Skip to content

Fix tie_word_embeddings handling for GGUF models #35085

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
SunMarc merged 3 commits into from
Dec 5, 2024
Merged

Fix tie_word_embeddings handling for GGUF models #35085

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
cedbced
fix tie_word_embeddings
Isotr0py Dec 4, 2024
e6c7862
fix
Isotr0py Dec 4, 2024
eba02e2
Merge branch 'main' into fix-llama-gguf
Isotr0py Dec 4, 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
7 changes: 6 additions & 1 deletion src/transformers/modeling_gguf_pytorch_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -291,7 +291,6 @@ def load_gguf_checkpoint(gguf_checkpoint_path, return_tensors=False):
# FIXME: Currnetly this implementation is only for flan-t5 architecture.
# It needs to be developed for supporting legacy t5.
elif "t5" in architecture or "t5encoder" in architecture:
parsed_parameters["config"]["tie_word_embeddings"] = False
parsed_parameters["config"]["is_gated_act"] = True
updated_architecture = "t5"
else:
Expand Down Expand Up @@ -326,6 +325,12 @@ def load_gguf_checkpoint(gguf_checkpoint_path, return_tensors=False):
if architecture + model_size not in GGUF_SUPPORTED_ARCHITECTURES:
raise ValueError(f"Architecture {architecture + model_size} not supported")

# Handle tie_word_embeddings, if lm_head.weight is not present in tensors,
# tie_word_embeddings is true otherwise false
parsed_parameters["config"]["tie_word_embeddings"] = all(
"output.weight" != tensor.name for tensor in reader.tensors
)

# List all key-value pairs in a columnized format
for gguf_key, field in reader.fields.items():
gguf_key = gguf_key.replace(architecture, updated_architecture)
Expand Down