DeepSeek-R1-Distill-Qwen-7B-GGUF / scores /deepseek-r1-distill-qwen-7b-q6_k.log
eaddario's picture
Generate perplexity and kld scores
75442e7 unverified
====== Perplexity statistics ======
Mean PPL(Q) : 25.139045 ± 0.245198
Mean PPL(base) : 24.931431 ± 0.241228
Cor(ln(PPL(Q)), ln(PPL(base))): 99.87%
Mean ln(PPL(Q)/PPL(base)) : 0.008293 ± 0.000501
Mean PPL(Q)/PPL(base) : 1.008327 ± 0.000505
Mean PPL(Q)-PPL(base) : 0.207614 ± 0.013014
====== KL divergence statistics ======
Mean KLD: 0.003335 ± 0.000014
Maximum KLD: 1.228119
99.9% KLD: 0.039909
99.0% KLD: 0.019418
99.0% KLD: 0.019418
Median KLD: 0.002067
10.0% KLD: 0.000037
5.0% KLD: 0.000005
1.0% KLD: -0.000004
Minimum KLD: -0.000159
====== Token probability statistics ======
Mean Δp: -0.007 ± 0.004 %
Maximum Δp: 22.155%
99.9% Δp: 8.711%
99.0% Δp: 4.504%
95.0% Δp: 2.060%
90.0% Δp: 1.096%
75.0% Δp: 0.141%
Median Δp: -0.000%
25.0% Δp: -0.153%
10.0% Δp: -1.126%
5.0% Δp: -2.115%
1.0% Δp: -4.485%
0.1% Δp: -8.494%
Minimum Δp: -55.308%
RMS Δp : 1.402 ± 0.011 %
Same top p: 96.985 ± 0.044 %