Skip to content

Extend TorchAOConfig to support mobile usecases #3587

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
merged 3 commits into from
Nov 14, 2025
+144 −36
Merged

Extend TorchAOConfig to support mobile usecases #3587

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c2ac422
up
metascroy Nov 11, 2025
f71bfa4
up
metascroy Nov 11, 2025
c30f1b8
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Nov 11, 2025
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
up
  • Loading branch information
@metascroy
metascroy committed Nov 11, 2025
commit f71bfa400eada6a2cb2fdcc9e313aec96fc65798
15 changes: 1 addition & 14 deletions unsloth/models/_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2009,19 +2009,6 @@ def error_out_no_vllm(*args, **kwargs):
AOBaseConfig = None
Int4WeightOnlyConfig = None



@dataclass
class TorchAOConfig:
qat_scheme: str = "int4"
base_config_and_filter_fns: List[
Tuple["AOBaseConfig", Optional[Callable]]
] = field(
default_factory=lambda: [
(Int4WeightOnlyConfig(group_size=128), None)
]
)

@dataclass
class TorchAOConfig:
qat_scheme: str = "int4"
Expand Down Expand Up @@ -2070,7 +2057,7 @@ def _untie_input_output_embeddings(model: torch.nn.Module) -> None:
if out_proj.weight.data_ptr() == in_emb.weight.data_ptr():
with torch.no_grad():
W = in_emb.weight.detach().clone()
out_proj.weight = nn.Parameter(W) # new storage, keeps dtype/device
out_proj.weight = torch.nn.Parameter(W) # new storage, keeps dtype/device

# 4) Prevent future automatic re-tying
def _no_tie(self):
Expand Down
Loading