Commit Graph

2382 Commits

Author SHA1 Message Date
a7048eea5f Merge branch 'main' into feat/import-with-vae 2023-01-24 09:07:41 -05:00
87c9398266 [enhancement] import .safetensors ckpt files directly (#2353)
This small fix makes it possible to import and run safetensors ckpt
files directly without doing a conversion step first.
2023-01-24 09:06:49 -05:00
9f32daab2d Merge branch 'main' into lstein-import-safetensors 2023-01-23 21:58:07 -05:00
884768c39d Make sure --free_gpu_mem still works when using CKPT-based diffuser model (#2367)
This PR attempts to fix `--free_gpu_mem` option that was not working in
CKPT-based diffuser model after #1583.

I noticed that the memory usage after #1583 did not decrease after
generating an image when `--free_gpu_mem` option was enabled.
It turns out that the option was not propagated into `Generator`
instance, hence the generation will always run without the memory saving
procedure.

This PR also related to #2326. Initially, I was trying to make
`--free_gpu_mem` works on 🤗 diffuser model as well.
In the process, I noticed that InvokeAI will raise an exception when
`--free_gpu_mem` is enabled.
I tried to quickly fix it by simply ignoring the exception and produce a
warning message to user's console.
2023-01-23 21:48:23 -05:00
bc2194228e stability improvements
- provide full traceback when a model fails to load
- fix VAE record for VoxelArt; otherwise load fails
2023-01-23 21:40:27 -05:00
10c3afef17 Merge branch 'main' into bugfix/free-gpu-mem-diffuser 2023-01-23 21:15:12 -05:00
66babb2e81 Japanese Localization Build 2023-01-24 09:07:29 +13:00
31a967965b Add Japanese Localization 2023-01-24 09:07:29 +13:00
b9c9b947cd update japanese translation 2023-01-24 09:07:29 +13:00
1eee08a070 add Japanese Translation 2023-01-24 09:07:29 +13:00
aca1b61413 [Feature] Add interactive diffusers model merger (#2388)
This PR adds `scripts/merge_fe.py`, which will merge any 2-3 diffusers
models registered in InvokeAI's `models.yaml`, producing a new merged
model that will be registered as well.

Currently this script will only work if all models to be merged are
known by their repo_ids. Local models, including those converted from
ckpt files, will cause a crash due to a bug in the diffusers
`checkpoint_merger.py` code. I have made a PR against
huggingface/diffusers which fixes this:
https://github.com/huggingface/diffusers/pull/2060
2023-01-23 09:27:05 -05:00
e18beaff9c Merge branch 'main' into feat/merge-script 2023-01-23 09:05:38 -05:00
d7554b01fd fix typo in prompt 2023-01-23 00:24:06 -08:00
70f8793700 Merge branch 'main' into feat/import-with-vae 2023-01-23 00:17:46 -08:00
ea61bf2c94 [bugfix] ckpt conversion script respects cache in ~/invokeai/models (#2395) 2023-01-23 00:07:23 -08:00
7dead7696c fixed setup.py to install the new scripts 2023-01-23 00:43:15 -05:00
ffcc5ad795 conversion script uses invokeai models cache by default 2023-01-23 00:35:16 -05:00
48deb3e49d add model merging documentation and launcher script menu entries 2023-01-23 00:20:28 -05:00
6c31225d19 create small module for merge importation logic 2023-01-22 18:07:53 -05:00
f0fe483915 Merge branch 'main' into feat/merge-script 2023-01-21 18:42:40 -05:00
4ee8d104f0 working, but needs diffusers PR to be accepted 2023-01-21 18:39:13 -05:00
89791d91e8 fix: use pad_token for padding (#2381)
Stable Diffusion 2 does not use eos_token for padding.

Fixes #2378
2023-01-21 13:30:03 -08:00
87f3da92e9 Merge branch 'main' into fix/sd2-padding-token 2023-01-21 13:11:02 -08:00
f169bb0020 fix long prompt weighting bug in ckpt codepath (#2382) 2023-01-21 15:14:14 -05:00
155efadec2 Merge branch 'main' into fix/sd2-padding-token 2023-01-21 21:05:40 +01:00
e94c8fa285 fix long prompt weighting bug in ckpt codepath 2023-01-21 12:08:21 +01:00
b3363a934d Update index.md (#2377) 2023-01-21 00:17:23 -05:00
599c558c87 Merge branch 'main' into patch-1 2023-01-20 23:54:40 -05:00
d35ec3398d fix: use pad_token for padding
Stable Diffusion does not use the eos_token for padding.
2023-01-20 19:25:20 -08:00
96a900d1fe correctly import diffusers models by their local path
- Corrects a bug in which the local path was treated as a repo_id
2023-01-20 20:13:43 -05:00
f00f7095f9 Add instructions for installing xFormers on linux (#2360)
I've written up the install procedure for xFormers on Linux systems.

