ComfyUI Node
Lora Extract K Jv2 (QQ)
A ComfyUI node in KJNodes/lora with 17 inputs and 0 outputs.
Lora Extract K Jv2 (QQ)
- finetuned_model
- original_model
◄filename_prefixloras/ComfyUI_v2_extracted_lora►
◄rank8►
◄lora_type▾►
◄algorithmsvd_linalg►
◄lowrank_iters7►
◄output_dtypefp16►
◄bias_difftrue►
◄adaptive_param0.15►
◄clamp_quantiletrue►
◄vram_usage_modebalanced►
◄blocks_in_gpu10►
◄batch_size4►
◄prefetch_ahead2►
◄memory_threshold0.85►
◄debug_modefalse►
CategoryKJNodes/lora
Inputs (17)
| Name | Type | Default | Description |
|---|---|---|---|
| finetuned_model | MODEL | — | |
| original_model | MODEL | — | |
| filename_prefix | STRING | loras/ComfyUI_v2_extracted_lora | — |
| rank | INT | 81–4096 | The rank to use for standard LoRA, or maximum rank limit for adaptive methods. |
| lora_type | COMBO | 6 options: standard, full, adaptive_ratio, adaptive_quantile, adaptive_energy, adaptive_fro | |
| algorithm | COMBO | svd_linalg | SVD algorithm to use, svd_lowrank is faster but less accurate. |
| lowrank_iters | INT | 71–100 | The number of subspace iterations for lowrank SVD algorithm. |
| output_dtype | COMBO | fp16 | 3 options: fp16, bf16, fp32 |
| bias_diff | BOOLEAN | true | — |
| adaptive_param | FLOAT | 0.150–1 | For ratio mode, this is the ratio of the maximum singular value. For quantile mode, this is the quantile of the singular values. For fro mode, this is the Frobenius norm retention ratio. |
| clamp_quantile | BOOLEAN | true | — |
| vram_usage_mode | COMBO | balanced | VRAM usage mode. Conservative=50% VRAM, Balanced=75% VRAM, Aggressive=90%+ VRAM |
| blocks_in_gpu | INT | 100–40 | Number of transformer blocks to keep in GPU memory (0=offload all, 40=keep all) |
| batch_size | INT | 41–8 | Number of layers to process simultaneously (higher=more VRAM, faster) |
| prefetch_ahead | INT | 20–10 | Number of blocks to prefetch ahead (reduces I/O bottlenecks) |
| memory_threshold | FLOAT | 0.850.5–0.95 | VRAM usage threshold before automatic offloading (0.5-0.95) |
| debug_mode | BOOLEAN | false | Enable detailed performance and VRAM usage debugging |
Outputs (0)
No outputs