List your company

Language Modeling on WikiText-103 (PPL, Bits, and Efficiency Delta)

446.8Perplexity (PPL)fp16
323.5521,155.4761,987.42,819.324May 26, 2026

Evaluation results

MethodLinks
fp16Precision=fp162026.05446.8160
HQMQQuantization Setting=s96_r6 + Med3×2026.05460.45.2513.6
HQMQQuantization Setting=s192_r42026.05469.54.0422.7
HQMQQuantization Setting=s24_r6 + Med3×2026.05542.24.7895.4
HQMQQuantization Setting=s96_r42026.05553.83.79107
HQMQQuantization Setting=s48_r42026.05688.13.54241.3
HQMQQuantization Setting=s24_r32026.051,0393.04593
NaivePrecision=int42026.051,66641,219
NaivePrecision=int32026.053,52833,081