So far the slider component was unable to take typed input due to a
bunch of issues that were a pain to solve. This PR fixes it.
Things to test:
- Moving the slider also updates the value in the input text box.
- Input text box next to slider can be changed in two ways: If you type
a manual value, the slider will be updated when you lose focus from the
input box. If you use the stepper icons to update the values, the slider
should update immediately.
- Make sure the reset buttons next to the slider are updating correctly
and make sure this updates both the slider and the input box values.
- Brush Size slider -> make sure the hotkeys are updating the input box
too.
- This replaces the original clipseg library with the transformers
version from HuggingFace.
- This should make it possible to register InvokeAI at PyPi and do a
fully automated pip-based install.
- Minor regression: it is no longer possible to specify which device the
clipseg model will be loaded into, and it will reside in CPU. However,
performance is more than acceptable.
- This replaces the original clipseg library with the transformers
version from HuggingFace.
- This should make it possible to register InvokeAI at PyPi and do
a fully automated pip-based install.
- Minor regression: it is no longer possible to specify which device
the clipseg model will be loaded into, and it will reside in CPU.
However, performance is more than acceptable.
Fix two deficiencies in the CLI's support for model management:
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.
This PR improves the console reporting of the process of recognizing
trigger tokens and loading their embeds.
1. Do not report "concept is not known to HuggingFace" if the trigger
term is in fact a local embedding trigger.
2. When a trigger term is first recognized during a session, report the
fact.
This should help debug embedding issues in the future.
Note that the local embeddings produced by the new InvokeAI TI training
script default to the format <trigger> with literal angle brackets. This
sets them off from the rest of the text well and will enable
autocomplete at some point in the future. However, this means that they
supersede like-named HuggingFace concepts, and may cause problems for
people uploading them to the HuggingFace repository (although that
problem already exists).
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.
