SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2203.09747 · 2022

Efficient Split-Mix Federated Learning for On-Demand and In-Situ Customization

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 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 10 have no contract verdict; 0 failures of our instrument; 1 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/2203.09747.json · badge: /api/ran/2203.09747.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
illidanlab/SplitMix canonical (classifier label)
classifier basis not recorded on the link
10 ran · 1 no recorded run
FunctionStatusWhere it lives
count_params_by_state Ran
ran on a synthesized input
illidanlab/SplitMix/nets/profile_func.py
code served (permissive licence) · get_code("624f5821bd9f2daa")
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 int, ndim 0, dtype family int, 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 ModelWithStateSize(nn.Module):
    """Model that implements state_size() method."""
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
        self.conv = nn.Conv1d(3, 5, kernel_size=2)
    
    def state_size(self):
        """Return total number of parameters."""
        total = 0
        for p in self.parameters():
            total += p.numel()
        return total
    
    def forward(self, x):
        return x

class ModelWithoutStateSize(nn.Module):
    """Model without state_size() method, uses state_dict()."""
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
        self.bias = nn.Parameter(torch.randn(2))
    
    def forward(self, x):
        return x

# Test with model that has state_size method
model_with_state_size = ModelWithStateSize()

# Test with model that doesn't have state_size method
model_without_state_size = ModelWithoutStateSize()

ARGS = {
    "model": model_with_state_size
}
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 int, ndim 0, dtype family int, 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
counter_norm Ran
ran on a synthesized input
illidanlab/SplitMix/nets/thop_op_hooks.py
code served (permissive licence) · get_code("c1f9c3bdd753a577")
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 1, dtype family float, shape (1,) 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

torch.manual_seed(0)

# The function takes input_size as a parameter
# Looking at the body: it's used in arithmetic (2 * input_size)
# and the result is wrapped in torch.DoubleTensor([...])
# So input_size should be a number (int or float)

input_size = 5

ARGS = {
    "input_size": input_size
}
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 1, dtype family float, shape (1,) — shape (1,) 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
counter_softmax Ran
ran on a synthesized input
illidanlab/SplitMix/nets/thop_op_hooks.py
code served (permissive licence) · get_code("fc1cb71e81ffc3fa")
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 1, dtype family float, shape (1,) 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

torch.manual_seed(0)

ARGS = {
    "batch_size": 2,
    "nfeatures": 4
}
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 1, dtype family float, shape (1,) — shape (1,) 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_bn_layer Ran
ran on a synthesized input
illidanlab/SplitMix/nets/bn_ops.py
code served (permissive licence) · get_code("f2666ef16161746f")
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 dict. 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)

# The function `get_bn_layer` takes a single parameter: bn_type (a string)
# It returns a dict with keys '1d' and '2d' that map to batch norm layer constructors
# We need to test it with valid bn_type values

# Valid bn_type options based on the code:
# - 'bn': returns standard PyTorch BatchNorm layers
# - 'none': returns Mock BatchNorm layers
# - 'd' + valid_type: returns DualNormLayer wrappers (but DualNormLayer not defined, so would fail)
# - 'fd' + valid_type: returns FilmDualNormLayer wrappers (but FilmDualNormLayer not defined, so would fail)

# The safest test is with 'bn' or 'none' since those are fully defined
ARGS = {
    'bn_type': 'bn'
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
get_model_fh Ran
ran on a synthesized input
illidanlab/SplitMix/fedavg.py
code served (permissive licence) · get_code("4fa0fb6d78f7d19d")
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 type. 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

# Create minimal stand-in model classes that match what the function expects
class DigitModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(784, 10)
    
    def forward(self, x):
        return self.fc(x.view(x.size(0), -1))

class AlexNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(256, 1000)
    
    def forward(self, x):
        return self.fc(x)

def resnet18(pretrained=False):
    class ResNet18(nn.Module):
        def __init__(self):
            super().__init__()
            self.fc = nn.Linear(512, 10)
        
        def forward(self, x):
            return self.fc(x)
    
    return ResNet18()

# Mock the module paths so imports work
import sys
from types import ModuleType

nets_module = ModuleType('nets')
nets_models_module = ModuleType('nets.models')
nets_models_module.DigitModel = DigitModel
nets_models_module.AlexNet = AlexNet
nets_module.models = nets_models_module

hetero_fl_module = ModuleType('nets.HeteFL')
preresne_module = ModuleType('nets.HeteFL.preresne')
preresne_module.resnet18 = resnet18
hetero_fl_module.preresne = preresne_module
nets_module.HeteFL = hetero_fl_module

sys.modules['nets'] = nets_module
sys.modules['nets.models'] = nets_models_module
sys.modules['nets.HeteFL'] = hetero_fl_module
sys.modules['nets.HeteFL.preresne'] = preresne_module

