This is an Extension for Forge Neo, which quantizes / casts models into different format / dtype to reduce memory usage and save disk space
Tip
- Use the Cast section to cast an arbitrary model (e.g. ControlNet) into
dtype - Use the Quantize section to quantize a diffusion model into
format
Note
Supported Formats: fp8_scaled / nvfp4 / mxfp8 / int8_tensorwise / convrot_w4a4 / asym_w4a8_int8 / w6a8_int8
Important
Supported Inputs: fp32 / fp16 / bf16 / fp8_scaled
Warning
This Extension does not include per-model config ; quality may not match other dedicated tools