kompress-v17 β€” 5x Must-Keep Weight

v8 data + 5x must-keep loss weight. Heretic 0.963 β€” beats v8's 0.955 but only 3.7% compression (v8: 15%). Middle ground in the weight tradeoff experiment. Use kompress-v8 for production.

Weight Model Heretic Compression
3x v8 0.955 15%
5x v17 0.963 3.7%
10x v16 0.972 2.8%

CONCLUSION

Ξ»=5 gives 0.963 heretic but 3.7% compression. +0.008 precision costs -11.3% compression vs v8.

USECASE

Tradeoff reference. Shows Ξ»=5 is not worth the compression loss.

Full benchmark β†’ | LoopKit β†’

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for PeetPedro/kompress-v17

Quantized
(9)
this model

Space using PeetPedro/kompress-v17 1