SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1807.03039 · 2018

Glow: Generative Flow with Invertible 1x1 Convolutions

Diederik Kingma, Prafulla Dhariwal, San Openai, Francisco

arXiv · PDF · Open in the Atlas

Code that ran

85 of 129 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (47 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 70 have no contract verdict; 12 failures of our instrument; 44 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/1807.03039.json · badge: /api/ran/1807.03039.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
openai/glow canonical (classifier label)
classifier basis: census
8 ran · 3 no recorded run
y0ast/Glow-PyTorch community
Papers with Code archive
18 ran · 3 no recorded run
keonlee9420/VAENAR-TTS community
Papers with Code archive
8 ran · 7 no recorded run
Naagar/Glow_NormalizingFlow_implimentation community
Papers with Code archive
8 ran · 3 no recorded run
rosinality/glow-pytorch community
Papers with Code archive
7 ran · 3 no recorded run
chrischute/glow community
Papers with Code archive
6 ran · 3 no recorded run
lifeitech/fce-2d community
Papers with Code archive
6 ran · 1 no recorded run
rhychen/Glow community
Papers with Code archive
5 ran · 1 no recorded run
ClaraBing/flow community
Papers with Code archive
4 ran · 4 no recorded run
KiUngSong/Generative-Models community
Papers with Code archive
4 ran · 4 no recorded run
vvvm23/glow community
Papers with Code archive
4 ran · 3 no recorded run
mahkons/flows community
Papers with Code archive
3 ran · 0 no recorded run
eyalbetzalel/GLOW community (classifier label)
classifier basis not recorded on the link
2 ran · 0 no recorded run
5yearsKim/Conditional-Normalizing-Flow community
Papers with Code archive
1 ran · 4 no recorded run
Zhangyanbo/iResNetLab community
Papers with Code archive
1 ran · 0 no recorded run
Discover304/SinGlow community (classifier label)
classifier basis not recorded on the link
0 ran · 3 no recorded run
samuelmat19/GLOW-tf2 community (classifier label)
classifier basis not recorded on the link
0 ran · 2 no recorded run
FunctionStatusWhere it lives
ActNorm Ran
ran on a synthesized input · constructed an object; no result was computed
rosinality/glow-pytorch/model.py
code served (permissive licence) · get_code("595f289e212c9b26")
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, ActNorm: 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
from torch import nn

torch.manual_seed(0)

ARGS = {
    "in_channel": 4,
    "logdet": 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 ActNorm — 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 ActNorm 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
ActNorm Ran
ran on a synthesized input · constructed an object; no result was computed
chrischute/glow/models/glow/glow.py
code served (permissive licence) · get_code("65fc1df1616202de")
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, ActNorm: 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

torch.manual_seed(0)

ARGS = {
    'num_features': 4,
    'scale': 1.0,
    'return_ldj': 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 ActNorm — 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 ActNorm 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
ActNorm Ran
ran on a synthesized input · constructed an object; no result was computed
vvvm23/glow/glow/glow.py
code served (permissive licence) · get_code("0312a33dd14621de")
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, ActNorm: 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

torch.manual_seed(0)

ARGS = {
    'nb_channels': 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 ActNorm — 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 ActNorm 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
ActNorm Ran
ran on a synthesized input · constructed an object; no result was computed
rhychen/Glow/models/glow.py
pointer only (licence: NONE) · get_code("dfdfca3351e28290")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, rhychen/Glow 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.

ActNorm Ran
ran on a synthesized input · constructed an object; no result was computed
ClaraBing/flow/models/glow/glow.py
code served (permissive licence) · get_code("3bff339fa799f507")
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, ActNorm: 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

torch.manual_seed(0)

# Setup for ActNorm with layer_type='fc'
ARGS = {
    'num_features': 4,
    'layer_type': 'fc',
    'scale': 1.0,
    'return_ldj': 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 ActNorm — 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 ActNorm 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
ActNorm Ran
ran on a synthesized input
KiUngSong/Generative-Models/Glow/Glow_pytorch.py
code served (permissive licence) · get_code("28fdcb25a21f1a5e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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
from torch import nn

class ActNorm(nn.Module):

    def __init__(self, in_channel):
        super().__init__()
        self.scale = nn.Parameter(torch.ones(1, in_channel, 1, 1))
        self.bias = nn.Parameter(torch.zeros(1, in_channel, 1, 1))
        self.register_buffer('init', torch.tensor(0, dtype=torch.uint8))

    def initialize(self, x):
        with torch.no_grad():
            mean = torch.mean(x.clone(), dim=[0, 2, 3], keepdim=True)
            std = torch.std(x.clone(), dim=[0, 2, 3], keepdim=True)
            self.scale.data.copy_(1 / (std + 1e-06))
            self.bias.data.copy_(-mean)

