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

Fixing rolling loglikelihood management #78

Merged
merged 5 commits into from
Mar 8, 2024
Merged

Fixing rolling loglikelihood management #78

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
024654c
wip on rolling management
clefourrier Mar 1, 2024
53c8ac5
Merge branch 'main' into clem_fix_rolling
clefourrier Mar 4, 2024
67fc7fb
Merge branch 'main' into clem_fix_rolling
clefourrier Mar 5, 2024
bcd7322
Merge branch 'main' into clem_fix_rolling
clefourrier Mar 6, 2024
664f1a0
Merge branch 'main' into clem_fix_rolling
clefourrier Mar 6, 2024
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
11 changes: 8 additions & 3 deletions src/lighteval/models/base_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -560,6 +560,7 @@ def loglikelihood_rolling(
requests,
override_bs=override_bs,
return_bool_score=False,
rolling=True,
)
return results

Expand All @@ -568,6 +569,7 @@ def _loglikelihood_tokens(
requests: list[LoglikelihoodRequest],
override_bs: int = -1,
return_bool_score: bool = True,
rolling: bool = False,
) -> list[LoglikelihoodReturn]:
dataset = LoglikelihoodDataset(requests=requests, dataset_splits=self.DATASET_SPLITS)
starting_batch_size = STARTING_BATCH_SIZE
Expand All @@ -576,9 +578,12 @@ def _loglikelihood_tokens(
for split_start, split_end in tqdm(dataset.splits_start_end_iterator()):
context_enc = dataset[0].tokenized_context
continuation_enc = dataset[0].tokenized_continuation
max_context_continuation_size_allowed = len(
(context_enc + continuation_enc)[-(self.max_length + 1) :][:-1]
)
if rolling: # we take all the sequence in rolling mode
max_context_continuation_size_allowed = len(context_enc + continuation_enc)
else: # in normal mode, we left cut the context if needed
max_context_continuation_size_allowed = len(
(context_enc + continuation_enc)[-(self.max_length + 1) :][:-1]
)

batch_size = self._get_batch_size(
override_bs=override_bs,
Expand Down
Loading