Skip to content

Update defaults for NorMuon #54

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
JohnLangford merged 3 commits into from
Apr 8, 2026
Merged

Update defaults for NorMuon #54

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f16084d
normuon default: adjust_lr = "spectral_norm"
Apr 7, 2026
199894b
default: use_gram_newton_schulz=True, use_polar_express=True
Apr 7, 2026
a5fd82e
Add clear error message when gram-newton-schulz package is missing
JohnLangford Apr 8, 2026
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
4 changes: 2 additions & 2 deletions dion/dion2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,9 +59,9 @@ def __init__(
epsilon: float = 1e-8,
adjust_lr: Optional[str] = "spectral_norm",
flatten: bool = False,
use_gram_newton_schulz: bool = False,
use_triton: bool = False,
use_polar_express: bool = False,
use_polar_express: bool = True,
use_gram_newton_schulz: bool = True,
newton_schulz_func: Optional[Callable] = None,
verbose: bool = False,
):
Expand Down
13 changes: 10 additions & 3 deletions dion/megabatch_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,9 +34,9 @@ def __init__(
distributed_mesh: Optional[Union[DeviceMesh, ProcessGroup]],
algo_name: str,
defaults: dict,
use_gram_newton_schulz: bool = False,
use_gram_newton_schulz: bool = True,
use_triton: bool = False,
use_polar_express: bool = False,
use_polar_express: bool = True,
newton_schulz_func: Optional[Callable] = None,
):
super().__init__(params, defaults)
Expand Down Expand Up @@ -75,7 +75,14 @@ def __init__(
)
self._newton_schulz_func = newton_schulz_func
elif use_gram_newton_schulz:
from gram_newton_schulz import GramNewtonSchulz
try:
from gram_newton_schulz import GramNewtonSchulz
except ImportError:
raise ImportError(
"use_gram_newton_schulz=True requires the 'gram-newton-schulz' package, "
"which is not installed. "
"Install it with: pip install gram-newton-schulz"
)
use_polar_express = True
_gns = GramNewtonSchulz(
ns_use_kernels=use_triton,
Expand Down
4 changes: 2 additions & 2 deletions dion/muon.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,9 +61,9 @@ def __init__(
nesterov: bool = False,
adjust_lr: Optional[str] = "spectral_norm",
flatten: bool = False,
use_gram_newton_schulz: bool = False,
use_gram_newton_schulz: bool = True,
use_triton: bool = False,
use_polar_express: bool = False,
use_polar_express: bool = True,
newton_schulz_func: Optional[Callable] = None,
):
if lr < 0.0:
Expand Down
6 changes: 3 additions & 3 deletions dion/normuon.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,11 +62,11 @@ def __init__(
cautious_wd: bool = False,
epsilon: float = 1e-8,
nesterov: bool = False,
adjust_lr: Optional[str] = "rms_norm",
adjust_lr: Optional[str] = "spectral_norm",
flatten: bool = False,
use_gram_newton_schulz: bool = False,
use_gram_newton_schulz: bool = True,
use_triton: bool = False,
use_polar_express: bool = False,
use_polar_express: bool = True,
newton_schulz_func: Optional[Callable] = None,
):
if lr < 0.0:
Expand Down