You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Right now, we handle compile flags in sd_inference/utils.py and pass in some combination of MLIR, keywords, device, IREE target, model name, etc. to compile a list of IREE compile flags to use.
This would be better managed by a repository of flag files, parametrized by pinned IREE "flag versions" (we move the SHA every time we need to update flags for a IREE-side change and just hardcode that SHA in utils.py, maybe) then by model, then by configuration (think the same configs we make unique .vmfb filenames for)
More on this later, but putting a bookmark here so it doesn't get lost.
The text was updated successfully, but these errors were encountered:
Right now, we handle compile flags in sd_inference/utils.py and pass in some combination of MLIR, keywords, device, IREE target, model name, etc. to compile a list of IREE compile flags to use.
This would be better managed by a repository of flag files, parametrized by pinned IREE "flag versions" (we move the SHA every time we need to update flags for a IREE-side change and just hardcode that SHA in utils.py, maybe) then by model, then by configuration (think the same configs we make unique .vmfb filenames for)
More on this later, but putting a bookmark here so it doesn't get lost.
The text was updated successfully, but these errors were encountered: