Back to Powerinfer

perplexity

examples/perplexity/README.md

latest524 B
Original Source

perplexity

TODO

Llama 2 70B Scorechart

QuantizationModel size (GiB)PerplexityDelta to fp16
Q4_036.203.55503.61%
Q4_140.203.51252.37%
Q5_044.203.47441.26%
Q2_K27.273.73398.82%
Q3_K_S27.863.70197.89%
Q3_K_M30.833.59324.72%
Q3_K_L33.673.56173.80%
Q4_K_S36.393.48521.57%
Q4_K_M38.543.47251.20%
Q5_K_S44.203.44830.50%
Q5_K_M45.413.44510.40%
Q6_K52.703.43670.16%
fp16128.53.4313-