Skip to content

Correct tie_embedding_weights and add logit checking #1288

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mattdangerw merged 6 commits into from
Oct 28, 2023
Merged

Correct tie_embedding_weights and add logit checking #1288

Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
f97cb90
Add tie_embedding_weights and logit checking
nkovela1 Oct 27, 2023
c1edcad
Fix formatting
nkovela1 Oct 27, 2023
2eec377
Remove comments
nkovela1 Oct 27, 2023
a8178ff
Change backbone default
nkovela1 Oct 27, 2023
e68d815
Change weights hashes
nkovela1 Oct 27, 2023
05ace8d
Fix first hash
nkovela1 Oct 28, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Fix formatting
  • Loading branch information
@nkovela1
nkovela1 committed Oct 27, 2023
commit c1edcad9ccc0f6dff034b39dcf1cd59d7899e449
14 changes: 10 additions & 4 deletions tools/checkpoint_conversion/convert_t5_checkpoints.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,8 +82,10 @@ def convert_checkpoints(hf_model):
hf_wts[f"{section}.embed_tokens.weight"]
)
if not keras_nlp_model.tie_embedding_weights:
keras_nlp_model.get_layer("token_embedding").reverse_embeddings.assign(
hf_wts[f"lm_head.weight"].transpose(1, 0).numpy()
keras_nlp_model.get_layer(
"token_embedding"
).reverse_embeddings.assign(
hf_wts["lm_head.weight"].transpose(1, 0).numpy()
)

# Query, key, value, and output projectors in self-attention
Expand Down Expand Up @@ -333,9 +335,13 @@ def check_output(
)

if keras_model.tie_embedding_weights:
keras_hidden_states = keras_hidden_states * (keras_model.hidden_dim ** -0.5)
keras_hidden_states = keras_hidden_states * (
keras_model.hidden_dim**-0.5
)

keras_logits = keras_model.token_embedding(keras_hidden_states, reverse=True)
keras_logits = keras_model.token_embedding(
keras_hidden_states, reverse=True
)
hf_logits = hf_out.logits
print("-> KerasNLP logits:", keras_logits[0:5])
print("-> HF logits:", hf_logits[0:5])
Expand Down