    def forward(self, x):
        if self.init.item() == 0:
            self.initialize(x)
            self.init.fill_(1)
        _, _, h, w = x.shape
        logdet = h * w * torch.sum(torch.log(torch.abs(self.scale)))
        return (self.scale * (x + self.bias), logdet)

    def reverse(self, x):
        return x / self.scale - self.bias


INIT_ARGS = {
    'in_channel': 4
}

ARGS = {
    'x': torch.randn(2, 4, 3, 5)
}

if __name__ == '__main__':
    model = ActNorm(**INIT_ARGS)
    output, logdet = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Logdet shape: {logdet.shape}")
    print(f"Logdet value: {logdet}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch, torchvision — 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ActNorm2d Ran
ran on a synthesized input · constructed an object; no result was computed
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("bf223397ab401254")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

ActNormFlow Ran
ran on a synthesized input · constructed an object; no result was computed
keonlee9420/VAENAR-TTS/model/glow.py
code served (permissive licence) · get_code("55790851e3ad4bfa")
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, ActNormFlow: 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

torch.manual_seed(0)

# ActNormFlow.__init__ takes one parameter: channels
# It creates parameters of shape [channels]
# The forward/inverse/init methods expect inputs of shape [batch, channels]

channels = 4
batch_size = 2

# Create a small input tensor to verify compatibility
# Shape: [batch_size, channels]
sample_input = torch.randn(batch_size, channels)

ARGS = {
    'channels': channels
}
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 ActNormFlow — 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 ActNormFlow 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
Actnorm Ran
ran on a synthesized input
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("3495f030ce4be2cb")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

Actnorm Ran
ran on a synthesized input · constructed an object; no result was computed
lifeitech/fce-2d/flows/glow.py
code served (permissive licence) · get_code("4a89b88cec7ece97")
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, Actnorm: 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

torch.manual_seed(0)

class Actnorm(nn.Module):

    def __init__(self, param_dim=(1, 2)):
        super().__init__()
        self.scale = nn.Parameter(torch.ones(param_dim))
        self.bias = nn.Parameter(torch.zeros(param_dim))
        self.register_buffer('initialized', torch.tensor(0).byte())

    def forward(self, x):
        if not self.initialized:
            self.bias.squeeze().data.copy_(x.transpose(0, 1).flatten(1).mean(1)).view_as(self.scale)
            self.scale.squeeze().data.copy_(x.transpose(0, 1).flatten(1).std(1, unbiased=False) + 1e-06).view_as(self.bias)
            self.initialized += 1
        z = (x - self.bias) / self.scale
        logdet = -self.scale.abs().log().sum()
        return (z, logdet)

    def inverse(self, z):
        x = z * self.scale + self.bias
        logdet = self.scale.abs().log().sum()
        return (x, logdet)


# Create an instance to test with
actnorm_instance = Actnorm(param_dim=(1, 2))

# Create a small test input tensor: batch_size=2, feature_dim=4
# Shape should be compatible with the forward pass
x_input = torch.randn(2, 4)

ARGS = {
    'param_dim': (1, 2)
}
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 Actnorm — 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 Actnorm 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
AdditiveCoupling Ran
ran on a synthesized input · constructed an object; no result was computed
vvvm23/glow/glow/glow.py
code served (permissive licence) · get_code("ef54ca33abc06335")
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, AdditiveCoupling: 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

torch.manual_seed(0)

ARGS = {
    'nb_channels': 8,
    'hidden_channels': 16,
    'grad_checkpoint': 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 AdditiveCoupling — 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 AdditiveCoupling 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
AffineCoupling Ran
ran on a synthesized input · constructed an object; no result was computed
rosinality/glow-pytorch/model.py
code served (permissive licence) · get_code("6d49cd29d96b9e28")
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, AffineCoupling: 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
from torch import nn

torch.manual_seed(0)

ARGS = {
    'in_channel': 8,
    'filter_size': 64,
    'affine': 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 AffineCoupling — 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 AffineCoupling 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
AffineCoupling Ran
ran on a synthesized input · constructed an object; no result was computed
vvvm23/glow/glow/glow.py
code served (permissive licence) · get_code("ccfc200e43a99a10")
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, AffineCoupling: 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

torch.manual_seed(0)

ARGS = {
    "nb_channels": 4,
    "hidden_channels": 8,
    "grad_checkpoint": 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 AffineCoupling — 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 AffineCoupling 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
AffineCoupling Ran
ran on a synthesized input · constructed an object; no result was computed
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("373806f87fa29229")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

AffineCoupling Ran
ran on a synthesized input · constructed an object; no result was computed
lifeitech/fce-2d/flows/glow.py
code served (permissive licence) · get_code("59684e4be6834453")
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, AffineCoupling: 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

torch.manual_seed(0)

ARGS = {
    "dim": 4,
    "width": 8,
}
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 AffineCoupling — 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 AffineCoupling 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
AffineCoupling Ran
ran on a synthesized input
KiUngSong/Generative-Models/Glow/Glow_pytorch.py
code served (permissive licence) · get_code("22cf5be44ad0609c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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
from torch import nn
from torch.nn import functional as F

class ZeroConv(nn.Module):

    def __init__(self, in_channel, out_channel, scale_factor=3):
        super().__init__()
        self.conv = nn.Conv2d(in_channel, out_channel, 3, 1, 0)
        nn.init.zeros_(self.conv.weight)
        nn.init.zeros_(self.conv.bias)
        self.scale = nn.Parameter(torch.zeros(1, out_channel, 1, 1))
        self.scale_factor = scale_factor

    def forward(self, x):
        x = F.pad(x, [1, 1, 1, 1], value=1)
        return self.conv(x) * torch.exp(self.scale * self.scale_factor)

class AffineCoupling(nn.Module):

    def __init__(self, in_channel, latent_dim=512):
        super().__init__()
        self.net = nn.Sequential(nn.Conv2d(in_channel // 2, latent_dim, 3, padding=1), nn.ReLU(inplace=True), nn.Conv2d(latent_dim, latent_dim, 1), nn.ReLU(inplace=True), ZeroConv(latent_dim, in_channel))

    def forward(self, x):
        x_a, x_b = x.chunk(2, 1)
        log_s, bias = self.net(x_b).chunk(2, 1)
        scale = torch.sigmoid(log_s + 2)
        logdet = torch.log(scale).view(x.shape[0], -1).sum(-1)
        return (torch.cat([scale * (x_a + bias), x_b], 1), logdet)

    def reverse(self, output):
        y_a, y_b = output.chunk(2, 1)
        log_s, bias = self.net(y_b).chunk(2, 1)
        scale = torch.sigmoid(log_s + 2)
        return torch.cat([y_a / scale - bias, y_b], 1)

INIT_ARGS = {
    'in_channel': 8,
    'latent_dim': 64
}

ARGS = {
    'x': torch.randn(2, 8, 4, 3)
}

if __name__ == '__main__':
    model = AffineCoupling(**INIT_ARGS)
    output, logdet = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Logdet shape: {logdet.shape}")
    print(f"Logdet: {logdet}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch, torchvision — 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
BaseAttention Ran
ran on a synthesized input · constructed an object; no result was computed
keonlee9420/VAENAR-TTS/model/glow.py
code served (permissive licence) · get_code("ceeeb1dd1e32ffaa")
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, BaseAttention: the run constructed the object and did not call it, so no result was computed. 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 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 SimpleAttention(nn.Module):
    def __init__(self, attention_dim):
        super(SimpleAttention, self).__init__()
        self.attention_dim = attention_dim
    
    def forward(self, inputs, memory, memory_lengths, query_lengths):
        batch_size = inputs.shape[0]
        q_time = inputs.shape[1]
        m_time = memory.shape[1]
        
        contexts = torch.zeros(batch_size, q_time, self.attention_dim)
        alignments = torch.softmax(torch.randn(batch_size, q_time, m_time), dim=-1)
        
        return contexts, alignments

attention_dim = 8
batch_size = 2
q_time = 5
m_time = 7
q_dim = 4
m_dim = 6

inputs = torch.randn(batch_size, q_time, q_dim)
memory = torch.randn(batch_size, m_time, m_dim)
memory_lengths = torch.tensor([7, 5], dtype=torch.int32)
query_lengths = torch.tensor([5, 3], dtype=torch.int32)

ARGS = {
    "attention_dim": attention_dim
}
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 BaseAttention — 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 BaseAttention 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
BaseTransform Ran
ran on a synthesized input · constructed an object; no result was computed
keonlee9420/VAENAR-TTS/model/glow.py
code served (permissive licence) · get_code("68d4c2f44e93569d")
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, BaseTransform: 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

torch.manual_seed(0)

ARGS = {
    'in_dim': 4,
    'out_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 BaseTransform — 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 BaseTransform 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
Conv2d Ran
ran on a synthesized input · constructed an object; no result was computed
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("d15a5ce5143a7f33")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

Conv2dZeros Ran
ran on a synthesized input · constructed an object; no result was computed
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("eddb8dfbe9cd3b72")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

Coupling Ran
ran on a synthesized input · constructed an object; no result was computed
chrischute/glow/models/glow/glow.py
code served (permissive licence) · get_code("3b10b0afd12bb72c")
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, Coupling: 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

torch.manual_seed(0)

# The Coupling class is incomplete in the provided code, but based on the context
# it appears to take in_channels and mid_channels as parameters.
# We need to create valid arguments for it.

