fix(model cache): keep patch models alive while they are patched in - #9397

Closed

lstein wants to merge 76 commits into

invoke-ai:mainfrom

lstein:lstein/fix/lora-unlocked-use

Commits on Jun 1, 2026

Commits on Jun 2, 2026

fix(model-manager): serialize model construction against VRAM moves to prevent meta-device corruption

lstein
and
claude
committed
a119b50
fix(backend): fix outpainting crash caused by model download collisions
lstein
committed
7011446
fix(backend): make DiskImageFileStorage thread-safe for parallel sessions

lstein
and
claude
committed
a1fe375
feat(ui): stack per-session progress bars during parallel generation

lstein
and
claude
committed
3db88d5
fix(ui): make $progressEvents module-local to satisfy knip

lstein
and
claude
committed
351758c
fix(ui): cap stacked tab progress bars to fit below the tab label

lstein
and
claude
committed
4f6613f
feat(config): support "auto" generation_devices to use all GPUs by default

lstein
and
claude
committed
2a65c4a

Commits on Jun 3, 2026

chore(frontend): typegen+openapi
lstein
committed
4209780
docs(multi-gpu): add configuration information
lstein
committed
914c577
chore(frontend): typegen + openapi again
lstein
committed
a928a75
Merge branch 'main' into lstein/feat/multi-gpu
lstein
authored
5e4e864
feat(settings): add Generation Devices selector to Settings dialog

lstein
and
claude
committed
48458c1
feat(settings): boldface the restart notice on Generation Devices

lstein
and
claude
committed
5100605
feat(settings): show GPU name in Generation Devices badges

lstein
and
claude
committed
200b3a3
chore(frontend): openapi
lstein
committed
cdcf7df

Commits on Jun 5, 2026

Merge branch 'main' into lstein/feat/multi-gpu
lstein
authored
d521ba4

Commits on Jun 12, 2026

feat(multi-gpu): surface per-session GPU number in logs and UI

lstein
and
claude
committed
8cef3cf
Merge remote-tracking branch 'upstream/main' into lstein/feat/multi-gpu
lstein
committed
3353e5a

Commits on Jun 23, 2026

Merge branch 'main' into lstein/feat/multi-gpu
lstein
authored
c48dfc8

Commits on Jun 24, 2026

Merge branch 'main' into lstein/feat/multi-gpu

lstein
and
claude
committed
c300a16

Commits on Jun 25, 2026

feat(multi-gpu): show per-device names in startup log and progress circles

lstein
and
claude
committed
1f55a4f
Merge branch 'main' into lstein/feat/multi-gpu
JPPhoto
authored
9c1e516

Commits on Jun 26, 2026

feat(model-cache): share one CPU copy of model weights across per-GPU caches

lstein
and
claude
committed
a1fde96
fix(session-queue): cancel all in-progress items in bulk-cancel APIs (multi-GPU)

lstein
and
claude
committed
fbd95a8
fix(multi-gpu): address review findings (cancel race, bulk delete, device guards, refcount leak)

lstein
and
claude
committed
ed2a330
fix(ci): ruff format + make CPU-incompatible device test mock CUDA

lstein
and
claude
committed
57e1e79
fix(multi-gpu): stop RAM blowup/swapping during concurrent generations

lstein
and
claude
committed
2d3802a

Commits on Jun 27, 2026

fix(qwen-image): reserve VAE working memory so decode/encode don't OOM

lstein
and
claude
committed
6903911
fix(flux2): tile reference-image VAE encode to avoid VRAM OOM

lstein
and
claude
committed
3e917dd
fix(multi-gpu): query execution device for VRAM-in-use accounting

lstein
and
claude
committed
902c77d
fix(qwen-image): calibrate VAE working-memory estimate to the 3D-conv decode peak

lstein
and
claude
committed
aa6fec8
feat(qwen-image): honor force_tiled_decode in the l2i node

lstein
and
claude
committed
43a46bd
fix(ui): stop progress disk flashing during indeterminate phases

lstein
and
claude
committed
b84d450

Commits on Jun 28, 2026

Merge branch 'main' into lstein/feat/multi-gpu
lstein
committed
b275c38

Commits on Jun 29, 2026

feat(multi-gpu): offload text encoders to idle GPUs

lstein
and
claude
committed
0037a21
fix(multi-gpu): adopt GGUF weights across devices to stop RAM spikes

lstein
and
claude
committed
b61a8e1
fix(multi-gpu): tie device #N label to cuda index, not filtered position

lstein
and
claude
committed
163cbb2
feat(multi-gpu): flash restart reminder when generation devices change

lstein
and
claude
committed
04e2071

Commits on Jun 30, 2026

Merge branch 'main' into lstein/feat/multi-gpu
lstein
committed
f9e6fba

Commits on Jul 1, 2026

Merge branch 'main' into lstein/feat/multi-gpu

lstein
and
claude
committed
1145318

Commits on Jul 6, 2026

Commits on Jul 10, 2026

Merge branch 'main' into lstein/feat/multi-gpu
JPPhoto
authored
025ce99

Commits on Jul 12, 2026

Merge branch 'main' into lstein/feat/multi-gpu
JPPhoto
authored
b289580
fix: address multi-GPU review findings from PR #9263 review

lstein
and
claude
committed
aee12c3

Commits on Jul 15, 2026

Commits on Jul 19, 2026

Merge remote-tracking branch 'origin/main' into lstein/feat/multi-gpu
lstein
committed
eedc43b

Commits on Jul 20, 2026

chore(frontend): lint:prettier
lstein
committed
998f3ed

Commits on Jul 21, 2026

Merge branch 'main' into lstein/feat/multi-gpu
JPPhoto
authored
d44e0b3

Commits on Jul 22, 2026

Commits on Jul 23, 2026

fix(backend): avoid MODEL_LOAD_LOCK self-deadlock when patching a LoRA on a cold cache

lstein
and
claude
committed
3f78434
fix: address JPPhoto's four merge blockers from the 2026-07-22 review

lstein
and
claude
committed
68edb02
chore: regenerate OpenAPI schema for the clear endpoint docstring

lstein
and
claude
committed
4545c10

Commits on Jul 24, 2026

Merge branch 'main' into lstein/feat/multi-gpu
lstein
authored
45141ae

Commits on Jul 27, 2026

fix(model cache): close lost-wakeup race in deferred RAM-budget reconcile

lstein
and
claude
committed
a3f7f49
fix(model cache): close remaining RAM-budget reconcile gaps

lstein
and
claude
committed
cef6c8c
Merge branch 'main' into lstein/feat/multi-gpu
JPPhoto
authored
13c8e51

Commits on Jul 28, 2026

Merge branch 'main' into lstein/feat/multi-gpu
lstein
authored
e19008e
fix(model cache): bound the admission grace and keep cached_model_keys stall-free

lstein
and
claude
committed
1282be4
Merge branch 'main' into lstein/feat/multi-gpu

lstein
and
claude
committed
4ce0a78
fix(model cache): release abandoned admission grace
JPPhoto
committed
7ffee4d
fix(model cache): keep grace release off the collecting thread

lstein
and
claude
committed
55a6fc4
fix(model cache): harden the deferred grace release

lstein
and
claude
committed
c7dddb6
fix(model cache): keep patch models alive while they are patched in

lstein
and
claude
committed
2bded03