SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2308.14929 · ICML · 2024

Maestro: Uncovering Low-Rank Structures via Trainable Decomposition

Hongyi Wang, Samuel Horváth, Stefanos Laskaridis, Shashank Rajput

arXiv · PDF · Open in the Atlas

Code that ran

15 of 18 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 2 violated it and 1 has no contract verdict; 9 failures of our instrument; 3 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2308.14929.json · badge: /api/ran/2308.14929.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
samuelhorvath/maestro-lod canonical (classifier label)
classifier basis: classifier_2025
8 ran · 0 no recorded run
SamuelHorvath/Maestro-LoD canonical (classifier label)
classifier basis: census
7 ran · 3 no recorded run
FunctionStatusWhere it lives
MaestroConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
samuelhorvath/maestro-lod/maestro/layers/conv.py
code served (permissive licence) · get_code("838b3ced47c739f8")
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 an instance of its own class, MaestroConv2d: the run constructed the object and did not call it, so no result was computed. 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
import numpy as np

torch.manual_seed(0)

ARGS = {
    'in_features': 4,
    'out_features': 3,
    'kernel_size': 3,
    'stride': 1,
    'padding': 0,
    'dilation': 1,
    'groups': 1,
    'bias': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MaestroConv2d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MaestroConv2d 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
bool_string Ran
ran, violated the contract we drafted
SamuelHorvath/Maestro-LoD/maestro_opts.py
code served (permissive licence) · get_code("396d2b49a5391613")
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) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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

def bool_string(s):
    if s not in {'False', 'True'}:
        raise ValueError('Not a valid boolean string')
    return s == 'True'

ARGS = {
    's': 'True'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
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
check_layer Ran
ran, honoured the contract we drafted
samuelhorvath/maestro-lod/maestro/layers/conv.py
code served (permissive licence) · get_code("a80d13a84b21f1cf")
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 float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 SimpleLayer(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

layer = SimpleLayer()

# Test case 1: p is just a plain value (not a tuple)
p_simple = 0.5

# Test case 2: p is a tuple with (value, layer)
p_tuple = (0.5, layer)

ARGS = {
    "layer": layer,
    "p": p_tuple
}
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 float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
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
create_conv2d_layer Ran
ran; our contract draft was wrong, not the code
SamuelHorvath/Maestro-LoD/maestro/models/lenet.py
code served (permissive licence) · get_code("18a414ece7a6c3c2")
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 Conv2d. 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

torch.manual_seed(0)

# Define a minimal MaestroConv2d class that mimics nn.Conv2d interface
class MaestroConv2d(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True):
        super().__init__()
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.kernel_size = kernel_size if isinstance(kernel_size, tuple) else (kernel_size, kernel_size)
        self.stride = stride
        self.padding = padding
        self.dilation = dilation
        self.groups = groups
        self.weight = nn.Parameter(torch.randn(out_channels, in_channels // groups, *self.kernel_size))
        if bias:
            self.bias = nn.Parameter(torch.randn(out_channels))
        else:
            self.bias = None
    
    def forward(self, x):
        return torch.nn.functional.conv2d(x, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups)

ARGS = {
    'decomposition': False,
    'in_channels': 3,
    'out_channels': 16,
    'kernel_size': 3,
    'stride': 1,
    'padding': 1,
    'dilation': 1,
    'groups': 1,
    'bias': True
}
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 Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d 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
create_linear_layer Ran
ran; our contract draft was wrong, not the code
SamuelHorvath/Maestro-LoD/maestro/models/utils.py
code served (permissive licence) · get_code("33eb693ad533396c")
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 Linear. 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

torch.manual_seed(0)

class MaestroLinear(nn.Module):
    def __init__(self, in_features, out_features, bias=True):
        super().__init__()
        self.in_features = in_features
        self.out_features = out_features
        self.weight = nn.Parameter(torch.randn(out_features, in_features))
        if bias:
            self.bias = nn.Parameter(torch.randn(out_features))
        else:
            self.register_parameter('bias', None)
    
    def forward(self, x):
        return torch.nn.functional.linear(x, self.weight, self.bias)

ARGS = {
    "decomposition": False,
    "in_features": 4,
    "out_features": 3,
    "bias": True
}
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 Linear — a Linear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Linear 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
create_norm_layer Ran
ran; our contract draft was wrong, not the code
samuelhorvath/maestro-lod/cifar_mnist_main.py
code served (permissive licence) · get_code("31759f14336427d4")
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) and returned BatchNorm2d. 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

torch.manual_seed(0)

