DiffusionAi/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI.git synced 2025-01-08 11:57:36 +08:00
Code Issues Packages Projects Releases Wiki Activity
15,060 Commits 335 Branches 286 Tags 409 MiB
ryan/model-offload-prototype-lora
Commit Graph

15060 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Ryan Dick
f6045682c0 Fix bug with partial offload of model buffers. 2024-12-10 22:19:17 +00:00
Ryan Dick
84a75ddb72 Fix bug in ModelCache that was causing it to offload more models from VRAM than necessary. 2024-12-10 20:38:37 +00:00
Ryan Dick
a9fb1c82a0 Fix handling of torch.nn.Module buffers in CachedModelWithPartialLoad. 2024-12-10 19:38:04 +00:00
Ryan Dick
cc7391e630 Enable LoRAPatcher.apply_smart_lora_patches(...) throughout the stack. 2024-12-10 17:27:33 +00:00
Ryan Dick
62d595f695 (minor) Rename num_layers -> num_loras in unit tests. 2024-12-10 16:41:52 +00:00
Ryan Dick
5e2080266e Add test_apply_smart_lora_patches_to_partially_loaded_model(...). 2024-12-10 16:38:48 +00:00
Ryan Dick
ed7bb7ea3d Add LoRAPatcher.smart_apply_lora_patches() 2024-12-10 16:26:34 +00:00
Ryan Dick
62407f7c6b Refactor LoRAPatcher slightly in preparation for a 'smart' patcher. 2024-12-10 15:36:36 +00:00
Ryan Dick
80128e1e14 Fix LoRAPatcher.apply_lora_wrapper_patches(...) 2024-12-10 03:10:23 +00:00
Ryan Dick
4c84d39e7d Finish consolidating LoRA sidecar wrapper implementations. 2024-12-10 02:54:32 +00:00
Ryan Dick
0c4a368555 Begin to consolidate the LoRA sidecar and LoRA layer wrapper implementations. 2024-12-10 01:16:01 +00:00
Ryan Dick
55dc762a91 Fix bias handling in LoRAModuleWrapper and add unit test that checks that all LoRA patching methods produce the same outputs. 2024-12-09 16:59:37 +00:00
Ryan Dick
d825d3856e Add LoRA wrapper patching to LoRAPatcher. 2024-12-09 16:35:23 +00:00
Ryan Dick
d94733f55a Add LoRA wrapper layer. 2024-12-09 15:17:50 +00:00
Ryan Dick
2144d21f80 Maintain a read-only CPU state dict copy in CachedModelWithPartialLoad. 2024-12-06 21:49:24 +00:00
Ryan Dick
958efa19d7 Memoize frequently accessed values in CachedModelWithPartialLoad. 2024-12-06 20:39:05 +00:00
Ryan Dick
11af57def3 More ModelCache logging improvements. 2024-12-06 18:38:36 +00:00
Ryan Dick
8b70a5b9bd Cleanup of ModelCache and added a bunch of debug logging. 2024-12-06 17:39:16 +00:00
Ryan Dick
5d9fdcd78d Fix a couple of bugs to get basic vanilla partial model load working with the model cache. 2024-12-06 00:50:58 +00:00
Ryan Dick
c7b84cf012 WIP - first pass at overhauling ModelCache to work with partial loads. 2024-12-05 23:03:40 +00:00
Ryan Dick
8e409e3436 Delete experimental torch device autocasting solutions and clean up TorchFunctionAutocastDeviceContext. 2024-12-05 19:36:44 +00:00
Ryan Dick
987393853c Create CachedModelOnlyFullLoad class. 2024-12-05 18:43:50 +00:00
Ryan Dick
91c5af1b95 Move CachedModelWithPartialLoad into the main model_cache/ directory. 2024-12-05 18:21:26 +00:00
Ryan Dick
5c67dd507a Get rid of ModelLocker. It was an unnecessary layer of indirection. 2024-12-05 16:59:40 +00:00
Ryan Dick
2ff928ec17 Move lock(...) and unlock(...) logic from ModelLocker to the ModelCache and make a bunch of ModelCache properties/methods private. 2024-12-05 16:11:40 +00:00
Ryan Dick
4327bbe77e Pull get_model_cache_key(...) out of ModelCache. The ModelCache should not be concerned with implementation details like the submodel_type. 2024-12-04 22:53:57 +00:00
Ryan Dick
ad1c0d37ef Rename model_cache_default.py -> model_cache.py. 2024-12-04 22:45:30 +00:00
Ryan Dick
9708d87946 Remove ModelCacheBase. 2024-12-04 22:05:34 +00:00
Ryan Dick
3ad44f7850 Move CacheStats to its own file. 2024-12-04 21:56:50 +00:00
Ryan Dick
9a482981b2 Move CacheRecord out to its own file. 2024-12-04 21:53:19 +00:00
Ryan Dick
6b02362b12 Rip out ModelLockerBase. 2024-12-04 21:47:11 +00:00
Ryan Dick
8fec4ec91c Tidy up CachedModel and improve unit test coverage. 2024-12-04 20:28:31 +00:00
Ryan Dick
693e421970 Alternative implementation with torch.nn.Linear module streaming. 2024-12-03 22:32:15 +00:00
Ryan Dick
dc14104bc8 Add TorchFunctionAutocastContext 2024-12-03 19:26:46 +00:00
Ryan Dick
f286a1d1f3 Remove debug logs. 2024-12-03 18:04:55 +00:00
Ryan Dick
9dc86b2b71 Add basic CachedModel class with features for partial load/unload. 2024-12-03 17:12:22 +00:00
Ryan Dick
2cab689b79 Naive TorchAutocastContext. 2024-12-03 14:55:43 +00:00
psychedelicious
f8c7adddd0 feat(ui): add vietnamese to language picker 
Closes #7384
 2024-12-02 08:12:14 -05:00
psychedelicious
17da1d92e9 fix(ui): remove "adding to" text on Invoke tooltip on Workflows/Upscaling tabs 
The "adding to" text indicates if images are going to the gallery or staging area. This info is relevant only to the canvas tab, but was displayed on Upscaling and Workflows tabs. Removed it from those tabs.
 2024-12-02 08:08:16 -05:00
psychedelicious
1cc57a4854 chore(ui): lint 2024-12-02 07:59:12 -05:00
psychedelicious
3993fae331 fix(ui): unable to invoke w/ empty inpaint mask or raster layer 
Removed the empty state checks for these layer types - it's always OK to invoke when they are empty.
 2024-12-02 07:59:12 -05:00
psychedelicious
1446526d55 tidy(ui): translation keys for canvas layer warnings 2024-12-02 07:59:12 -05:00
psychedelicious
62c024e725 feat(ui): add gallery image ctx menu items to create ref image from image 
Appears these actions disappeared at some point. Restoring them.
 2024-12-02 07:52:58 -05:00
psychedelicious
1e92bb4e94 fix(ui): ref image defaults to prev ref image's image selection 
A redux selector is used to get the "default" IP Adapter. The selector uses the model list query result to select an IP Adapter model to be preset by default.

