Skip to content

Bug fixes in reward utils #3

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
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Bug fixes in reward utils #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 3 additions & 2 deletions kernel-coder/nano_r1_script.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -229,7 +229,7 @@ def extract_first_code(output_string: str, code_language_types: list[str]) -> st

return None

def format_reward_func(completion: str, EOS_TOKEN: str) -> float:
def format_reward_func(completion: str, EOS_TOKEN: str) -> tuple[float, str | None]:
"""
Format: <think>...</think>anything

Expand All @@ -238,7 +238,8 @@ def format_reward_func(completion: str, EOS_TOKEN: str) -> float:
EOS_TOKEN (str): End of sequence token

Returns:
float: Reward score
Tuple containing the reward score and the first extracted code block if
available.
"""

code = None
Expand Down
3 changes: 2 additions & 1 deletion kernel-coder/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -187,7 +187,8 @@ def compute_token_log_probs(

logits = outputs.logits / temperature # Shape: [batch_size, seq_len, vocab_size]
shift_logits = logits[..., :-1, :] # Shape: [batch_size, seq_len-1, vocab_size]
shift_labels = inputs["labels"][..., 1:] # Shape: [batch_size, seq_len-1]
# Clone to avoid in-place modification of `inputs["labels"]`
shift_labels = inputs["labels"][..., 1:].clone() # Shape: [batch_size, seq_len-1]
shift_labels_mask = inputs["labels_mask"][..., 1:] # Shape: [batch_size, seq_len-1]

# Create mask for valid labels
Expand Down