# Test case 1: BatchNorm enabled
ARGS = {
    "num_channels": 64,
    "batch_norm": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type BatchNorm2d — a BatchNorm2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a BatchNorm2d 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
decompose_conv2d Ran
ran; our contract draft was wrong, not the code
SamuelHorvath/Maestro-LoD/maestro/layers/conv.py
code served (permissive licence) · get_code("11a9508cc4b6b3fa")
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 MaestroConv2d. 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
import numpy as np

torch.manual_seed(0)

# Create a simple Conv2d layer to decompose
# in_channels=3, out_channels=4, kernel_size=3x3
conv_layer = nn.Conv2d(
    in_channels=3,
    out_channels=4,
    kernel_size=3,
    stride=1,
    padding=1,
    dilation=1,
    groups=1,
    bias=True
)

ARGS = {
    "layer": conv_layer,
    "device": "cpu"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MaestroConv2d — a MaestroConv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MaestroConv2d 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
decompose_linear Ran
ran; our contract draft was wrong, not the code
samuelhorvath/maestro-lod/maestro/layers/linear.py
code served (permissive licence) · get_code("db892158fa6646a4")
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) and returned MaestroLinear. 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

torch.manual_seed(0)

# Create a small nn.Linear layer to decompose
# Use distinct dimensions: in_features=4, out_features=3
layer = nn.Linear(in_features=4, out_features=3, bias=True)

# Initialize with some data so SVD works well
layer.weight.data = torch.randn(3, 4)
layer.bias.data = torch.randn(3)

ARGS = {
    "layer": layer,
    "device": None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MaestroLinear — a MaestroLinear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MaestroLinear 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
do_not_decompose_linear Ran
ran, violated the contract we drafted
samuelhorvath/maestro-lod/maestro/layers/linear.py
code served (permissive licence) · get_code("db6320dbc007c7bd")
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) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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.parameter import Parameter
import numpy as np

torch.manual_seed(0)

# Create a MaestroLinear layer with specific dimensions
# in_features=5, out_features=3, so inner_dim=min(5,3)=3
layer = nn.Module()
layer.in_features = 5
layer.out_features = 3
layer.inner_dim = min(5, 3)

# Create the in_network and out_network as the function expects
layer.in_network = nn.Linear(5, 3, bias=False)
layer.out_network = nn.Linear(3, 3, bias=True)

# Initialize with some weights
layer.in_network.weight = Parameter(torch.randn(3, 5))
layer.out_network.weight = Parameter(torch.randn(3, 3))
layer.out_network.bias = Parameter(torch.randn(3))

ARGS = {
    "layer": layer
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
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
get_submodel_and_name Ran
ran; our contract draft was wrong, not the code
SamuelHorvath/Maestro-LoD/maestro/layers/decomposition.py
code served (permissive licence) · get_code("54c0134104fb7718")
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 tuple, 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__()
        self.layer1 = nn.Linear(4, 3)
        self.layer2 = nn.Sequential(
            nn.Linear(3, 2),
            nn.ReLU()
        )
        self.nested = nn.Module()
        self.nested.sublayer = nn.Linear(2, 1)

model = SimpleModel()

ARGS = {
    "model": model,
    "name": "layer1.weight"
}
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 tuple, ndim 1, dtype family float — a tuple 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
group_lasso_criterion Ran
ran, honoured the contract we drafted
SamuelHorvath/Maestro-LoD/maestro/layers/utils.py
code served (permissive licence) · get_code("88bb5d3e4db2e20f")
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 Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 ModuleWithGroupLasso(nn.Module):
    def __init__(self):
        super().__init__()
        self.inner_dim = 4
        self.weight = nn.Parameter(torch.randn(4, 3))
    
    def group_lasso(self, hierarchical=False):
        # Return a tensor that can be summed
        # Simulate group lasso regularization
        if hierarchical:
            return torch.abs(self.weight).sum(dim=1, keepdim=True)
        else:
            return torch.abs(self.weight).sum(dim=1, keepdim=True)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.layer1 = ModuleWithGroupLasso()
        self.layer2 = ModuleWithGroupLasso()
        self.linear = nn.Linear(3, 2)
    
    def forward(self, x):
        return x

model = SimpleModel()

