Repository navigation
Reference and packers: Model L rows (Model M's stage tables) - #100
Merged
Merged
Conversation
Model L's ten Swin stages are Model M's stages with its own weights, so its rows in the model-keyed tables are Model M's rows, the same objects: the reference stage table and window offsets, the bneck and mid stage lists, the pipeline's Swin order. The stage runner accepts model="L" for them. Model M unchanged. Refs #89 Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Model L's ten Swin stages (enc1 … dec1) are Model M's stages with Model L's own weights: same shapes, window offsets
and blobs. This adds Model L's rows to the model-keyed tables of #92 as Model M's rows, the same objects. Model L's
extra width-32 blocks stay in
clean/l/reference/table.py. Nothing changes for Model M.clean/reference/swin.py:MODEL_STAGES["L"]andWINDOW_OFFSETS["L"]are Model M's.clean/runner/packing.py:MODEL_BNECK_STAGES["L"]andMODEL_MID_STAGES["L"]are Model M's lists.clean/runner/pipeline.py:MODEL_SWIN_ORDER["L"]is Model M's order.clean/runner/stage_runner.pyneeds no edit. ItsMODELScomes from the reference stage table, so it is now("M", "L", "K"), andRunner(..., model="L")runs these stages with Model M's kernels.Refs #89
Tested
clean/tests/test_model_tables.py, new classModelLRows:pack_stage/stage_blob_sizeas Model L give Model M's bytes for every stage;window_offsetsentry of the Model L table (clean/l/reference/table.py).The existing checks are updated for the new model list (
MODELS, the "known:" list of an unknown model).clean.tests.test_model_tables,clean.l.tests.test_l_reference,clean.tests.test_reference: 36 tests, OK.Model M identity: the 69-hash fingerprint of Reference and packers: model-keyed stage tables (Model M unchanged) #92 (tables, packed blobs in both layouts, reference outputs) is
identical to the one recorded on main before Reference and packers: model-keyed stage tables (Model M unchanged) #92.
Not tested