* Restrict resamplers for img2img and high res fix to the ones that
PIL.Image actually supports, since it uses that to di the resampling.
Removed: Antialias, Affine, Cubic. Added: Hamming.
* Set list of available schedulers to CPU only when high res fix
is selected in the web ui. Set list to all schdulers when high res fix
is deselected.
* Put hi res fix in its own Accordian in the txt2img UI instead of
grouping it with Advanced Options.
* Moves the imports of winreg and Tk, into the functions that use them,
with winreg behind a guard clause. This should hopefully mean that if
you're not on Window or not using `ui=app` we won't trip over either
of these due to them not being there.
* On windows insist on the presence of webview2 as the embeddable
browser for `ui=app`. If we can't find it, effectively switch back to
`ui=web`. This should prevent pywebview trying to use MSHTML, whilst
saying its deprecated, and apparently we are too much for poor old IE11
* Add webview2 runtime droppings to .gitignore.
* If we can't bind to args.server_port get another suitable port from
the OS and advise the user that we did this in the UI.
* Make `ui=web` mode use 'SHARK AI Studio' as its title. This makes it
consistent with `ui=app`.
* Replace the generic gradio favicon with a nod swirl one instead.
* SD/API: Add missing a1111 API features for Koboldcpp
* Refactors SD api functions into their own file
* Adds the following apis implemented by a1111 as needed by koboldcpp:
- adds /sdapi/v1/sd-models (lists available models)
- adds /sdapi/v1/options (only the bare minimum needed)
* Adds optional CORS support, use the '--api_accept_origin' command line
argument to activate and configure.
* Extends existing APIs to include optional sampler/scheduler selection
* Extends /sdapi/v1/textimg to recognise the method used by koboldcpp
to select the model.
* Where possible take values not provided to the API in the request from
the existing relevant command line parameters rather than hardcoding
them.
* return a 400 response when a request doesn't have required properties.
* changed default schedulers and models for some apis to ones that
actually seem to work.
* Update api_test.py to include the new APIs.
* Update api_test.py to include a '--verbose' command line option.
* SD/API: Take more API values from args
* Take LoRA from '--use_lora' command line arg if specified
* Take device from '--device' command line arg if specified (substring
match, so a short name such as 'vulkan://0' should work)
* SD/API: add more endpoints and pydantic typing
* Mount the whole of /sdapi from index.py as a FastAPI application,
rather than each endpoint individually
* Add the following additional API endpoints:
* /sdapi/v1/samplers
* /sdapi/v1/cmd-flags
* Make scheduler/sampler selection checking and fallback much more
robust.
* Support aliasing some A1111 scheduler/sampler names to the diffusers
ones we are using.
* Expand response /sdapi/v1/options to add a few more things.
* Split non-api functions and variables into their own utils.py file.
* Support 'n_iter' request property and the return of multiple images
from generation endpoints. Equivalent of '--batch_count', batch_size
is stil hardcoded at 1
* Include (some) hires_fix request properties in txt2img endpoint
* Rework endpoints using pydantic model classes for better request
validation and so we get much improved swagger api docs at
/sdapi/docs and redoc at /sdapi/redoc
* SD/API Delete commented out code from index.py
* Delete some code that is no longer needed by the SD API in index.py
(and one line sdapi_v1.py) that I'd previously only commented out.
* SD/UI: Add shark_sd_koboldcpp.md document
* Add documentation on how to set up Koboldcpp with SHARK
* Link this and the existing blender set up document from the main
README.md
* SD/API Improve stencil options in img2img endpoint
In /sdapi/v1/img2img:
* Add zoedepth to the controlnet use_stencil options
* Require and use second image as stencil mask for controlnet scribble
Add a new flag `-Xiree_compile` to forward extra compiler arguments to
`iree-compile`. This flag can be set multiple times to pass more than
one extra argument.
This commit adds the support for sharded Falcon-7b-GPTQ and
Falcon-180B-GPTQ. This commit also adds the support for 4-way
sharding of the Falcon model for the device ROCM.
Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>
* correct my sloppy copy/paste for the untuned unet default compilation
flags that introduced an extra 'detach' into what should have been
'iree-global-opt-convert-1x1-filter-conv2d-to-matmul'
* Allow entry of a huggingface model id or civitai download url to be
done in the main model selection dropdown on SD tabs
* Remove separate textbox for entering huggingface model id or civitai
download url on SD Tabs
* Remove 'None' option from the model selection dropdown (no longer
needed) on SD tabs
* Update png metadata drop zone on txt2img tab to work with a single
argument for model selection
* Update UI generate functions on SD tabs to work with single argument
model selection
* Update API code for changes to the UI generate functions
* Move info about the custom model path to the logging textarea on SD
tabs
* Set the input seed for the random number generator when
generating repeatable seeds to exclude any negative numbers
in the parsed seed input. The makes seeds generated for
different batch counts consistent where they have the same
input for the initial seed or set of seeds.
* Switch most compile flows to use ireec.compile_file.
* re-add input type to compile_str path.
* Check if mlir_module exists before checking if it's a path or pyobject.
* Fix some save_dir cases
Print averaged results at the end of all iterations. Increase the
default number of iterations to 5.
Example:
```
Number of iterations: 5
Prefill: avg. 0.03 s, stddev 0.00
Decode: avg. 43.34 tokens/s, stdev 0.13
```
Also remove the -2 in the number of generated tokens -- I did not find
any evidence we need it.
