Kroma Pruned INT8 Experiments

( Not my Model )


 <a href="https://civitai.red/models/2957000/kroma-pruned-int8-experiments?modelVersionId=3349618" rel="noopener noreferrer nofollow" target="_blank">
 https://civitai.red/models/2957000/kroma-pruned-int8-experiments?modelVersionId=3349618
 </a>




 Experimental structurally pruned releases of
 
 Kroma v0.3 Turbo by lodestones
 
 . I was just curious as to how the removal of pieces of architecture would affect the model, so just experimental models with a guarantee of quality loss. Once again, my goal is to reduce peak VRAM/RAM usage, especially for lower end hardware.




 
 Pruned v1
 


 Structural changes:

<ul>
 <li>
 
 Removed 1 full transformer block: original block 2
 
 </li>
 <li>
 
 Main block count reduced from 28 → 27
 
 </li>
 <li>
 
 Pruned 1,536 MLP channels from every remaining block
 
 </li>
 <li>
 
 MLP width reduced from 16,384 → 14,848
 
 </li>
</ul>

 The pruning is physical: the removed block and MLP channels are no longer stored in the checkpoint rather than simply being bypassed.