ARGS = {
    'in_channels': 4,
    'mid_channels': 8,
}
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 Coupling — 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 Coupling 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
CouplingLayer Ran
ran on a synthesized input · constructed an object; no result was computed
mahkons/flows/models/CouplingLayer.py
code served (permissive licence) · get_code("7e3c8937d2cbfe61")
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, CouplingLayer: 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

torch.manual_seed(0)

# image_shape: tuple of (channels, height, width)
image_shape = (3, 4, 4)

# hidden_channels: int for the number of hidden channels in ResnetBlocks
hidden_channels = 8

# num_resnet: int for the number of ResnetBlock layers
num_resnet = 2

# mask: tensor of shape matching image_shape with binary values
# Create a checkerboard-like mask for alternating channel coupling
mask = torch.zeros(image_shape, dtype=torch.float32)
mask[0, :, :] = 1.0  # First channel is masked (kept)

ARGS = {
    'image_shape': image_shape,
    'hidden_channels': hidden_channels,
    'num_resnet': num_resnet,
    'mask': mask,
}
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 CouplingLayer — 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 CouplingLayer 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
FFN Ran
ran on a synthesized input
keonlee9420/VAENAR-TTS/model/glow.py
code served (permissive licence) · get_code("b5a032a39e673350")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, 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

# Define initialization arguments for FFN.__init__
INIT_ARGS = {
    'in_features': 4,
    'hidden1': 8,
    'hidden2': 4,
}

# Define arguments for FFN.forward
ARGS = {
    'inputs': torch.randn(2, 4),
    'mask': None,
}

# Construct and call the class
if __name__ == '__main__':
    ffn = nn.Module()
    
    # Manually construct FFN to verify it works
    class LinearNorm(nn.Module):
        def __init__(self, in_features, out_features, activation=None, use_bias=True, kernel_initializer='glorot_uniform', bias_initializer='zeros'):
            super(LinearNorm, self).__init__()
            self.linear = nn.Linear(in_features=in_features, out_features=out_features, bias=use_bias)
            if kernel_initializer == 'glorot_uniform':
                nn.init.xavier_uniform_(self.linear.weight)
            elif kernel_initializer == 'zeros':
                nn.init.zeros_(self.linear.weight)
            if use_bias:
                if bias_initializer == 'zeros':
                    nn.init.constant_(self.linear.bias, 0.0)
                else:
                    raise NotImplementedError
            self.activation = activation if activation is not None else nn.Identity()

        def forward(self, x):
            x = self.activation(self.linear(x))
            return x

    class FFN(nn.Module):
        def __init__(self, in_features, hidden1, hidden2):
            super(FFN, self).__init__()
            self.dense1 = LinearNorm(in_features, hidden1, activation=nn.ReLU())
            self.dense2 = LinearNorm(hidden1, hidden2, activation=None)
            self.layer_norm = nn.LayerNorm(hidden2)

        def forward(self, inputs, mask=None):
            dense1_outs = self.dense1(inputs)
            dense2_outs = self.dense2(dense1_outs)
            outs = dense2_outs + inputs
            outs = self.layer_norm(outs)
            if mask is not None:
                outs = outs.masked_fill(mask.unsqueeze(-1), 0.0)
            return outs

    model = FFN(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Flow Ran
ran on a synthesized input · constructed an object; no result was computed
mahkons/flows/models/CouplingLayer.py
code served (permissive licence) · get_code("21fa24b8b336cbae")
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, Flow: 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

torch.manual_seed(0)

class ConcreteFlow(nn.Module):
    def __init__(self):
        super(ConcreteFlow, self).__init__()
        self.device = torch.device('cpu')
        self.linear = nn.Linear(4, 4)
    
    def to(self, device):
        super(ConcreteFlow, self).to(device)
        self.device = device
        return self
    
    def forward_flow(self, x):
        """returns forward flow computation result and log of jacobian"""
        z = self.linear(x)
        log_jac = torch.zeros(x.shape[0], device=self.device)
        return z, log_jac
    
    def inverse_flow(self, x):
        """returns inverse flow computation result and log of jacobian"""
        z = self.linear(x)
        log_jac = torch.zeros(x.shape[0], device=self.device)
        return z, log_jac

flow_instance = ConcreteFlow()

ARGS = {}
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 Flow — 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 Flow 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
FlowLevel Ran
ran on a synthesized input · constructed an object; no result was computed
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("f12c4e2f139f27fb")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

FlowLevel Ran
ran on a synthesized input · constructed an object; no result was computed
lifeitech/fce-2d/flows/glow.py
code served (permissive licence) · get_code("00947c93efacb3dd")
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, FlowLevel: 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

torch.manual_seed(0)

ARGS = {
    'dim': 4,
    'width': 32,
    'depth': 2
}
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 FlowLevel — 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 FlowLevel 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
FlowStep Ran
ran on a synthesized input · constructed an object; no result was computed
lifeitech/fce-2d/flows/glow.py
code served (permissive licence) · get_code("5e726e30234d700d")
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, FlowStep: 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

torch.manual_seed(0)

ARGS = {
    'dim': 4,
    'width': 128
}
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 FlowStep — 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 FlowStep output the recorded ndim 1 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
FlowStep Ran
ran on a synthesized input
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("6e65ead7e62d95d2")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

GaussianPrior Ran
ran on a synthesized input · constructed an object; no result was computed
rhychen/Glow/models/glow.py
pointer only (licence: NONE) · get_code("29d8d0bee755fdac")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, rhychen/Glow 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.

Gaussianize Ran
ran on a synthesized input · constructed an object; no result was computed
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("a3e5d01daa126e24")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

Gaussianize Ran
ran on a synthesized input · constructed an object; no result was computed
rhychen/Glow/models/glow.py
pointer only (licence: NONE) · get_code("329f607dc4f85586")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, rhychen/Glow 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.

Glow Ran
ran on a synthesized input · constructed an object; no result was computed
lifeitech/fce-2d/flows/glow.py
code served (permissive licence) · get_code("0e8254537237ccf3")
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, Glow: 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 torch.nn.functional as F
import torch.distributions as D

torch.manual_seed(0)

ARGS = {
    "width": 64,
    "depth": 2,
    "n_levels": 1,
    "data_dim": 2,
}
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 Glow — 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 Glow 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
Inv1x1 Ran
ran on a synthesized input
KiUngSong/Generative-Models/Glow/Glow_pytorch.py
code served (permissive licence) · get_code("658c3b5f01c3cd7e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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
from torch import nn
from einops import rearrange
from torch.nn import functional as F

class Inv1x1(nn.Module):

    def __init__(self, in_channel):
        super().__init__()
        w_shape = [in_channel, in_channel]
        w_init, _ = torch.qr(torch.randn(*w_shape))
        P, L, U = torch.lu_unpack(*torch.lu(w_init))
        s = torch.diag(U)
        self.register_buffer('L_mask', torch.tril(torch.ones(w_shape), -1))
        self.register_buffer('eye', torch.eye(*w_shape))
        self.register_buffer('P', P)
        self.register_buffer('s_sign', torch.sign(s))
        self.log_s = nn.Parameter(torch.log(torch.abs(s)))
        self.L = nn.Parameter(L)
        self.U = nn.Parameter(U)

    def get_weight(self):
        weight = self.P @ (self.L * self.L_mask + self.eye) @ (self.U * self.L_mask.t() + torch.diag(self.s_sign * torch.exp(self.log_s)))
        return weight

    def forward(self, x):
        _, _, h, w = x.shape
        weight = self.get_weight()
        weight = rearrange(weight, 'w1 w2 -> w1 w2 1 1')
        logdet = h * w * torch.sum(self.log_s)
        return (F.conv2d(x, weight), logdet)

    def reverse(self, x):
        weight = self.get_weight()
        weight_inv = rearrange(weight.inverse(), 'w1 w2 -> w1 w2 1 1')
        return F.conv2d(x, weight_inv)


INIT_ARGS = {
    'in_channel': 4
}

ARGS = {
    'x': torch.randn(2, 4, 3, 5)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
einops, torch, torchvision — 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
InvConv Ran
ran on a synthesized input · constructed an object; no result was computed
chrischute/glow/models/glow/glow.py
code served (permissive licence) · get_code("c108e525040ebd71")
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, InvConv: 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 torch.nn.functional as F
import numpy as np

torch.manual_seed(0)

# Create arguments for InvConv
num_channels = 4

ARGS = {
    'num_channels': num_channels,
}
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 InvConv — 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 InvConv 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
InvConv Ran
ran on a synthesized input · constructed an object; no result was computed
vvvm23/glow/glow/glow.py
code served (permissive licence) · get_code("d21e5cea397ce344")
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, InvConv: 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

torch.manual_seed(0)

ARGS = {
    'nb_channels': 4,
    'lu': 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 InvConv — 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 InvConv 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
InvConv Ran
ran on a synthesized input · constructed an object; no result was computed
ClaraBing/flow/models/glow/glow.py
code served (permissive licence) · get_code("3fee1f342bcbbca1")
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, InvConv: 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)
np.random.seed(0)

