🚨 [v5] Refactor RoPE for layer types #39847
zucchini-nlp
merged
103 commits into
huggingface:main
from
zucchini-nlp:rope-refactor-version-2
update
30aaa216
batch update model code
0e5d07bf
typos
0712f629
too many diffs, dump
4fc73559
dump again
b6162401
another dump
06cd2a87
fix copies
f66ad573
make `rope_scaling_dict` self attr
7dc077ff
fix a few more tests
4ac0f189
another update
9ad42e93
fix a few more tests, hopefully last ones
98944d5d
fox copies
12137694
a huuuge merge conflict resolved!
d787da72
fix copies again
00d4b3d2
fix newly added models, I hate rebasing on main
f9d4de3f
update config files
d695f5a1
modular files
303f218f
fix rope utils test
3229fbae
docstring has to be indented more, why?
1914d826
oops forgot to update some modualr files
fccb637e
copy from doesn't copy decorators?
709c414e
fix overriden test as well
b00d90c3
add a new test
c8120cf6
fix failing tests again
a352362f
update docstrings
2f54cb36
fix phi3
11edd474
Merge branch 'main' into rope-refactor-version-2
6bc850d3
fix two models
206f03d1
fix copies
8a9085fd
forgot to add
88328d2a
stupid bug from modular conversion
53247489
Merge remote-tracking branch 'upstream/main' into rope-refactor-versi…
62518de1
fix slow tests
01e79a89
update to call rotary emb once per model forward
1a4ccc76
3K tests failing?!
13d30ad9
update
df6a3c56
update more models
5c129c53
fix copies
c54bf4a1
fix the rest of tests hopefully
9b2b3577
merge main
373a8f37
fix after rebase
54ad1dce
fix the rope tests
c576edb0
fix docs omni
dc9ad8d0
change a bit
686b1a7e
models with layer types
182a6001
why it was deleted?
fdf68186
fix a few tests
c786413a
fix last test!
722541f3
delete extra empty lines
6a7321e3
add a test case
5a9cb987
more changes
403f56fd
fix models
a6c41247
typing hint for nested rope params
193eb23f
merge main
d1eeb420
missed when resolving conflicts
235ef331
delete layer types and fix typo
6efc1832
fix copies
cf7097d4
fix copies
234f2333
update docs text
a425778b
docs
8886f85b
huuge update all models
52192d9d
fix copies
74a4b4f3
rename attr to align with new format
362e88db
delete redundant rope tests
f458fd06
trigger ci
a7cf9928
merge main
05cd1ec1
update the case
f3172faf
this is why i hate rebasing
5739b6eb
maybe fixed?
ecde27c7
oops
9ecfa5ee
now fix?
5e379579
fix last tests and copies
2b279e07
merge main
96172118
gante
approved these changes
on 2025-10-08
fix copies?
878a9335
fix minimax and gemma3n
617f1aec
update typo
b983014b
deprecation end version
f7c50438
final fix copies :fingers-crossed:
07fa6303
oh my, add the docs in toctree
89beae3a
oke, this is really the last fix
721de32a
zucchini-nlp
changed the title
kill me please...
d530a860
fix copies and hope that tests won't start failing again
a5b397ba
use rope scaling if saved
c66337c5
fix slow tests
1bfc0133
fix cwm and unrelated deepseek
baf23d96
Merge branch 'main' into rope-refactor-version-2
3de40c3f
fix last
87278a9d
update
9dcd94d8
hope it works now, it took so long
dca1b94f
lets keep None for now, I will try to remove after checking tests
fcdff3ba
some more fixes, i find and replace does not always find all cases
9f6d9633
last fix of tests
ff23a682
Merge branch 'main' (commit b84c0b31) into rope-refactor-version-2
e4617dc1
arthur's comment for extra foreward kwargs
3c955d2c
delete unused code
b4186d1f
Merge branch 'main' into rope-refactor-version-2
cd175981
fix slow qwen tests
717ddc6a
delete layer types from models
5fdae079
faulty modular conversion
51ca43ab
fix qwen omni
080a7423
merge main
10493054
fix copies and style
5f126b71
ArthurZucker
approved these changes
on 2025-10-16
zucchini-nlp
commented
on 2025-10-16
zucchini-nlp
commented
on 2025-10-16
zucchini-nlp
commented
on 2025-10-16
address my comment
1cba3b83
Login to write a write a comment.
Login via GitHub