silveroxides
9f8b337584
changes to LoRAs
2026-02-07 16:02:40 +01:00
silveroxides
6a119e63f5
Bump version from 0.2.6 to 0.2.7
2026-01-23 19:21:02 +01:00
silveroxides
4097326a82
Add pad/crop for mismatched tensors and interpolation mode for weight redistribution across rank
2026-01-23 19:04:51 +01:00
silveroxides
31eee604de
Bump version from 0.2.5 to 0.2.6
2026-01-23 18:54:09 +01:00
silveroxides
2bfecf9c2d
Fix resize_tensors to handle cases where model B is larger than model A
2026-01-23 18:53:40 +01:00
silveroxides
1ba0932d6b
Bump version from 0.2.4 to 0.2.5
2026-01-23 18:42:23 +01:00
silveroxides
b1ec55ff42
improve and correct merger operations based on papers
2026-01-23 18:41:35 +01:00
silveroxides
8deb4fd739
perf: Add low_memory mode to reduce RAM usage in LoRA merge
...
- MemoryEfficientSafeOpen now supports low_memory=True which disables mmap
- Uses direct file reads with readinto() to avoid OS page caching
- _deserialize_tensor avoids bytearray copy when input is already bytearray
- merge_loras_to_model now uses low_memory=True for base model
This should reduce peak RAM from 2x model size to 1x + overhead
2026-01-15 03:13:51 +01:00
silveroxides
e9f7cda706
fix: Free cpu_base after storing to avoid duplicate RAM usage
2026-01-15 02:51:17 +01:00
silveroxides
cffb92aa06
fix: Add HuggingFace LoRA format support (.lora.down.weight/.lora.up.weight)
...
- Add detection for HuggingFace format in detect_lora_format()
- Strip trailing .lora suffix in extract_core_layer_lora() for proper key matching
- Applied fix to resize_lora_via_base, merge_multi_loras_via_base, and merge_loras_to_model
2026-01-15 02:43:42 +01:00
silveroxides
ce8c6aab97
feat: Add LoRAMergeToModel node for merging LoRAs directly into model
...
- New merge_loras_to_model() function that merges LoRAs into base model
weights and saves the result without extraction
- New LoRAMergeToModel node with simplified inputs (no extraction params)
- Outputs merged model to the base model's directory
2026-01-15 02:30:51 +01:00
silveroxides
6f3fd0900f
add support for full diff LoRA in multi LoRA merge
2026-01-12 02:36:56 +01:00
Silver
eaed2003a0
Bump version from 0.2.3 to 0.2.4
2026-01-12 01:23:09 +01:00
Silver
78192c594a
Merge pull request #5 from silveroxides/feature/resize-with-base
...
add new LoRA features and up to 4 way LoRA resize+merge node.
2026-01-12 01:22:46 +01:00
silveroxides
ede6fde5f3
add new feature to resize LoRA by first adding it to base model
2026-01-12 01:21:19 +01:00
Silver
0726ec952b
Update pyproject.toml
2026-01-09 23:02:30 +01:00
silveroxides
c6b65d6721
fix: frobenius/cumulative norm calculation now relative to max_rank
2026-01-03 20:55:27 +01:00
Silver
32281ad2d5
Bump version from 0.2.0 to 0.2.1
2026-01-03 20:38:38 +01:00
Silver
68fab8d3d0
Merge pull request #4 from silveroxides/feature/lora-resize
...
Feature/lora resize
2026-01-03 20:38:13 +01:00
silveroxides
61b2f506d0
feat: add svd_niter parameter for SVD accuracy control
2026-01-03 20:27:56 +01:00
silveroxides
75d1f0f5c9
perf: use svd_lowrank for fixed rank resize (10x faster)
2026-01-03 20:14:25 +01:00
silveroxides
3a71ad9952
feat: add max_rank inputs to dynamic extraction nodes
2026-01-03 20:12:58 +01:00
silveroxides
37090dd7ba
feat: add LoRA resize nodes and fix sv_ratio convention
2026-01-03 20:04:56 +01:00
silveroxides
f58289d042
document development phases
2026-01-03 19:37:39 +01:00
Silver
1796f1bea9
Merge pull request #3 from silveroxides/feature/lora-extract
...
Feature/lora extract and version bumped to 0.2.0
2026-01-02 17:01:41 +01:00
Silver
3df073f810
Bump version from 0.1.7 to 0.2.0
2026-01-02 17:01:02 +01:00
silveroxides
0e73c5d889
Update README.md, add images and workflows
2026-01-02 16:59:25 +01:00
silveroxides
eef7377061
Added .contiguous() to saves
2026-01-01 21:18:08 +01:00
silveroxides
65c1ae37b7
add svd_lowrank for fixed rank extraction and put experimental chunk large layers default to false
2026-01-01 21:08:15 +01:00
silveroxides
b7707862b2
Add pinned memory for faster merging and extraction
2026-01-01 20:49:49 +01:00
silveroxides
6d7b2d1403
refactor nodes and implement memory management
2026-01-01 20:29:08 +01:00
silveroxides
bfbbcd9515
fix exception handler
2026-01-01 19:04:25 +01:00
silveroxides
18679cca2c
Split lycoris extraction modes into different nodes
2026-01-01 07:21:37 +01:00
silveroxides
da13f9f51d
add lycoris diff extraction
2026-01-01 07:06:19 +01:00
silveroxides
92d3c5582c
add lora extraction, progress bars and more
2026-01-01 06:36:59 +01:00
Silver
bb923e8474
Merge pull request #2 from silveroxides/feature/v3-migration
...
migrate to v3 schema and add LoRA extraction node
2026-01-01 06:35:56 +01:00
silveroxides
a3116cbb11
migrate to v3 schema and add LoRA extraction node
2026-01-01 05:43:24 +01:00
silveroxides
b6fae8d0d9
Bump version to 0.1.7
2025-12-21 20:32:19 +01:00
silveroxides
de27186b90
merge feature mismatch-handling
2025-12-21 20:31:17 +01:00
silveroxides
1131811117
add layer filtering
2025-12-18 22:23:33 +01:00
silveroxides
c5df75dce8
fix handling of mismatched layers
2025-12-18 22:04:04 +01:00
silveroxides
19bc0472dc
add handling of mismatched layers
2025-12-18 21:43:32 +01:00
silveroxides
6c08202ef5
update .gitignore
2025-12-18 21:22:10 +01:00
silveroxides
68f22fa87b
fix Extract namespace conflict
2025-11-24 19:16:03 +01:00
silveroxides
5ea63abfbe
fix PowerUp namespace conflict
2025-11-24 19:13:45 +01:00
silveroxides
232dc76db2
refactor names to be more universal
2025-11-24 18:37:18 +01:00
silveroxides
31989beefd
Bump version to 0.1.6
2025-11-09 06:57:51 +01:00
silveroxides
0d035ef072
Add embedding nodes
2025-11-09 06:57:23 +01:00
silveroxides
78b3198c14
Bump version to 0.1.5
2025-10-19 10:20:26 +02:00
silveroxides
79593701aa
Add model merging nodes
2025-10-19 10:19:52 +02:00
silveroxides