transformers
🔴 🚨 Resizing tokens embeddings: initialize from old embeddings' normal distribution.
#33325
Merged

🔴 🚨 Resizing tokens embeddings: initialize from old embeddings' normal distribution. #33325

abuelnasr0
abuelnasr0
LysandreJik
Rocketknight1
Rocketknight1
Rocketknight1 approved these changes on 2024-09-05
HuggingFaceDocBuilderDev
abuelnasr0 abuelnasr0 requested a review from Rocketknight1 Rocketknight1 1 year ago
abuelnasr0
Rocketknight1
Rocketknight1 approved these changes on 2024-09-06
ArthurZucker
ArthurZucker commented on 2024-09-06
abuelnasr0 abuelnasr0 requested a review from ArthurZucker ArthurZucker 1 year ago
abuelnasr0 abuelnasr0 requested a review from Rocketknight1 Rocketknight1 1 year ago
ArthurZucker
ArthurZucker approved these changes on 2024-09-13
abuelnasr0 abuelnasr0 changed the title Resizing tokens embeddings: initialize from old embeddings' normal distribution. 🔴 🔴 Resizing tokens embeddings: initialize from old embeddings' normal distribution. 1 year ago
abuelnasr0 abuelnasr0 changed the title 🔴 🔴 Resizing tokens embeddings: initialize from old embeddings' normal distribution. 🔴 🚨 Resizing tokens embeddings: initialize from old embeddings' normal distribution. 1 year ago
ArthurZucker
ArthurZucker commented on 2024-09-17
abuelnasr0
abuelnasr0
Rocketknight1
abuelnasr0
abuelnasr0
ArthurZucker
ArthurZucker commented on 2024-10-01
abuelnasr0 abuelnasr0 force pushed 1 year ago
ArthurZucker
ArthurZucker approved these changes on 2024-10-03
ArthurZucker
abuelnasr0 intilize new embeddings from normal distrib
25c92e1e
abuelnasr0 Fix typo in comments
a95639c3
abuelnasr0 Fix typo in comments
d850b995
abuelnasr0 Fix style
3f445078
abuelnasr0 Fix variables naming
5ea5f828
abuelnasr0 Add tests
d1d81d52
abuelnasr0 Fix style
f3aaf0af
abuelnasr0 code consistency nit
bdef61af
abuelnasr0 Add deepspeed support
15a7b5ab
abuelnasr0 Add deepspeed support
6e40b4f6
abuelnasr0 Conver embeddings weights to float32 before computations
aba7d8c2
abuelnasr0 Add deepspeed tests
4f1b0fa5
abuelnasr0 Cover when vocab_size is smaller than embedding_size
dea8e285
abuelnasr0 Style fix
84f8cfa7
abuelnasr0 Add tests for vocab_size smaller than hiddin_size
2923e858
abuelnasr0 Style fix
188ba1bd
abuelnasr0 Nits in tests
22ac85c3
abuelnasr0 Nits in tests
3e42f66e
abuelnasr0 Check for deepspeed before importing it
226f31c7
abuelnasr0 Increase vocab_size for positive definite covariance matrix test
cef744fa
abuelnasr0 Add warning
6583cd5e
abuelnasr0 Add multivariate_resizing flag and implement resizing for lm_heads
7577cd49
abuelnasr0 Fix typo
0472bacc
abuelnasr0 Fix wrong bias indexing
fd4ad000
abuelnasr0 Fix bias is zero check
6ff2bca9
abuelnasr0 remove multivariate_resizing flag from tests
12e61c61
abuelnasr0 Intialize bias from old bias normal distribution
eb80c339
abuelnasr0 Fixup
ef6bdbc4
abuelnasr0 Code usability
5cdce5f3
abuelnasr0 Use mean_resizing instead of multivariate_resizing
f4a9cf46
abuelnasr0 Fix up
fc436d7e
abuelnasr0 abuelnasr0 force pushed to fc436d7e 1 year ago
abuelnasr0 Fix comments and docs
8e60a368
abuelnasr0
ArthurZucker
ArthurZucker ArthurZucker merged 78ef5832 into main 1 year ago
Rocketknight1
abuelnasr0

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone