To be consistent with max_cache_size, the amount of memory to hold in
VRAM for model caching is now controlled by the max_vram_cache_size
configuration parameter.
Image files are immutable and we expect deletion to result in no further requests for a given image, so we can set the max-age to something thicc.
Resolves#3426
This PR is to allow FP16 precision to work on Macs with MPS. In
addition, it centralizes the torch fixes/workarounds required for MPS
into a new backend utility `mps_fixes.py`. This is conditionally
imported in `api_app.py`/`cli_app.py`.
Many MANY thanks to @StAlKeR7779 for patiently working to debug and fix
these issues.
- No longer fail root directory probing if invokeai.yaml is missing
(test is now whether a `models/core` directory exists).
- Migrate script does not overwrite previously-installed models.
- Can run migrate script on an existing 2.3 version directory
with --from and --to pointing to same 2.3 root.
The list models route should just be the base route path, and should use query parameters as opposed to path parameters (which cannot be optional)
Removed defaults for update model route - for the purposes of the API, we should always be explicit with this
This PR is to allow FP16 precision to work on Macs with MPS. In addition, it centralizes the torch fixes/workarounds
required for MPS into a new backend utility file `mps_fixes.py`. This is conditionally imported in `api_app.py`/`cli_app.py`.
Many MANY thanks to StAlKeR7779 for patiently working to debug and fix these issues.
