ComfyUI Node
Mpi LoRA Extract (Files, Low Memory)
Extracts a LoRA from two model FILES one layer at a time (low memory, no GPU needed) and saves it under output/. A scaled-FP8 fine-tune is dequantized first.
Mpi LoRA Extract (Files, Low Memory)
◄finetuned_name▾►
◄original_name▾►
◄filename_prefixloras/extracted_lora►
◄rank64►
◄algorithmsvd_lowrank►
◄lowrank_iters7►
◄output_dtypefp16►
◄bias_difftrue►
◄clamp_quantilefalse►
CategoryMpiNodes/KJ
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| finetuned_name | COMBO | The fine-tuned model (diffusion_models or checkpoints). FP16, BF16 or scaled FP8. | |
| original_name | COMBO | The original base model it was fine-tuned from. | |
| filename_prefix | STRING | loras/extracted_lora | — |
| rank | INT | 641–4096 | — |
| algorithm | COMBO | svd_lowrank | SVD algorithm; svd_lowrank is faster but less accurate. |
| lowrank_iters | INT | 71–100 | — |
| output_dtype | COMBO | fp16 | 3 options: fp16, bf16, fp32 |
| bias_diff | BOOLEAN | true | — |
| clamp_quantile | BOOLEAN | false | — |
Outputs (0)
No outputs