Add beta4 INT8 ConvRot skip-edges
#4
by cicalooo - opened
Upload 10Eros_Max_h3_TURBO-hybrid_beta4_int8_convrot_skip_edges.safetensors: native ComfyUI int8_tensorwise + ConvRot (groupsize 256). Blocks 0, 1, 48, 49 remain BF16; blocks 2-47 quantized (184 projections). Source: 10Eros_Max_h3_TURBO-hybrid_beta4.safetensors.
TenStrip changed pull request status to merged