Skip to content

[Misc][Quantization] Clarify the intent of GGUF FusedMoE weight materialization #30310

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
a4lg wants to merge 1 commit into
base: main
Choose a base branch
from
+5 −2
Open

[Misc][Quantization] Clarify the intent of GGUF FusedMoE weight materialization #30310

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
7 changes: 5 additions & 2 deletions vllm/model_executor/layers/fused_moe/layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1207,10 +1207,13 @@ def weight_loader(
if full_load:
shard_dim += 1

# Materialize GGUF UninitializedParameter
# Materialize GGUF UninitializedParameter accounting merged weights
if is_gguf_weight and isinstance(param, UninitializedParameter):
# GGUF currently requires full load (3D tensors).
assert full_load
Copy link
Member

@Isotr0py Isotr0py Dec 9, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Perhaps add a message to clarify what's happening when assertion is false?

Copy link
Contributor Author

@a4lg a4lg Dec 9, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm, rather, it would be better to improve the comment (i.e. what is truly necessary). I'll consider changing around this line.

final_shape = list(loaded_weight.shape)
if shard_id in ["w1", "w3"]:
# w1 and w3 are merged per expert.
if shard_id in {"w1", "w3"}:
final_shape[1] *= 2
final_shape[shard_dim] = final_shape[shard_dim] // self.tp_size
param.materialize(final_shape, dtype=loaded_weight.dtype)
Expand Down