4 Commits

Author SHA1 Message Date
CalamitousFelicitousness b8cf912e4e feat(lora): run the full adapter family chain on every native arch
zimage, chroma, ernie and krea2 chained only lora/lokr/loha/oft while
flux2 and anima ran all eight families, so ia3/glora/norm/full files
(e.g. full-diff extractions with diff/diff_b keys) reported not loaded
on the short-chain arches. The generic family loaders are arch-agnostic;
wire the missing four into each chain.

- add ia3/glora/norm/full wrappers and chain entries to the four arch
  modules, with matching suffix/marker re-exports
- add a chain-level full-diff test per suite: zimage covers the legacy
  attention.out alias and the fused-qkv skip, chroma the proj rename,
  ernie the passthrough
2026-07-14 08:06:45 +01:00
CalamitousFelicitousness 2d47b1f761 test(lora): cover OneTrainer lora_transformer_ format for native loaders
Add an end-to-end regression per diffusers arch (chroma, flux2, zimage, ernie)
that loads a lora_transformer_ diffusers-flat state dict and checks the
expected modules bind. The chroma suite also asserts the shared
resolve_group_targets passthrough runs above each arch's own resolve_targets,
which returns nothing for this prefix.
2026-06-14 18:22:41 +01:00
CalamitousFelicitousness 6937ea803f refactor(ernie): migrate to generic native_loader
Replaces ernie's four family loaders with thin wrappers binding
native_loader's generics to ernie's prefix tuples and resolve_targets.

ErnieImageAttention has fully split to_q / to_k / to_v with no fused QKV
and ErnieImageFeedForward has three separate Linear modules, so
resolve_targets is a straight passthrough across every recognized prefix.

BARE_DIFFUSERS_PREFIXES covers layers., adaLN_modulation., final_norm.,
final_linear. for bare-diffusers exports (e.g. via save_lora_adapter).

parse_key returns (prefix_used, base, suffix) instead of the old
(network_key, suffix); parse test updated.
2026-05-18 23:34:41 +01:00
CalamitousFelicitousness c7e8e7a029 test(ernie): offline tests for native adapter loaders
Covers ernie's four-family surface (LoRA, LoKR, LoHA, OFT).

ErnieImageAttention has fully split to_q / to_k / to_v with no fused QKV,
and ErnieImageFeedForward has three separate Linear modules. The loader
has no chunking, renames, or fused-target dispatch.

Mock matches diffusers.ErnieImageTransformer2DModel
(ErnieImageSharedAdaLNBlock with self_attention + mlp + RMSNorms, plus
module-level adaLN_modulation Sequential and final_linear).

Formats exercised:

- BFL / AI-toolkit
- kohya
- BFL LoKR
2026-05-18 23:19:46 +01:00