<h2 id="flux-depth-fp8-(blockwise)-irevwwmhc">FLUX Depth FP8 (Blockwise)</h2><ul><li>FP8 Conversion of Flux Dev Depth</li><li>Load Cover image for workflow, Example Depth image does not have workflow.</li><li>20-40% speed up over full model for older cards or if loading TE's into GPU</li></ul>Note that this model is in FP8 E4M3 it is possible that a scaled conversion with mixed E5M2 and E4M3 would be beneficial but based off of the info from Blackforest I stuck with the higher precision.BLOCKWISEThe following blocks where not converted"time_<a target="_blank" rel="ugc" href="http://in.in">in.in</a>_layer.bias","time_<a target="_blank" rel="ugc" href="http://in.in">in.in</a>_layer.weight","time_in.out_layer.bias","time_in.out_layer.weight","txt_in.bias","txt_in.weight","vector_<a target="_blank" rel="ugc" href="http://in.in">in.in</a>_layer.bias","vector_<a target="_blank" rel="ugc" href="http://in.in">in.in</a>_layer.weight","vector_in.out_layer.bias","vector_in.out_layer.weight","final_layer.adaLN_modulation.1.bias","final_layer.adaLN_modulation.1.weight","final_layer.linear.bias","final_layer.linear.weight","guidance_<a target="_blank" rel="ugc" href="http://in.in">in.in</a>_layer.bias","guidance_<a target="_blank" rel="ugc" href="http://in.in">in.in</a>_layer.weight","guidance_in.out_layer.bias","guidance_in.out_layer.weight","img_in.bias", "img_in.weight"

Files with SHA256: a2f8c58ccb1060be795186e3cd45f368d45bd938d42717a463e3b7a8099a8c51