Skip to content
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

Jump to bottom

Update generation.py to add support for repetition penalty #733

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+12 −0
Open

Update generation.py to add support for repetition penalty #733

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
12 changes: 12 additions & 0 deletions llama/generation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,6 +110,7 @@ def generate(
temperature: float = 0.6,
top_p: float = 0.9,
logprobs: bool = False,
repetition_penalty: float = (1.0/0.9),
echo: bool = False,
) -> Tuple[List[List[int]], Optional[List[List[float]]]]:
params = self.model.params
Expand All @@ -133,6 +134,17 @@ def generate(
input_text_mask = tokens != pad_id
for cur_pos in range(min_prompt_len, total_len):
logits = self.model.forward(tokens[:, prev_pos:cur_pos], prev_pos)
if repetition_penalty != 1.0:
logits_new = logits.clone()
batch_size = len(tokens)
for i in range(batch_size):
for token in set(tokens[i].tolist()):
# if score < 0 then repetition penalty has to multiplied to reduce the previous token probability
if logits[0, i, token] < 0:
logits_new[0, i, token] = logits[0, i, token] * repetition_penalty
else:
logits_new[0, i, token] = logits[0, i, token] / repetition_penalty
logits = logits_new
if logprobs:
token_logprobs[:, prev_pos + 1 : cur_pos + 1] = -F.cross_entropy(
input=logits.transpose(1, 2),
Expand Down