Skip to content
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

fix: Unexpected tracer in muon optimizer when sharding. #1193

Merged
merged 1 commit into from
Feb 24, 2025
Merged

fix: Unexpected tracer in muon optimizer when sharding. #1193

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
27 changes: 19 additions & 8 deletions optax/contrib/_muon.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,6 +97,7 @@ class MuonState(NamedTuple):
"""State for the Adam algorithm."""
count: chex.Array # shape=(), dtype=jnp.int32.
mu: base.Updates
ns_coeffs: chex.Array # shape=(), dtype=jnp.int32.


def scale_by_muon(
Expand Down Expand Up @@ -142,15 +143,19 @@ def scale_by_muon(
<https://arxiv.org/abs/2409.20325>`_, 2024
"""
mu_dtype = utils.canonicalize_dtype(mu_dtype)
ns_coeffs_ = jnp.asarray(ns_coeffs)
if ns_coeffs_.ndim > 2 or ns_coeffs_.shape[-1] != 3:
raise ValueError(
f'ns_coeffs must have shape (3,) or (n, 3), got {ns_coeffs_.shape}'
)

def init_fn(params):
mu = otu.tree_zeros_like(params, dtype=mu_dtype) # First moment
return MuonState(count=jnp.zeros([], jnp.int32), mu=mu)
ns_coeffs_ = jnp.asarray(ns_coeffs)
if ns_coeffs_.ndim > 2 or ns_coeffs_.shape[-1] != 3:
raise ValueError(
f'ns_coeffs must have shape (3,) or (n, 3), got {ns_coeffs_.shape}'
)
return MuonState(
count=jnp.zeros([], jnp.int32),
mu=mu,
ns_coeffs=ns_coeffs_,
)

def update_fn(updates, state, params=None):
del params
Expand All @@ -168,7 +173,9 @@ def update_fn(updates, state, params=None):
mu_hat = otu.tree_bias_correction(mu, beta, count_inc)
# Apply Newton-schulz orthogonalization.
updates = jax.tree.map(
lambda x: orthogonalize_via_newton_schulz(x, ns_coeffs_, ns_steps, eps),
lambda x: orthogonalize_via_newton_schulz(
x, state.ns_coeffs, ns_steps, eps
),
mu_hat,
)
if adaptive:
Expand All @@ -178,7 +185,11 @@ def update_fn(updates, state, params=None):
lambda x, y: jnp.einsum('ij,ij,ab->ab', x, y, y), mu_hat, updates
)
mu = otu.tree_cast(mu, mu_dtype)
return updates, MuonState(count=count_inc, mu=mu)
return updates, MuonState(
count=count_inc,
mu=mu,
ns_coeffs=state.ns_coeffs,
)
return base.GradientTransformation(init_fn, update_fn)


Expand Down