* Adding metal_utils for iree_utils
* Add patch for making compile API work for both MEGABYTE and MiniGPT4 (#1559)
-- It also modifies the mega_test.py script
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
* [SD] Update unet in_channels API and add PIL metadata to spec. (#1560)
* Fix deprecation warning for unet config.
* Include PIL metadata instead of hidden imports in SD spec.
* Fixing iree-metal-target-platform
* adding metal to txt2img pipeline
* Fixing Copyright date
* removing debug prints
* black lint formating
* fixing device dump
---------
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: Abhishek Varma <avarma094@gmail.com>
Co-authored-by: Ean Garvey <87458719+monorimet@users.noreply.github.com>
Co-authored-by: powderluv <powderluv@users.noreply.github.com>
* Change retrieval of Iree options used when loading an existing
unet512 vmfb to look up the "unet" options rather than attempt to
find a non-existent set of options for "unet512"
Co-authored-by: powderluv <powderluv@users.noreply.github.com>
* Switches a few places in the SD pipeline where an assumption of
max_length=64 was being made, to using the actual max_length
as passed into the pipeline. This prevents unet512 always being
used and producing different images than previously when
--max_length=77
* Fixes randomize seed buttons that stopped working.
* Update now deprecated method to set initial colums for output
gallery to the newer undeprecated one.
single endpoint in apps/language/models/scripts/vicuna.py
removed main functions from pipelines
replaced divergent utils compile with shark_importer
adds support for different precisions
* On startup report that cleaning up of temp files is taking place, in
case it takes a long time.
* Have the output gallery tab delete any zero length temporary files
generated by gradio < 3.32.0 for its gallery control whenever it
needs to update that control with images. This prevents such
files multiplying out of control.
Model config generator takes a PyTorch model as input and generates a JSON file with model layers and other propperties that define sharding on a particular hardware.
This fix the gallery component of the output gallery dumping temporary
files into the standard folders rather than shark_tmp so those files never
got cleared out on restart and would build up.