ARGS = {
    'data': 'Digits',
    'model': 'digit'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
nets, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type type — a type output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a type 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
get_slim_ratios_from_str Ran
ran on a synthesized input
illidanlab/SplitMix/nets/slimmable_models.py
code served (permissive licence) · get_code("2cfcc2cd96c72207")
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 ndarray, ndim 1, dtype family float, shape (8,) 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 numpy as np

# Test case 1: mode starting with 'ln'
ARGS_1 = {
    'mode': 'ln',
    'sort': True
}

# Test case 2: mode with simple integers
ARGS_2 = {
    'mode': '2-4-8',
    'sort': True
}

# Test case 3: mode with fractions (d notation)
ARGS_3 = {
    'mode': '1d2-1d4-3d4',
    'sort': False
}

# Choose one for execution
ARGS = ARGS_1
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (8,) — shape (8,) 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
is_film_dual_norm Ran
ran on a synthesized input
illidanlab/SplitMix/nets/bn_ops.py
code served (permissive licence) · get_code("8df4be770640f030")
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 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

torch.manual_seed(0)

ARGS = {
    "bn_type": "fd_norm"
}
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 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
kaiming_uniform_in_ Ran
ran on a synthesized input
illidanlab/SplitMix/nets/models.py
code served (permissive licence) · get_code("8bfb46d01decb53f")
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 2, dtype family float, shape (4, 3) 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
import math

torch.manual_seed(0)

# Create a tensor to initialize
# The function uses nn.init._calculate_correct_fan(tensor, mode)
# which needs a tensor with at least 2 dimensions
tensor = torch.randn(4, 3)

ARGS = {
    'tensor': tensor,
    'a': 0,
    'mode': 'fan_in',
    'scale': 1.0,
    'nonlinearity': 'leaky_relu'
}
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 2, dtype family float, shape (4, 3) — shape (4, 3) 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
parse_lognorm_slim_schedule Ran
ran on a synthesized input
illidanlab/SplitMix/nets/slimmable_models.py
code served (permissive licence) · get_code("42ad6079acf38b70")
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 ndarray, ndim 1, dtype family float, shape (100,) 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 numpy as np

# Analyzing the function:
# - train_slim_ratios: a list/array of numeric values (sorted internally)
# - mode: a string starting with 'ln' followed by parameters like 'ln{s}_{scale}'
# - client_num: an integer representing number of clients

# The function parses a lognormal distribution schedule
# train_slim_ratios is used with min(), sorted()
# mode is parsed by slicing [len('ln'):] and splitting by '_'
# client_num is used as an integer for multiplication and array size

ARGS = {
    'train_slim_ratios': [0.3, 0.5, 0.7, 0.9],
    'mode': 'ln0.5_1.0',
    'client_num': 100
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (100,) — shape (100,) 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
scale_init_param Ran
ran on a synthesized input
illidanlab/SplitMix/nets/models.py
code served (permissive licence) · get_code("62d1f031086e9acf")
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 math
import torch
import torch.nn as nn
from torch.nn.modules.conv import _ConvNd

torch.manual_seed(0)

# Create a simple Linear layer for testing
linear_layer = nn.Linear(in_features=4, out_features=3)

# Create a simple Conv2d layer for testing
conv_layer = nn.Conv2d(in_channels=4, out_channels=3, kernel_size=3)

ARGS = {
    'm': linear_layer,
    'scale_in': 1.0
}
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
get_model_fh Not yet run
unverified (harvested, no recorded run)
illidanlab/SplitMix/fed_hfl.py
code served (permissive licence) · get_code("92176cf63686ae8d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 10 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
counter_norm 1 other paper
1 later
2211.06088 (2022-11)
RepGhost: A Hardware-Efficient Ghost Module via Re-parameterization

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

Federated learning (FL) provides a distributed learning framework for multiple participants to collaborate learning without sharing raw data. In many practical FL scenarios, participants have heterogeneous resources due to disparities in hardware and inference dynamics that require quickly loading models of different sizes and levels of robustness. The heterogeneity and dynamics together impose significant challenges to existing FL approaches and thus greatly limit FL's applicability. In this paper, we propose a novel Split-Mix FL strategy for heterogeneous participants that, once training is done, provides in-situ customization of model sizes and robustness. Specifically, we achieve customization by learning a set of base sub-networks of different sizes and robustness levels, which are later aggregated on-demand according to inference requirements. This split-mix strategy achieves customization with high efficiency in communication, storage, and inference. Extensive experiments demonstrate that our method provides better in-situ customization than the existing heterogeneous-architecture FL methods. Codes and pre-trained models are available: https://github.com/illidanlab/SplitMix.

For agents

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

get_harvested_code_for_paper("2203.09747")
get_code_for_paper("2203.09747")
have("2203.09747")

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

curl https://syntology.ai/api/ran/2203.09747.json

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

[![Syntology run record](https://syntology.ai/api/ran/2203.09747.svg)](https://syntology.ai/paper/2203.09747)

Connect an agent — have() is free.