Skip to content

Fix loading of BF16 bare nn.Parameters (LTX-2 learnable_registers) #467

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

Open
hc20k wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix loading of BF16 bare nn.Parameters (LTX-2 learnable_registers) #467

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
6 changes: 6 additions & 0 deletions loader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,6 +142,12 @@ def gguf_sd_loader(path, handle_prefix="model.diffusion_model.", is_text_model=F
if len(shape) <= 1 and tensor.tensor_type == gguf.GGMLQuantizationType.BF16:
state_dict[sd_key] = dequantize_tensor(state_dict[sd_key], dtype=torch.float32)

# GGMLLayer only intercepts weight/bias, so anything else (e.g. bare
# nn.Parameters such as LTX-2's learnable_registers) reaches torch's
# default load path and must already be a real float tensor.
elif not sd_key.endswith((".weight", ".bias")) and is_quantized(state_dict[sd_key]):
state_dict[sd_key] = dequantize_tensor(state_dict[sd_key], dtype=torch.float32)

# keep track of loaded tensor types
tensor_type_str = getattr(tensor.tensor_type, "name", repr(tensor.tensor_type))
qtype_dict[tensor_type_str] = qtype_dict.get(tensor_type_str, 0) + 1
Expand Down