The selector is memoized, so if we mutate the returned default IP Adapter state, it mutates the result of the selector for all consumers.

For example, the `image` property of the default IP Adapter selector result is `null`. When we set the `image` property of the selector result while creating an IP Adapter, this does not trigger the selector to recompute its result. We end up setting the image for the selector result directly, and all other consumers now have that same image set.

Solution - we need to clone the selector result everywhere it is used. This was missed in a few spots, causing the issue.
 2024-12-02 07:48:39 -05:00
psychedelicious
db6398fdf6 feat(ui): less confusing empty state for rg ref images 
It was easy to misunderstand the empty state for a regional guidance reference image. There was no label, so it seemed like it was the whole region that was empty.

This small change adds the "Reference Image" heading to the empty state, so it's clear that the empty state messaging refers to this reference image, not the whole regional guidance layer.
 2024-12-02 07:46:10 -05:00
Riccardo Giovanetti
ebd73a2ac2 translationBot(ui): update translation (Italian) 
Currently translated at 98.7% (1622 of 1643 strings)

Co-authored-by: Riccardo Giovanetti <riccardo.giovanetti@gmail.com>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/it/
Translation: InvokeAI/Web UI
 2024-12-02 02:13:51 -08:00
Hosted Weblate
8ee95cab00 translationBot(ui): update translation files 
Updated by "Cleanup translation files" hook in Weblate.

Co-authored-by: Hosted Weblate <hosted@weblate.org>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/
Translation: InvokeAI/Web UI
 2024-12-02 02:13:51 -08:00
Linos
d1184201a8 translationBot(ui): update translation (Vietnamese) 
Currently translated at 100.0% (1643 of 1643 strings)

translationBot(ui): update translation (Vietnamese)

Currently translated at 100.0% (1638 of 1638 strings)

Co-authored-by: Linos <linos.coding@gmail.com>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/vi/
Translation: InvokeAI/Web UI
 2024-12-02 02:13:51 -08:00
Nik Nikovsky
5887891654 translationBot(ui): update translation (Polish) 
Currently translated at 4.9% (81 of 1638 strings)

Co-authored-by: Nik Nikovsky <zejdzztegomaila@gmail.com>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/pl/
Translation: InvokeAI/Web UI
 2024-12-02 02:13:51 -08:00
Riku
765ca4e004 translationBot(ui): update translation (German) 
Currently translated at 69.7% (1142 of 1638 strings)

Co-authored-by: Riku <riku.block@gmail.com>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/de/
Translation: InvokeAI/Web UI
 2024-12-02 02:13:51 -08:00