Learned-rounding INT8 ConvRot quantization of
Kroma v0.2 Turbo, a Krea.2-based diffusion model by
Lodestones. The 25.64 GB BF16 checkpoint is reduced to approximately
13.49 GB.
The image input and final output projections, timestep embedding and modulation paths, text projection, and text-fusion transformer remain BF16. Norms and biases also retain their original precision. This protects the shared conditioning and boundary paths while quantizing the large transformer core.
1ctq -i <input-model>.safetensors -o kroma-v0.2-turbo_int8_convrot_learned.safetensors `
2 --krea2 `
3 --int8 --scaling_mode row `
4 --convrot --convrot-group-size 256 `
5 --comfy_quant --save-quant-metadata `
6 --low-memory --device cuda
Quantization is lossy, so outputs are not bit-identical to the original BF16 checkpoint.
