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

debugged 2042 past key val inputs #12

Merged
merged 1 commit into from
Jun 7, 2023
Merged

debugged 2042 past key val inputs #12

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
26 changes: 14 additions & 12 deletions src/transformers/generation/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2036,8 +2036,20 @@ def contrastive_search(
if self.config.is_encoder_decoder
else (outputs.hidden_states,)
)

print (top_k_ids)

# Replicates the new past_key_values to match the `top_k` candidates
new_key_values = []
for layer in model_kwargs["past_key_values"]:
items = []
# item is either the key or the value matrix
for item in layer:
if low_memory:
items.append(item.repeat_interleave(1, dim=0))
else:
items.append(item.repeat_interleave(top_k, dim=0))
new_key_values.append(items)
model_kwargs["past_key_values"] = new_key_values
print (top_k_ids.view(-1, 1))

# if the used memory exceeds a threshold, do not batch
if low_memory:
Expand All @@ -2056,16 +2068,6 @@ def contrastive_search(
print (outputs)

else:
# Replicates the new past_key_values to match the `top_k` candidates
new_key_values = []
for layer in model_kwargs["past_key_values"]:
items = []
# item is either the key or the value matrix
for item in layer:
items.append(item.repeat_interleave(top_k, dim=0))
new_key_values.append(items)
model_kwargs["past_key_values"] = new_key_values

# compute the candidate tokens by the language model and collect their hidden_states
# assembles top_k_ids into batch of size k (leading to OOM for large models)
next_model_inputs = self.prepare_inputs_for_generation(top_k_ids.view(-1, 1), **model_kwargs)
Expand Down