Skip to content

Replace from_legacy_cache method with constructors #2767

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

Merged
BenjaminBossan merged 2 commits into from
Sep 8, 2025
Merged

Replace from_legacy_cache method with constructors #2767

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9a7c141
Replace from_legacy_cache method with init
SP1029 Sep 3, 2025
d6cba39
Add version based selection
SP1029 Sep 6, 2025
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
20 changes: 18 additions & 2 deletions src/peft/peft_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -786,14 +786,30 @@ def get_prompt(
past_key_values = new_cache
elif peft_config.num_transformer_submodules == 1:
# Dont' apply this to encoder-decoder models and not to models requiring special processing.
past_key_values = DynamicCache.from_legacy_cache(past_key_values)
# TODO: remove from_legacy_cache once transformers < 4.56 is dropped
transformers_lt_4_56 = packaging.version.parse(transformers.__version__) < packaging.version.parse(
"4.56.0.dev0"
)
if transformers_lt_4_56:
past_key_values = DynamicCache.from_legacy_cache(past_key_values)
else:
past_key_values = DynamicCache(past_key_values)

elif (peft_config.num_transformer_submodules == 2) and getattr(
self.base_model, "_supports_cache_class", True
):
# Dont' apply this to encoder-decoder models that don't support new Cache format yet
# If we don't apply this, prefix-tuning fails to update cross-attn cache
# TODO: remove check for _supports_cache_class once transformers 4.53 is no longer supported
past_key_values = EncoderDecoderCache.from_legacy_cache(past_key_values)
# TODO: remove from_legacy_cache once transformers < 4.56 is dropped
transformers_lt_4_56 = packaging.version.parse(transformers.__version__) < packaging.version.parse(
"4.56.0.dev0"
)
if transformers_lt_4_56:
past_key_values = EncoderDecoderCache.from_legacy_cache(past_key_values)
else:
past_key_values = EncoderDecoderCache(past_key_values)

past_key_values.cross_attention_cache = DynamicCache()
# invalidate the cross attention cache, since we add virtual tokens to the encoder
for key in past_key_values.is_updated.keys():
Expand Down
Loading