Instructions:
1. Download LoRA .safetensors files of your choice and place in
`INVOKEAIROOT/loras`. Unlike the draft version of this, the file
names can contain underscores and alphanumerics. Names with
arbitrary unicode characters are not supported.
2. Add `withLora(lora-file-basename,weight)` to your prompt. The
weight is optional and will default to 1.0. A few examples, assuming
that a LoRA file named `loras/sushi.safetensors` is present:
```
family sitting at dinner table eating sushi withLora(sushi,0.9)
family sitting at dinner table eating sushi withLora(sushi, 0.75)
family sitting at dinner table eating sushi withLora(sushi)
```
Multiple `withLora()` prompt fragments are allowed. The weight can be
arbitrarily large, but the useful range is roughly 0.5 to 1.0. Higher
weights make the LoRA's influence stronger.
In my limited testing, I found it useful to reduce the CFG to avoid
oversharpening. Also I got better results when running the LoRA on top
of the model on which it was based during training.
Don't try to load a SD 1.x-trained LoRA into a SD 2.x model, and vice
versa. You will get a nasty stack trace. This needs to be cleaned up.
3. You can change the location of the `loras` directory by passing the
`--lora_directory` option to `invokeai.
Documentation can be found in docs/features/LORAS.md.
- This variant, exemplified by "easynegative.safetensors" has a single
'embparam' key containing a Tensor.
- Also refactored code to make it easier to read.
- Handle both pickle and safetensor formats.
- Fixes longstanding bug in the token vector size code which caused
.pt files to be assigned the wrong token vector length. These
were then tossed out during directory scanning.
- Fixed the test for token length; tested on several .pt and .bin files
- Also added a __main__ entrypoint for CLI.py, to make pdb debugging a bit
more convenient.
When encountering a bad embedding, InvokeAI was asking about reconfiguring models. This is because the embedding load error was never handled - it now is.
* new OffloadingDevice loads one model at a time, on demand
* fixup! new OffloadingDevice loads one model at a time, on demand
* fix(prompt_to_embeddings): call the text encoder directly instead of its forward method
allowing any associated hooks to run with it.
* more attempts to get things on the right device from the offloader
* more attempts to get things on the right device from the offloader
* make offloading methods an explicit part of the pipeline interface
* inlining some calls where device is only used once
* ensure model group is ready after pipeline.to is called
* fixup! Strategize slicing based on free [V]RAM (#2572)
* doc(offloading): docstrings for offloading.ModelGroup
* doc(offloading): docstrings for offloading-related pipeline methods
* refactor(offloading): s/SimpleModelGroup/FullyLoadedModelGroup
* refactor(offloading): s/HotSeatModelGroup/LazilyLoadedModelGroup
to frame it is the same terms as "FullyLoadedModelGroup"
---------
Co-authored-by: Damian Stewart <null@damianstewart.com>
This PR adds a new attributer to ldm.generate, `embedding_trigger_strings`:
```
gen = Generate(...)
strings = gen.embedding_trigger_strings
strings = gen.embedding_trigger_strings()
```
The trigger strings will change when the model is updated to show only
those strings which are compatible with the current
model. Dynamically-downloaded triggers from the HF Concepts Library
will only show up after they are used for the first time. However, the
full list of concepts available for download can be retrieved
programatically like this:
```
from ldm.invoke.concepts_lib import HuggingFAceConceptsLibrary
concepts = HuggingFaceConceptsLibrary()
trigger_strings = concepts.list_concepts()
```