ARGS = {
    'model': model,
    'hierarchical': False
}
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 Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
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
initialise_model Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
samuelhorvath/maestro-lod/cifar_mnist_main.py
code served (permissive licence) · get_code("fb70500cc298c8d1")
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) and returned tuple. 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

torch.manual_seed(0)

# Create a minimal args object with batch_norm attribute
class Args:
    def __init__(self, batch_norm=True):
        self.batch_norm = batch_norm

# Create minimal stand-in models for the imports that would fail
class MockResNet18(nn.Module):
    def __init__(self, norm_layer):
        super().__init__()
        self.norm_layer = norm_layer
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3)
        self.bn1 = norm_layer(64)
        self.relu = nn.ReLU(inplace=True)
    
    def forward(self, x):
        return self.relu(self.bn1(self.conv1(x)))

class MockVGG19(nn.Module):
    def __init__(self, norm_layer):
        super().__init__()
        self.norm_layer = norm_layer
        self.conv1 = nn.Conv2d(3, 64, kernel_size=3, padding=1)
        self.bn1 = norm_layer(64)
    
    def forward(self, x):
        return self.bn1(self.conv1(x))

class MockMaestroLeNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 32, kernel_size=5)
    
    def forward(self, x):
        return self.conv1(x)

# Monkey-patch the imports to return our mock models
import sys
from types import ModuleType

maestro_models_resnets = ModuleType('maestro.models.resnets')
maestro_models_resnets.resnet18 = lambda norm_layer: MockResNet18(norm_layer)

maestro_models_vggs = ModuleType('maestro.models.vggs')
maestro_models_vggs.VGG = lambda arch, norm_layer: MockVGG19(norm_layer)

maestro_models_lenet = ModuleType('maestro.models.lenet')
maestro_models_lenet.MaestroLeNet = MockMaestroLeNet

sys.modules['maestro'] = ModuleType('maestro')
sys.modules['maestro.models'] = ModuleType('maestro.models')
sys.modules['maestro.models.resnets'] = maestro_models_resnets
sys.modules['maestro.models.vggs'] = maestro_models_vggs
sys.modules['maestro.models.lenet'] = maestro_models_lenet

ARGS = {
    'model_str': 'resnet18',
    'device': 'cpu',
    'args': Args(batch_norm=True)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
maestro, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
maestro_to_full_linear Ran
ran; our contract draft was wrong, not the code
samuelhorvath/maestro-lod/maestro/layers/linear.py
code served (permissive licence) · get_code("c4f3e05b55394eae")
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) and returned Linear. 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.parameter import Parameter

torch.manual_seed(0)

# Create a MaestroLinear layer with small dimensions
in_features = 4
out_features = 3
inner_dim = min(in_features, out_features)

maestro_layer = nn.Module()
maestro_layer.in_features = in_features
maestro_layer.out_features = out_features
maestro_layer.inner_dim = inner_dim

# Create the inner and outer networks
maestro_layer.in_network = nn.Linear(in_features, inner_dim, bias=False)
maestro_layer.out_network = nn.Linear(inner_dim, out_features, bias=True)

# Initialize with some weights
maestro_layer.in_network.weight.data = torch.randn(inner_dim, in_features)
maestro_layer.out_network.weight.data = torch.randn(out_features, inner_dim)
maestro_layer.out_network.bias.data = torch.randn(out_features)

# Create low_rank mask
maestro_layer.low_rank = torch.ones(inner_dim, dtype=torch.bool)

# Convert to MaestroLinear type for proper typing
from torch.nn.modules.module import Module

class MaestroLinear(Module):
    __constants__ = ['in_features', 'out_features']
    in_features: int
    out_features: int

    def __init__(self, in_features: int, out_features: int, bias: bool=True) -> None:
        super(MaestroLinear, self).__init__()
        self.in_features = in_features
        self.out_features = out_features
        self.inner_dim = min(self.in_features, self.out_features)
        self.in_network = nn.Linear(self.in_features, self.inner_dim, bias=False)
        self.out_network = nn.Linear(self.inner_dim, self.out_features, bias=bias)
        self.reset_parameters()
        self.low_rank = (torch.ones(self.inner_dim) >= 0).cpu()
        self.is_od = True

    def reset_parameters(self) -> None:
        self.in_network.reset_parameters()
        self.out_network.reset_parameters()

maestro_layer = MaestroLinear(in_features=4, out_features=3, bias=True)

