Implement ensure_weight_tying for trainable_token_indices (#2864) #2870
ENH: Implement ensure_weight_tying for trainable_token_indices (#2864)
423c6161
maintinaers comments addressed
232c6e7d
make style ran
213b47fc
Merge remote-tracking branch 'origin/main' into trainable-tokens-weig…
3874be0a
comments fixed
30d2d018
Merge main into trainable-tokens-weight-tying and resolve conflict in…
3f19ed39
Use _get_module_names_tied_with_embedding helper for cleaner code
88f5f22e
test fixes
c62aa564
Fix embedding name matching for nested paths and ensure weight tying …
769b8b0e
Apply ruff formatting
8252ccfb
Fix embedding name matching to use full paths and endswith for disamb…
616fb80a
Add _tied_weights_keys to MegaModel test and enable ensure_weight_tyi…
456cd367
Add tests for targeting both embedding and tied layers explicitly
6ed72f7d
Use mapping format for _tied_weights_keys in MegaModel test
7e1821df
Clarify docstrings for tied layer tests
5f86f3a2
Remove unnecessary reference to maintainer in docstring
69dbbe7d
Move BartConfig and BartModel imports to top of file
19692508
Rename MegaModel to CompositeModel throughout tests
eb20b839
style changes
d86bac89
Apply doc-builder style formatting to docstrings
518a3a74
Fix CI failure by setting sub-models _tied_weights_keys to None
b56d14c3
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub