SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1904.02877 · 2019

Single-Path NAS: Designing Hardware-Efficient ConvNets in less than 4 Hours

arXiv · PDF · Open in the Atlas

Code that ran

4 of 16 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; 1 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 0 failures of our instrument; 12 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/1904.02877.json · badge: /api/ran/1904.02877.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
dstamoulis/single-path-nas canonical (classifier label)
classifier basis: census
4 ran · 12 no recorded run
FunctionStatusWhere it lives
convnet_encoder Ran
ran on a synthesized input
dstamoulis/single-path-nas/nas-search/plot-progress/parse_search_output.py
code served (permissive licence) · get_code("146419f540edce62")
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

torch.manual_seed(0)

# The function takes a 'network' parameter which is indexed and iterated over
# Looking at the body: network[block_cnt] is accessed, checking equality with strings
# So network should be a list of strings representing layer types

network = [
    '3x3-3', 'skip', '5x5-6', '3x3-6',  # stage 0
    '5x5-3', 'skip', '3x3-3', '5x5-6',  # stage 1
    '3x3-6', 'skip', '5x5-3', '3x3-3',  # stage 2
    '5x5-6', 'skip', '3x3-3', '5x5-3',  # stage 3
    '3x3-6', 'skip', '5x5-6', '3x3-3',  # stage 4
]

ARGS = {
    'network': network
}
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
encode_single_path_nas_arch Ran
ran on a synthesized input
dstamoulis/single-path-nas/nas-search/plot-progress/parse_search_output.py
code served (permissive licence) · get_code("0ff5f3be7fcd4b71")
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 numpy as np

torch.manual_seed(0)

# The function expects `inds` to be a list/array with 20 rows,
# where each row is a list of 3 float values (either 0.0 or 1.0)
# The function iterates through 20 layers and checks each inds_row

inds = [
    [0.0, 1.0, 0.0],  # layer 0 -> idx 0
    [0.0, 1.0, 1.0],  # layer 1 -> idx 1
    [1.0, 1.0, 0.0],  # layer 2 -> idx 2
    [1.0, 1.0, 1.0],  # layer 3 -> idx 3
    [0.0, 0.0, 0.0],  # layer 4 -> idx 4
    [0.0, 0.0, 1.0],  # layer 5 -> idx 4
    [1.0, 0.0, 0.0],  # layer 6 -> idx 4
    [1.0, 0.0, 1.0],  # layer 7 -> idx 4
    [0.0, 1.0, 0.0],  # layer 8 -> idx 0
    [0.0, 1.0, 1.0],  # layer 9 -> idx 1
    [1.0, 1.0, 0.0],  # layer 10 -> idx 2
    [1.0, 1.0, 1.0],  # layer 11 -> idx 3
    [0.0, 0.0, 0.0],  # layer 12 -> idx 4
    [0.0, 0.0, 1.0],  # layer 13 -> idx 4
    [1.0, 0.0, 0.0],  # layer 14 -> idx 4
    [1.0, 0.0, 1.0],  # layer 15 -> idx 4
    [0.0, 1.0, 0.0],  # layer 16 -> idx 0
    [0.0, 1.0, 1.0],  # layer 17 -> idx 1
    [1.0, 1.0, 0.0],  # layer 18 -> idx 2
    [1.0, 1.0, 1.0],  # layer 19 -> idx 3
]

ARGS = {
    'inds': inds,
    'hard': False,
}
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 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
round_filters Ran
ran, honoured the contract we drafted
dstamoulis/single-path-nas/nas-search/singlepath_supernet.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from nsarang/MnasNet, is pointer only (licence: NONE) · get_code("285e4d1d9a3faeba")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from nsarang/MnasNet, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

round_filters Ran
ran on a synthesized input
dstamoulis/single-path-nas/runtime-modeling/model_def.py
code served (permissive licence) · get_code("7d1529c236ba14b5")
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 numpy as np


class GlobalParams:
    """Stand-in class for global parameters."""
    def __init__(self, depth_multiplier, depth_divisor, min_depth):
        self.depth_multiplier = depth_multiplier
        self.depth_divisor = depth_divisor
        self.min_depth = min_depth


# Create a valid GlobalParams instance
global_params = GlobalParams(
    depth_multiplier=1.0,
    depth_divisor=8,
    min_depth=16
)

# Test with a filters value
filters = 32

ARGS = {
    "filters": filters,
    "global_params": global_params
}
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 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
Indicator Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/superkernel.py
code served (permissive licence) · get_code("43e89725b46a75d5")
How it ran

no invocation recorded.

build_dropout_rate Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/nas_utils.py
code served (permissive licence) · get_code("3ea91536d334ba52")
How it ran

no invocation recorded.

build_learning_rate Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/nas_utils.py
code served (permissive licence) · get_code("713b0060c42cdcd3")
How it ran

no invocation recorded.

build_runtime_lambda Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/nas_utils.py
code served (permissive licence) · get_code("bbafe1b852b9e3f6")
How it ran

no invocation recorded.

conv_kernel_initializer Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/singlepath_supernet.py
code served (permissive licence) · get_code("17a1e9aed5cc4cee")
How it ran

no invocation recorded.

dense_kernel_initializer Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/singlepath_supernet.py
code served (permissive licence) · get_code("e430f1fd246b1b51")
How it ran

no invocation recorded.

distorted_bounding_box_crop Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/preprocessing.py
code served (permissive licence) · get_code("f45ff6307a313357")
How it ran

no invocation recorded.

export Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/search_main.py
code served (permissive licence) · get_code("e75aa2c199a09405")
How it ran

no invocation recorded.

parse_indicators_single_path_nas Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/plot-progress/parse_search_output.py
code served (permissive licence) · get_code("0ee0acc6ddbf5670")
How it ran

no invocation recorded.

preprocess_for_eval Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/preprocessing.py
code served (permissive licence) · get_code("3a458358e2a9447b")
How it ran

no invocation recorded.

preprocess_for_train Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/preprocessing.py
code served (permissive licence) · get_code("bc13cc6b340f0924")
How it ran

no invocation recorded.

sample_gumbel Not yet run
unverified (harvested, no recorded run)
dstamoulis/single-path-nas/nas-search/superkernel.py
code served (permissive licence) · get_code("fdd42e9c322647b8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 5 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 5 later) and 11 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
conv_kernel_initializer 4 other papers
4 later
1905.11946 (2019-05)
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
1911.04252 (2019-11)
Self-training with Noisy Student improves ImageNet classification
2006.14536 (2020-06)
Smooth Adversarial Training
and 1 more not named here
dense_kernel_initializer 4 other papers
4 later
1905.11946 (2019-05)
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
1911.04252 (2019-11)
Self-training with Noisy Student improves ImageNet classification
2006.14536 (2020-06)
Smooth Adversarial Training
and 1 more not named here
round_filters 3 other papers
1 dated earlier, 2 later
1807.11626 (2018-07)
MnasNet: Platform-Aware Neural Architecture Search for Mobile
1907.09595 (2019-07)
MixConv: Mixed Depthwise Convolutional Kernels
2004.01961 (2020-04)
Neural Architecture Search for Lightweight Non-Local Networks
distorted_bounding_box_crop 2 other papers
2 later
2006.14536 (2020-06)
Smooth Adversarial Training
2008.03703 (2020-08)
What Neural Networks Memorize and Why: Discovering the Long Tail via Influence Estimation
sample_gumbel 2 other papers
1 dated earlier, 1 later
1711.02085 (2017-11)
Neural Speed Reading via Skim-RNN
2107.04689 (2021-07)
Lifelong Teacher-Student Network 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

Can we automatically design a Convolutional Network (ConvNet) with the highest image classification accuracy under the runtime constraint of a mobile device? Neural architecture search (NAS) has revolutionized the design of hardware-efficient ConvNets by automating this process. However, the NAS problem remains challenging due to the combinatorially large design space, causing a significant searching time (at least 200 GPU-hours). To alleviate this complexity, we propose Single-Path NAS, a novel differentiable NAS method for designing hardware-efficient ConvNets in less than 4 hours. Our contributions are as follows: 1. Single-path search space: Compared to previous differentiable NAS methods, Single-Path NAS uses one single-path over-parameterized ConvNet to encode all architectural decisions with shared convolutional kernel parameters, hence drastically decreasing the number of trainable parameters and the search cost down to few epochs. 2. Hardware-efficient ImageNet classification: Single-Path NAS achieves 74.96% top-1 accuracy on ImageNet with 79ms latency on a Pixel 1 phone, which is state-of-the-art accuracy compared to NAS methods with similar constraints (<80ms). 3. NAS efficiency: Single-Path NAS search cost is only 8 epochs (30 TPU-hours), which is up to 5,000x faster compared to prior work. 4. Reproducibility: Unlike all recent mobile-efficient NAS methods which only release pretrained models, we open-source our entire codebase at: https://github.com/dstamoulis/single-path-nas.

For agents

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

get_harvested_code_for_paper("1904.02877")
get_code_for_paper("1904.02877")
have("1904.02877")

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

curl https://syntology.ai/api/ran/1904.02877.json

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

[![Syntology run record](https://syntology.ai/api/ran/1904.02877.svg)](https://syntology.ai/paper/1904.02877)

Connect an agent — have() is free.