ARGS = {
    'layer': maestro_layer,
    'device': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Linear — a Linear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Linear 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
parse_maestro_opts Ran
ran; our contract draft was wrong, not the code
SamuelHorvath/Maestro-LoD/maestro_opts.py
code served (permissive licence) · get_code("4998b554a5d3b133")
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 ArgumentParser. 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 argparse
import time

torch_seed = 0

ARGS = {
    'parser': argparse.ArgumentParser(description='Test parser for parse_maestro_opts')
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArgumentParser — a ArgumentParser output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ArgumentParser 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
validate Ran
ran, honoured the contract we drafted
samuelhorvath/maestro-lod/imagenet_main.py
code served (permissive licence) · get_code("c63928dac29e7a1d")
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) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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
from torch.utils.data import DataLoader, TensorDataset
import torch.distributed as dist

torch.manual_seed(0)

# Create a minimal model that accepts images and val_od_sampler
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 10)
    
    def forward(self, x, val_od_sampler=None):
        return self.linear(x)

# Create a minimal criterion
criterion = nn.CrossEntropyLoss()

# Create minimal val_loader with (images, target) tuples
images = torch.randn(2, 4)  # batch_size=2, features=4
targets = torch.randint(0, 10, (2,))  # 10 classes
dataset = TensorDataset(images, targets)
val_loader = DataLoader(dataset, batch_size=2)

# Create a minimal args object
class Args:
    def __init__(self):
        self.gpu = None
        self.print_freq = 1
        self.distributed = False
        self.world_size = 1

args = Args()

# Create model
model = MinimalModel()

# Create val_od_sampler (can be None or a simple object)
val_od_sampler = None

ARGS = {
    'val_loader': val_loader,
    'model': model,
    'criterion': criterion,
    'args': args,
    'val_od_sampler': val_od_sampler
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple 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
create_linear_layer Not yet run
unverified (harvested, no recorded run)
SamuelHorvath/Maestro-LoD/maestro/models/lenet.py
code served (permissive licence) · get_code("752dd8c48fb5d47b")
How it ran

no invocation recorded.

do_not_decompose_conv2d Not yet run
unverified (harvested, no recorded run)
SamuelHorvath/Maestro-LoD/maestro/layers/conv.py
code served (permissive licence) · get_code("a02b80587a98bf46")
How it ran

no invocation recorded.

maestro_to_full_conv2d Not yet run
unverified (harvested, no recorded run)
SamuelHorvath/Maestro-LoD/maestro/layers/conv.py
code served (permissive licence) · get_code("991b739d35882910")
How it ran

no invocation recorded.

Also present in other papers' code

Of 18 samples lifted from this paper's linked repositories, 18 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 16 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.

FunctionAlso present inEarliest-datedOthers named
bool_string 1 other paper
1 dated earlier
2007.05084 (2020-07)
Attack of the Tails: Yes, You Really Can Backdoor Federated Learning

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

Deep Neural Networks (DNNs) have been a large driver for AI breakthroughs in recent years. However, these models have been getting increasingly large as they become more accurate and safe. This means that their training becomes increasingly costly and time-consuming and typically yields a single model to fit all targets. Various techniques have been proposed in the literature to mitigate this, including pruning, sparsification, or quantization of model weights and updates. While achieving high compression rates, they often incur significant computational overheads at training or lead to non-negligible accuracy penalty. Alternatively, factorization methods have been leveraged for low-rank compression of DNNs. Similarly, such techniques (e.g., SVD) frequently rely on heavy iterative decompositions of layers and are potentially sub-optimal for non-linear models, such as DNNs. We take a further step in designing efficient low-rank models and propose MAESTRO, a framework for trainable low-rank layers. Instead of iteratively applying a priori decompositions, the low-rank structure is baked into the training process through LOD, a low-rank ordered decomposition. Not only is this the first time importance ordering via sampling is applied on the decomposed DNN structure, but it also allows selecting ranks at a layer granularity. Our theoretical analysis demonstrates that in special cases LOD recovers the SVD decomposition and PCA . Applied to DNNs, MAESTRO enables the extraction of lower footprint models that preserve performance. Simultaneously, it enables the graceful trade-off between accuracy-latency for deployment to even more constrained devices without retraining.

For agents

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

get_harvested_code_for_paper("2308.14929")
get_code_for_paper("2308.14929")
have("2308.14929")

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

curl https://syntology.ai/api/ran/2308.14929.json

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

[![Syntology run record](https://syntology.ai/api/ran/2308.14929.svg)](https://syntology.ai/paper/2308.14929)

Connect an agent — have() is free.