Will perform strided perplexity calculation -> adjusting context size from 3072 to 3264
llama_model_loader: loaded meta data with 39 key-value pairs and 363 tensors from granite-4.1-8b-Q7_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:   81 tensors
llama_model_loader: - type q5_1:    4 tensors
llama_model_loader: - type q8_0:  146 tensors
llama_model_loader: - type q6_K:  132 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 7.68 GiB (7.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   :   8.751241 ±   0.066500
Mean PPL(base)                :   8.691178 ±   0.065443
Cor(ln(PPL(Q)), ln(PPL(base))):  99.82%
Mean ln(PPL(Q)/PPL(base))     :   0.006887 ±   0.000464
Mean PPL(Q)/PPL(base)         :   1.006911 ±   0.000467
Mean PPL(Q)-PPL(base)         :   0.060063 ±   0.004141

====== KL divergence statistics ======
Mean    KLD:   0.003568 ±   0.000040
Maximum KLD:   2.888946
99.9%   KLD:   0.140997
99.0%   KLD:   0.037132
95.0%   KLD:   0.011608
90.0%   KLD:   0.006961
Median  KLD:   0.001456
10.0%   KLD:   0.000015
 5.0%   KLD:   0.000003
 1.0%   KLD:  -0.000000
 0.1%   KLD:  -0.000004
Minimum KLD:  -0.000012

====== Token probability statistics ======
Mean    Δp: -0.007 ± 0.005 %
Maximum Δp: 81.280%
99.9%   Δp: 12.574%
99.0%   Δp:  4.968%
95.0%   Δp:  2.242%
90.0%   Δp:  1.259%
75.0%   Δp:  0.222%
Median  Δp: -0.000%
25.0%   Δp: -0.211%
10.0%   Δp: -1.245%
 5.0%   Δp: -2.232%
 1.0%   Δp: -5.277%
 0.1%   Δp: -14.266%
Minimum Δp: -57.371%
RMS Δp    :  1.812 ± 0.025 %
Same top p: 97.235 ± 0.043 %