CalamitousFelicitousness 9b37a1535c fix(lora): exact lora application on sdnq-quantized layers ...

Baking a lora into a quantized weight requantizes it, and on low-bit
formats round-to-nearest erases sub-step deltas (uint4 retains roughly
2/group_size of the signal). Plain lora deltas now ride the sdnq svd
side-channel: factors append to svd_up/svd_down with the down factor
hadamard-rotated, applied by the dequantizer at full precision in every
forward mode. Apply and remove are exact and take no weight backup.

- non-factorable families (dora, lokr, loha, oft, cp mid, dense bias)
  fall back to requantize with a per-pass summary warning
- native fuse now honors the quantized-model guard; fuse requantized in
  place on every network swap and accumulated drift
- layers that fell back on a mixed set restore from backup before
  re-entering the factor path; untargeted quantized layers are no
  longer flagged
- test/test-sdnq-lora-factors.py pins the erasure law, factor-path
  exactness, memory accounting and set transitions

2026-08-25 00:25:22 +01:00

..

__init__.py

linting

2026-07-03 22:05:33 +02:00

extra_networks_lora.py

fix(lora): exact lora application on sdnq-quantized layers

2026-08-25 00:25:22 +01:00

lora_apply.py

fix(lora): exact lora application on sdnq-quantized layers

2026-08-25 00:25:22 +01:00

lora_common.py

modernize typing

2026-02-19 09:15:37 +01:00

lora_convert.py

fix(lora): handle flattened clip text model in kohya te keys

2026-07-08 03:05:30 +01:00

lora_diffusers.py

fix(lora): exact lora application on sdnq-quantized layers

2026-08-25 00:25:22 +01:00

lora_extract.py

fix(lora): handle flattened clip text model in kohya te keys

2026-07-08 03:05:30 +01:00

lora_load.py

fix(lora): exact lora application on sdnq-quantized layers

2026-08-25 00:25:22 +01:00

lora_nunchaku.py

cleanup logger

2026-02-19 11:09:13 +01:00

lora_overrides.py

fix(lora): exact lora application on sdnq-quantized layers

2026-08-25 00:25:22 +01:00

lora_sdnq.py

fix(lora): exact lora application on sdnq-quantized layers

2026-08-25 00:25:22 +01:00

lora_timers.py

modernize typing

2026-02-19 09:15:37 +01:00

lyco_helpers.py

full codespell coverage

2026-06-04 12:36:10 +02:00

native_adapter.py

fix(lora): refuse a network whose deltas do not fit the model

2026-08-21 02:12:16 +01:00

network_boft.py

feat(flux2): native BOFT (butterfly-OFT) support

2026-05-10 00:05:37 +01:00

network_full.py

Major lora refactor: works on my machine edition

2024-11-23 21:57:03 -06:00

network_glora.py

fix(lora): set GLora dim so alpha scaling is applied

2026-05-09 22:04:42 +01:00

network_hada.py

feat(lora): add NetworkModuleHadaChunk for fused-weight LoHA targets

2026-05-09 23:16:09 +01:00

network_ia3.py

Major lora refactor: works on my machine edition

2024-11-23 21:57:03 -06:00

network_lokr.py

refactor(lora): move NetworkModuleLokrSliceChunk to network_lokr

2026-05-18 23:18:31 +01:00

network_lora.py

perf(lora): skip the discarded init when building lora modules

2026-08-10 02:42:05 +01:00

network_norm.py

lint fixes

2024-11-29 08:01:26 -05:00

network_oft.py

fix(lora): guard OFT constraint for LyCORIS oft_diag files

2026-05-09 22:04:17 +01:00

network.py

fix(lora): refuse a network whose deltas do not fit the model

2026-08-21 02:12:16 +01:00

networks.py

fix(lora): exact lora application on sdnq-quantized layers

2026-08-25 00:25:22 +01:00