nodes/silveroxides-ComfyUI-ModelUtils
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

add new feature to resize LoRA by first adding it to base model

Browse Source
This commit is contained in:
silveroxides
2026-01-12 01:21:19 +01:00
parent 0726ec952b
commit ede6fde5f3
4 changed files with 1042 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/DEVELOPMENT.md
+25
View File
@@ -294,3 +294,28 @@ ComfyUI-ModelUtils/
| Metadata reading | **100x+** (header only) |
| Parallel I/O | **2-4x** (ThreadPoolExecutor) |
| Memory management | Auto-unload models before heavy ops |
| **LoRA Resize Fast** | **~91x** (merge-to-base extraction) |
---
## Phase 13: Fast LoRA Resize (2026-01-11)
### Summary
Added fast LoRA resize via base model extraction path, achieving ~91x speedup.
### Changes
- Added `resize_lora_via_base()` function using extraction code path
- Added 4 new fast resize nodes requiring base model input
### Files Modified
| File | Changes |
|------|---------|
| `nodes/lora_resize.py` | Added `resize_lora_via_base()`, 4 new node classes |
| `__init__.py` | Registered new nodes |
### New Nodes
- `LoRAResizeViaBaseFixed` - Fast fixed rank resize
- `LoRAResizeViaBaseRatio` - Fast SV ratio resize
- `LoRAResizeViaBaseFrobenius` - Fast Frobenius resize
- `LoRAResizeViaBaseCumulative` - Fast cumulative resize