DeepSeek-R1-Distill-Qwen-7B-GGUF / scores /deepseek-r1-distill-qwen-7b-q5_k_s.log
eaddario's picture
Generate perplexity and kld scores
75442e7 unverified
====== Perplexity statistics ======
Mean PPL(Q) : 25.403544 ± 0.248578
Mean PPL(base) : 24.931431 ± 0.241228
Cor(ln(PPL(Q)), ln(PPL(base))): 99.79%
Mean ln(PPL(Q)/PPL(base)) : 0.018759 ± 0.000634
Mean PPL(Q)/PPL(base) : 1.018936 ± 0.000646
Mean PPL(Q)-PPL(base) : 0.472114 ± 0.017348
====== KL divergence statistics ======
Mean KLD: 0.009320 ± 0.000039
Maximum KLD: 2.831377
99.9% KLD: 0.127923
99.0% KLD: 0.059484
99.0% KLD: 0.059484
Median KLD: 0.005393
10.0% KLD: 0.000103
5.0% KLD: 0.000013
1.0% KLD: -0.000002
Minimum KLD: -0.000203
====== Token probability statistics ======
Mean Δp: -0.016 ± 0.006 %
Maximum Δp: 42.542%
99.9% Δp: 15.234%
99.0% Δp: 7.440%
95.0% Δp: 3.334%
90.0% Δp: 1.787%
75.0% Δp: 0.232%
Median Δp: -0.000%
25.0% Δp: -0.256%
10.0% Δp: -1.792%
5.0% Δp: -3.371%
1.0% Δp: -7.776%
0.1% Δp: -15.189%
Minimum Δp: -60.684%
RMS Δp : 2.343 ± 0.015 %
Same top p: 95.099 ± 0.056 %