DeepSeek-R1-Distill-Qwen-7B-GGUF / scores /deepseek-r1-distill-qwen-7b-q4_k_m.log
eaddario's picture
Generate perplexity and kld scores
75442e7 unverified
====== Perplexity statistics ======
Mean PPL(Q) : 25.409863 ± 0.248331
Mean PPL(base) : 24.931431 ± 0.241228
Cor(ln(PPL(Q)), ln(PPL(base))): 99.58%
Mean ln(PPL(Q)/PPL(base)) : 0.019008 ± 0.000899
Mean PPL(Q)/PPL(base) : 1.019190 ± 0.000916
Mean PPL(Q)-PPL(base) : 0.478433 ± 0.023588
====== KL divergence statistics ======
Mean KLD: 0.027115 ± 0.000104
Maximum KLD: 2.016436
99.9% KLD: 0.391155
99.0% KLD: 0.179995
99.0% KLD: 0.179995
Median KLD: 0.015400
10.0% KLD: 0.000310
5.0% KLD: 0.000046
1.0% KLD: 0.000000
Minimum KLD: -0.000275
====== Token probability statistics ======
Mean Δp: -0.073 ± 0.010 %
Maximum Δp: 66.267%
99.9% Δp: 25.309%
99.0% Δp: 12.706%
95.0% Δp: 5.590%
90.0% Δp: 2.959%
75.0% Δp: 0.362%
Median Δp: -0.000%
25.0% Δp: -0.466%
10.0% Δp: -3.166%
5.0% Δp: -5.937%
1.0% Δp: -13.474%
0.1% Δp: -26.235%
Minimum Δp: -66.500%
RMS Δp : 4.030 ± 0.024 %
Same top p: 91.745 ± 0.071 %