# InvConv takes num_channels as its only parameter
num_channels = 4

ARGS = {
    "num_channels": num_channels,
}
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 InvConv — 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 InvConv 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
InvConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
rosinality/glow-pytorch/model.py
code served (permissive licence) · get_code("767354f41d756da7")
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, InvConv2d: 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
from torch import nn

torch.manual_seed(0)

# The InvConv2d constructor takes in_channel as a parameter
# The forward method expects input of shape (batch, in_channel, height, width)
# The reverse method also works with the same weight structure

in_channel = 4
batch_size = 2
height = 3
width = 5

# Create a simple input tensor with shape (batch, in_channel, height, width)
input_tensor = torch.randn(batch_size, in_channel, height, width)

ARGS = {
    'in_channel': in_channel
}
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 InvConv2d — 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 InvConv2d 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
InvConv2dLU Ran
ran on a synthesized input · constructed an object; no result was computed
rosinality/glow-pytorch/model.py
code served (permissive licence) · get_code("ba0d2a54b554b41a")
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, InvConv2dLU: 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
from torch import nn
import numpy as np

torch.manual_seed(0)

# The function InvConv2dLU.__init__ takes only one parameter: in_channel
# We need to provide a small integer value for the number of input channels

ARGS = {
    'in_channel': 4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type InvConv2dLU — 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 InvConv2dLU 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
Invertible1x1Conv Ran
ran on a synthesized input
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("990ea29970665475")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

Invertible1x1Conv Ran
ran on a synthesized input · constructed an object; no result was computed
lifeitech/fce-2d/flows/glow.py
code served (permissive licence) · get_code("df166ea8c0d1968a")
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, Invertible1x1Conv: 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

torch.manual_seed(0)

class Invertible1x1Conv(nn.Module):

    def __init__(self, dim=2):
        super().__init__()
        w = torch.randn(dim, dim)
        w = torch.linalg.qr(w)[0]
        self.w = nn.Parameter(w)

    def forward(self, x):
        logdet = torch.slogdet(self.w)[-1]
        return (x @ self.w.t(), logdet)

    def inverse(self, z):
        w_inv = self.w.t().inverse()
        logdet = -torch.slogdet(self.w)[-1]
        return (z @ w_inv, logdet)


ARGS = {
    "dim": 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 Invertible1x1Conv — 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 Invertible1x1Conv 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
InvertibleConv Ran
ran on a synthesized input
rhychen/Glow/models/glow.py
pointer only (licence: NONE) · get_code("eb9cb175c704641f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, rhychen/Glow 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.

InvertibleConv1x1 Ran
ran on a synthesized input
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("a8a027187c32cb9a")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

InvertibleLinearFlow Ran
ran on a synthesized input · constructed an object; no result was computed
keonlee9420/VAENAR-TTS/model/glow.py
code served (permissive licence) · get_code("21e86fbd65ad8813")
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, InvertibleLinearFlow: 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)

# InvertibleLinearFlow takes one parameter: channels
# Looking at the forward/inverse methods, it expects:
# - inputs: torch.Tensor of shape [batch, channels]
# - inputs_lengths: optional, torch.Tensor of shape [batch]

# Create a small instance
channels = 4
flow = nn.Linear(channels, channels)  # Stand-in, not actually used in ARGS

# The ARGS dict should contain arguments to construct InvertibleLinearFlow
ARGS = {
    'channels': 4
}
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 InvertibleLinearFlow — 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 InvertibleLinearFlow 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
LinearNorm Ran
ran on a synthesized input · constructed an object; no result was computed
keonlee9420/VAENAR-TTS/model/glow.py
code served (permissive licence) · get_code("dc67665961a19370")
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, LinearNorm: 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

torch.manual_seed(0)

ARGS = {
    'in_features': 4,
    'out_features': 3,
    'activation': nn.ReLU(),
    'use_bias': True,
    'kernel_initializer': 'glorot_uniform',
    'bias_initializer': 'zeros'
}
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 LinearNorm — 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 LinearNorm 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
LinearZeros Ran
ran on a synthesized input · constructed an object; no result was computed
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("d5c82ba7ba37b3bf")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

NN Ran
ran on a synthesized input · constructed an object; no result was computed
chrischute/glow/models/glow/glow.py
code served (permissive licence) · get_code("fffbbad6131561fe")
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, NN: 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

torch.manual_seed(0)

ARGS = {
    "in_channels": 3,
    "mid_channels": 16,
    "out_channels": 6,
    "use_act_norm": 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 NN — 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 NN 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
NN Ran
ran on a synthesized input · constructed an object; no result was computed
ClaraBing/flow/models/glow/glow.py
code served (permissive licence) · get_code("bf470a3be388ef5e")
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, NN: 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

torch.manual_seed(0)

ARGS = {
    'in_dim': 4,
    'mid_dim': 8,
    'out_dim': 3,
    'use_act_norm': False,
    'layer_type': 'conv'
}
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 NN — 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 NN 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
NN Ran
ran on a synthesized input · constructed an object; no result was computed
5yearsKim/Conditional-Normalizing-Flow/models/glow/glow.py
pointer only (licence: NONE) · get_code("61eb0c88853c62f1")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, 5yearsKim/Conditional-Normalizing-Flow 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.

PLUMatrix Ran
ran on a synthesized input · constructed an object; no result was computed
Zhangyanbo/iResNetLab/INN/utilities.py
pointer only (licence: GPL-3.0) · get_code("268ef31122cf2607")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, Zhangyanbo/iResNetLab, is not in the permissive set Syntology serves inline. 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.

Permute2d Ran
ran on a synthesized input · constructed an object; no result was computed
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("43a599d6552d0ece")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

PositionalEncoding Ran
ran on a synthesized input
keonlee9420/VAENAR-TTS/model/glow.py
code served (permissive licence) · get_code("cbbd1ff4b9219b9a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, 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

device = torch.device('cpu')

INIT_ARGS = {}

METHOD = 'positional_encoding'

ARGS = {
    'len': 8,
    'dim': 6,
    'device': device,
    'step': 1.0
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Preprocess Ran
ran on a synthesized input · constructed an object; no result was computed
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("1f34cf836f6a0eaa")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

ResnetBlock Ran
ran on a synthesized input · constructed an object; no result was computed
mahkons/flows/models/CouplingLayer.py
code served (permissive licence) · get_code("f4519d2bed1c4804")
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, ResnetBlock: 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

torch.manual_seed(0)

ARGS = {
    'n_channels': 4,
    'with_batch_norm': True,
    'with_weight_norm': 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 ResnetBlock — 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 ResnetBlock 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
Split Ran
ran on a synthesized input
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("d4de2af5339bde99")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

Split2d Ran
ran on a synthesized input · constructed an object; no result was computed
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("db30220ddc114661")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

Squeeze Ran
ran on a synthesized input · constructed an object; no result was computed
Naagar/Glow_NormalizingFlow_implimentation/glow.py
pointer only (licence: NONE) · get_code("ff7465a2f72088f8")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Naagar/Glow_NormalizingFlow_implimentation 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.

Squeeze Ran
ran on a synthesized input
rhychen/Glow/models/glow.py
pointer only (licence: NONE) · get_code("b9d7092efd7b3931")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, rhychen/Glow 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.

SqueezeLayer Ran
ran on a synthesized input · constructed an object; no result was computed
y0ast/Glow-PyTorch/model.py
pointer only (licence: NOASSERTION) · get_code("ee6ee2d65cbbd221")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, y0ast/Glow-PyTorch (NOASSERTION). 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.

ZeroConv Ran
ran on a synthesized input
KiUngSong/Generative-Models/Glow/Glow_pytorch.py
code served (permissive licence) · get_code("0bb1ec9e2327d771")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, dtype family float, shape (2, 4, 5, 5) 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
from torch import nn

class ZeroConv(nn.Module):

    def __init__(self, in_channel, out_channel, scale_factor=3):
        super().__init__()
        self.conv = nn.Conv2d(in_channel, out_channel, 3, 1, 0)
        nn.init.zeros_(self.conv.weight)
        nn.init.zeros_(self.conv.bias)
        self.scale = nn.Parameter(torch.zeros(1, out_channel, 1, 1))
        self.scale_factor = scale_factor

    def forward(self, x):
        x = torch.nn.functional.pad(x, [1, 1, 1, 1], value=1)
        return self.conv(x) * torch.exp(self.scale * self.scale_factor)


INIT_ARGS = {
    'in_channel': 3,
    'out_channel': 4,
    'scale_factor': 3
}

ARGS = {
    'x': torch.randn(2, 3, 5, 5)
}

if __name__ == '__main__':
    model = ZeroConv(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 4, 5, 5) — shape (2, 4, 5, 5) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ZeroConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
rosinality/glow-pytorch/model.py
code served (permissive licence) · get_code("f301e9a7e81f3c69")
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, ZeroConv2d: 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
from torch import nn

torch.manual_seed(0)

# ZeroConv2d takes in_channel, out_channel, and padding as arguments
# in_channel: number of input channels
# out_channel: number of output channels
# padding: padding parameter (default=1)

# Looking at the forward method:
# - input is expected to be a tensor that gets padded and passed through conv
# - The conv is Conv2d(in_channel, out_channel, 3, padding=0)
# - A typical input would be (batch, in_channel, height, width)

in_channel = 3
out_channel = 5
padding = 1

ARGS = {
    "in_channel": in_channel,
    "out_channel": out_channel,
    "padding": padding
}
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 ZeroConv2d — 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 ZeroConv2d 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

Showing the first 60. An agent gets every one from get_harvested_code_for_paper("1807.03039").

