SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.04669 · ICML · 2026

Delving into Muon and Beyond: Deep Analysis and Extensions

Xianbiao Qi, Rong Xiao, Yelin He, Jiaquan Ye, Marco Chen

arXiv · PDF · Open in the Atlas

Code that ran

3 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 4 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.04669.json · badge: /api/ran/2602.04669.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
Ocram7/BeyondMuon canonical (classifier label)
classifier basis: census
3 ran · 4 no recorded run
FunctionStatusWhere it lives
build_norm Ran
ran; our contract draft was wrong, not the code
Ocram7/BeyondMuon/model.py
code served (permissive licence) · get_code("ee769a4ab47b0aca")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned LayerNorm. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.nn import functional as F

torch.manual_seed(0)

class LayerNorm(nn.Module):
    """LayerNorm with affine scale and optional bias."""

    def __init__(self, ndim, bias):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(ndim))
        self.bias = nn.Parameter(torch.zeros(ndim)) if bias else None

    def forward(self, x):
        return F.layer_norm(x, self.weight.shape, self.weight, self.bias, 1e-05)

def build_norm(config, norm_name, *, dim=None):
    if norm_name == 'identity':
        return nn.Identity()
    if norm_name == 'layernorm':
        return LayerNorm(dim or config.n_embd, bias=config.bias)
    raise ValueError(f'Unsupported norm: {norm_name}')

# Create a minimal config class with required attributes
class MinimalConfig:
    def __init__(self, n_embd=4, bias=True):
        self.n_embd = n_embd
        self.bias = bias

config = MinimalConfig(n_embd=4, bias=True)

ARGS = {
    'config': config,
    'norm_name': 'layernorm',
    'dim': 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LayerNorm — a LayerNorm output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LayerNorm output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
build_parameter_groups Ran
ran; our contract draft was wrong, not the code
Ocram7/BeyondMuon/optimizer_factory.py
code served (permissive licence) · get_code("d5add16393b18146")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        # Matrix parameters (dim >= 2)
        self.linear1 = nn.Linear(3, 4)
        self.linear2 = nn.Linear(5, 2)
        # Vector parameters (dim < 2)
        self.bias_extra = nn.Parameter(torch.randn(6))
        self.scale = nn.Parameter(torch.randn(1))

model = SimpleModel()

weight_decay = 0.01
lr_matrix = 0.001
lr_vector = 0.01

ARGS = {
    "model": model,
    "weight_decay": weight_decay,
    "lr_matrix": lr_matrix,
    "lr_vector": lr_vector,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
resolve_optimizer_spec Ran
ran; our contract draft was wrong, not the code
Ocram7/BeyondMuon/optimizer_factory.py
code served (permissive licence) · get_code("736d7a0145b97da9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned OptimizerSpec. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from dataclasses import dataclass

torch.manual_seed(0)

@dataclass(frozen=True)
class OptimizerSpec:
    name: str
    family: str
    spectral_exponent: float

OPTIMIZER_SPECS = {'adam': OptimizerSpec('Adam', 'adam', 1.0), 'adams': OptimizerSpec('AdamS', 'adam', 0.5), 'adamq': OptimizerSpec('AdamQ', 'adam', 0.25), 'adamz': OptimizerSpec('AdamZ', 'adam', 0.0), 'msgd': OptimizerSpec('mSGD', 'msgd', 1.0), 'msgds': OptimizerSpec('mSGDS', 'msgd', 0.5), 'msgdq': OptimizerSpec('mSGDQ', 'msgd', 0.25), 'msgdz': OptimizerSpec('mSGDZ', 'msgd', 0.0)}

def _normalize_name(name):
    return (name or '').replace('_', '').replace('-', '').replace(' ', '').lower()

def resolve_optimizer_spec(optimizer_variant):
    variant_key = _normalize_name(optimizer_variant)
    if variant_key not in OPTIMIZER_SPECS:
        choices = ', '.join((spec.name for spec in OPTIMIZER_SPECS.values()))
        raise ValueError(f'Unknown optimizer_variant={optimizer_variant!r}; choose: {choices}')
    return OPTIMIZER_SPECS[variant_key]

ARGS = {
    'optimizer_variant': 'adam'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type OptimizerSpec — a OptimizerSpec output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a OptimizerSpec output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
coupled_newton_schulz_sqrt_invsqrt Not yet run
unverified (harvested, no recorded run)
Ocram7/BeyondMuon/optimizers/spectral_ns.py
code served (permissive licence) · get_code("b63182389c6302a0")
How it ran

no invocation recorded.

maybe_wrap_zero Not yet run
unverified (harvested, no recorded run)
Ocram7/BeyondMuon/optimizer_factory.py
code served (permissive licence) · get_code("dc340d50a2e38871")
How it ran

no invocation recorded.

muon_update Not yet run
unverified (harvested, no recorded run)
Ocram7/BeyondMuon/optimizers/spectral_ns.py
code served (permissive licence) · get_code("7e5bfd49da2b921d")
How it ran

no invocation recorded.

zero_power_via_quintic_ns Not yet run
unverified (harvested, no recorded run)
Ocram7/BeyondMuon/optimizers/spectral_ns.py
code served (permissive licence) · get_code("146c439e2209ff8c")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 7 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

The Muon optimizer has recently attracted considerable attention for its strong empirical performance and use of orthogonalized updates on matrix-shaped parameters, yet its underlying mechanisms and relationship to adaptive optimizers such as Adam remain insufficiently understood. In this work, we aim to address these questions through a unified spectral perspective. Specifically, we view Muon as the p = 0 endpoint of a family of spectral transformations of the form U Σ p V ⊤ , and consider additional variants with p = 1 2 , p = 1 4 , and p = 1. These transformations are applied to both first-moment updates, as in momentum SGD, and to root-mean-square (RMS) normalized gradient updates as in Adam. To enable efficient computation, we develop a coupled Newton iteration that avoids explicit singular value decomposition. Across controlled experiments, we find that RMS-normalized updates yield more stable optimization than first-moment updates. Moreover, while spectral compression provides strong stabilization benefits under firstmoment updates, the Muon update (p = 0) does not consistently outperform Adam. These results suggest that Muon is best understood as an effective form of spectral normalization, but not a universally superior optimization method. Our source code will be released at https: //github.com/Ocram7/BeyondMuon.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2602.04669")
get_code_for_paper("2602.04669")
have("2602.04669")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2602.04669.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2602.04669.svg)](https://syntology.ai/paper/2602.04669)

Connect an agent — have() is free.