Skip to content

Fix mixed precision crash in TopP and Random samplers #2585

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

Open
AlejandroPG06 wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Fix mixed precision crash in TopP and Random samplers #2585

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
49defce
Fix mixed precision crash in TopP and Random samplers
AlejandroPG06 Feb 6, 2026
47bc15e
Update keras_hub/src/samplers/random_sampler.py
AlejandroPG06 Feb 6, 2026
9b96e6f
Update keras_hub/src/samplers/top_p_sampler.py
AlejandroPG06 Feb 6, 2026
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
4 changes: 3 additions & 1 deletion keras_hub/src/samplers/random_sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,8 +45,10 @@ def __init__(

def get_next_token(self, probabilities):
# Sample the next token from the probability distribution.
# tf does not support half precision multinomial sampling, so make
# sure we have full precision here.
next_token_id = random.categorical(
ops.log(probabilities),
ops.cast(ops.log(probabilities), "float32"),
1,
seed=self.seed_generator,
dtype="int32",
Expand Down
4 changes: 3 additions & 1 deletion keras_hub/src/samplers/top_p_sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,7 +79,9 @@ def get_next_token(self, probabilities):
ops.zeros(ops.shape(sorted_preds), dtype=sorted_preds.dtype),
)
sorted_next_token = random.categorical(
ops.log(probabilities),
# tf does not support half precision multinomial sampling, so make
# sure we have full precision here.
ops.cast(ops.log(probabilities), "float32"),
1,
seed=self.seed_generator,
dtype="int32",
Expand Down
Loading