Also present in other papers' code

Of 129 samples lifted from this paper's linked repositories, 129 carry a function identity and 0 do not. Of the 129 distinct functions with an identity, 15 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 12 later) and 114 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
gaussian_log_p
brick of this paper's assembly: helper
5 other papers
5 later · 1 only as an assembly brick
1912.01219 (2019-12)
WaveFlow: A Compact Flow-based Model for Raw Audio
2206.05075 (2022-06)
Diffeomorphic Counterfactuals with Generative Models
and 2 more not named here
gaussian_sample
brick of this paper's assembly: sampler
4 other papers
4 later
1912.01219 (2019-12)
WaveFlow: A Compact Flow-based Model for Raw Audio
2206.05075 (2022-06)
Diffeomorphic Counterfactuals with Generative Models
and 1 more not named here
Flow
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 2 only as an assembly brick
1605.08803 (2016-05)
Density estimation using Real NVP
1705.07057 (2017-05)
Masked Autoregressive Flow for Density Estimation
ZeroConv2d
brick of this paper's assembly: helper
2 other papers
2 later · 1 only as an assembly brick
2105.07239 (2021-05)
AgeFlow: Conditional Age Progression and Regression with Normalizing Flows
2305.08457 (2023-05)
MolHF: A Hierarchical Normalizing Flow for Molecular Graph Generation
calc_z_shapes 2 other papers
2 later
2206.05075 (2022-06)
Diffeomorphic Counterfactuals with Generative Models
2306.10161 (2023-06)
Building the Bridge of Schrödinger: A Continuous Entropic Optimal Transport Benchmark
compute_same_pad
brick of this paper's assembly: helper
2 other papers
2 later · 1 only as an assembly brick
2002.11896 (2020-02)
Gradient Boosted Normalizing Flows
136750141 (2022 (year only))
Deep Bayesian Video Frame Interpolation
ActNorm
brick of this paper's assembly: helper
1 other paper
1 later
2006.14154 (2020-06)
Strictly Batch Imitation Learning by Energy-based Distribution Matching
CouplingLayer
brick of this paper's assembly: entry
1 other paper
1 dated earlier · 1 only as an assembly brick
1605.08803 (2016-05)
Density estimation using Real NVP
ResnetBlock
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
1605.08803 (2016-05)
Density estimation using Real NVP
calc_loss 1 other paper
1 later
2306.10161 (2023-06)
Building the Bridge of Schrödinger: A Continuous Entropic Optimal Transport Benchmark
flatten_eps 1 other paper
1 later
2106.12182 (2021-06)
Fairness for Image Generation with Uncertain Sensitive Attributes
mean_dim
brick of this paper's assembly: helper
1 other paper
1 later
2106.03357 (2021-06)
Evaluating State-of-the-Art Classification Models Against Bayes Optimality
squeeze
brick of this paper's assembly: helper, model
1 other paper
1 later
2106.03357 (2021-06)
Evaluating State-of-the-Art Classification Models Against Bayes Optimality
squeeze2d
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2002.11896 (2020-02)
Gradient Boosted Normalizing Flows
unsqueeze2d
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2002.11896 (2020-02)
Gradient Boosted Normalizing Flows

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

Flow-based generative models (Dinh et al., 2014) are conceptually attractive due to tractability of the exact log-likelihood, tractability of exact latent-variable inference, and parallelizability of both training and synthesis. In this paper we propose Glow, a simple type of generative flow using an invertible 1 × 1 convolution. Using our method we demonstrate a significant improvement in log-likelihood on standard benchmarks. Perhaps most strikingly, we demonstrate that a generative model optimized towards the plain log-likelihood objective is capable of efficient realisticlooking synthesis and manipulation of large images. The code for our model is available at https://github.com/openai/glow.

For agents

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

get_harvested_code_for_paper("1807.03039")
get_code_for_paper("1807.03039")
have("1807.03039")

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

curl https://syntology.ai/api/ran/1807.03039.json

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

[![Syntology run record](https://syntology.ai/api/ran/1807.03039.svg)](https://syntology.ai/paper/1807.03039)

Connect an agent — have() is free.