean ln(PPL(Q)/PPL(base)) : 0.007481 ± 0.000431
Mean PPL(Q)/PPL(base) : 1.007509 ± 0.000434
Mean PPL(Q)-PPL(base) : 0.187218 ± 0.011259
====== KL divergence statistics ======
Mean KLD: 0.000361 ± 0.000002
Maximum KLD: 0.215854
99.9% KLD: 0.004745
99.0% KLD: 0.002203
99.0% KLD: 0.002203
Median KLD: 0.000212
10.0% KLD: 0.000002
5.0% KLD: -0.000000
1.0% KLD: -0.000014
Minimum KLD: -0.000127
====== Token probability statistics ======
Mean Δp: 0.003 ± 0.001 %
Maximum Δp: 24.559%
99.9% Δp: 2.974%
99.0% Δp: 1.523%
95.0% Δp: 0.686%
90.0% Δp: 0.364%
75.0% Δp: 0.050%
Median Δp: 0.000%
25.0% Δp: -0.047%
10.0% Δp: -0.356%
5.0% Δp: -0.670%
1.0% Δp: -1.484%
0.1% Δp: -2.907%
Minimum Δp: -7.835%
RMS Δp : 0.468 ± 0.005 %
Same top p: 98.983 ± 0.026 %