I need help with the Windows install; I don't know what the build
dependencies (compiler, etc) are. This section of the docs is currently
empty.

Please see `docs/installation/070_INSTALL_XFORMERS.md`
2023-01-20 17:57:12 -05:00
279ffcfe15 Merge branch 'main' into lstein/xformers-instructions 2023-01-20 17:29:39 -05:00
9b73292fcb add pip install documentation for xformers 2023-01-20 17:28:14 -05:00
da81165a4b Update index.md 2023-01-20 19:03:12 +01:00
f3ff386491 [enhancement] Reorganize form for textual inversion training (#2375)
- Add num_train_epochs
- Reorganize widgets so all sliders that control # of steps are together
2023-01-20 10:58:26 -05:00
da524f159e Merge branch 'main' into feat/enhance-ti-training-ui 2023-01-20 10:28:27 -05:00
2d1eeec063 Save HFToken only if it is present (#2370)
Fixes https://github.com/invoke-ai/InvokeAI/issues/2083
2023-01-19 22:16:19 -05:00
a8bb1a1109 Save HFToken only if it is present 2023-01-19 21:47:27 -05:00
d9fa505412 [feat] Provide option to disable xformers from command line (#2373)
Starting `invoke.py` with --no-xformers will disable
memory-efficient-attention support if xformers is installed.

For symmetry, `--xformers` will enable support, but this is already the
default if xformers is available.
2023-01-19 19:15:57 -05:00
02ce602a38 Merge branch 'main' into feat/disable-xformers 2023-01-19 18:45:59 -05:00
9b1843307b [enhancement] Reorganize form for textual inversion training
- Add num_train_epochs
- Reorganize widgets so all sliders that control # of steps are together
2023-01-19 18:43:12 -05:00
f0010919f2 Merge branch 'main' into bugfix/free-gpu-mem-diffuser 2023-01-19 18:03:36 -05:00
d113b4ad41 [bugfix] suppress extraneous warning messages generated by diffusers (#2374)
This commit suppresses a few irrelevant warning messages that the
diffusers module produces:

1. The warning that turning off the NSFW detector makes you an
irresponsible person.
2. Warnings about running fp16 models stored in CPU (we are not running
them in CPU, just caching them in CPU RAM)
2023-01-19 18:00:31 -05:00
895505976e [bugfix] suppress extraneous warning messages generated by diffusers
This commit suppresses a few irrelevant warning messages that the
diffusers module produces:

1. The warning that turning off the NSFW detector makes you an
irresponsible person.
2. Warnings about running fp16 models stored in CPU (we are not running
   them in CPU, just caching them in CPU RAM)
2023-01-19 16:49:40 -05:00
171f4aa71b [feat] Provide option to disable xformers from command line
Starting `invoke.py` with --no-xformers will disable
memory-efficient-attention support if xformers is installed.

--xformers will enable support, but this is already the
default.
2023-01-19 16:16:35 -05:00
33a5c83c74 during ckpt->diffusers tell user when custom autoencoder can't be loaded
- When a ckpt or safetensors file uses an external autoencoder and we
  don't know which diffusers model corresponds to this (if any!), then
  we fallback to using stabilityai/sd-vae-ft-mse
- This commit improves error reporting so that user knows what is happening.
2023-01-19 12:05:49 -05:00
7ee0edcb9e when converting a ckpt/safetensors model, preserve vae in diffusers config
- After successfully converting a ckt file to diffusers, model_manager
  will attempt to create an equivalent 'vae' entry to the resulting
  diffusers stanza.

- This is a bit of a hack, as it relies on a hard-coded dictionary
  to map ckpt VAEs to diffusers VAEs. The correct way to do this
  would be to convert the VAE to a diffusers model and then point
  to that. But since (almost) all models are using vae-ft-mse-840000-ema-pruned,
  I did it the easy way first and will work on the better solution later.
2023-01-19 11:02:49 -05:00
7bd2220a24 fix two bugs in model import
1. !import_model did not allow user to specify VAE file. This is now fixed.
2. !del_model did not offer the user the opportunity to delete the underlying
   weights file or diffusers directory. This is now fixed.
2023-01-19 01:30:58 -05:00
284b432ffd add triton install instructions 2023-01-18 22:34:36 -05:00
be58a6bfbc Merge branch 'main' into bugfix/free-gpu-mem-diffuser 2023-01-19 10:21:06 +07:00