Add beta4 INT8 ConvRot (full, no skip-edges)
#3
by cicalooo - opened
Upload 10Eros_Max_h3_TURBO-hybrid_beta4_int8_convrot.safetensors: native ComfyUI int8_tensorwise + ConvRot (groupsize 256), all transformer blocks 0-49 quantized (200 projections). Source: 10Eros_Max_h3_TURBO-hybrid_beta4.safetensors. No skip-edges.
TenStrip changed pull request status to merged