SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1605.08803 · 2016

Density estimation using Real NVP

Laurent Dinh, Jascha Sohl, Dickstein Google, Brain Bengio

arXiv · PDF · Open in the Atlas

Code that ran

48 of 70 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (26 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 36 have no contract verdict; 10 failures of our instrument; 22 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/1605.08803.json · badge: /api/ran/1605.08803.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
mahkons/flows community
Papers with Code archive
9 ran · 2 no recorded run
xlwan/KRnet community
not in the paper's declared repositories
7 ran · 5 no recorded run
chrischute/real-nvp community (classifier label)
classifier basis not recorded on the link
5 ran · 0 no recorded run
fmu2/realNVP community
Papers with Code archive
4 ran · 7 no recorded run
hse-cs/probaforms community
Papers with Code archive
3 ran · 2 no recorded run
ispamm/realnvp-demo-pytorch community
Papers with Code archive
3 ran · 2 no recorded run
wjy5446/pytorch-Real-NVP community
not in the paper's declared repositories
3 ran · 0 no recorded run
e-hulten/real-nvp-2d community
Papers with Code archive
3 ran · 0 no recorded run
li012589/NeuralRG community
Papers with Code archive
3 ran · 0 no recorded run
simonwestberg/DD2412-Glow community
Papers with Code archive
2 ran · 0 no recorded run
rhychen/Glow community
Papers with Code archive
2 ran · 0 no recorded run
ssumin6/real_nvp community
Papers with Code archive
1 ran · 0 no recorded run
ikostrikov/pytorch-flows community
Papers with Code archive
1 ran · 0 no recorded run
CompPhysVienna/paper_NF_for_rare_events community
Papers with Code archive
1 ran · 0 no recorded run
copy not recorded community
not in the paper's declared repositories
1 ran · 0 no recorded run
sshish/NF community
Papers with Code archive
0 ran · 2 no recorded run
taesung89/real-nvp community
Papers with Code archive
0 ran · 2 no recorded run
FunctionStatusWhere it lives
AbstractCoupling Ran
ran on a synthesized input · constructed an object; no result was computed
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("f972e7e0b4588d1e")
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, AbstractCoupling: 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)

class HyperparameterConfig:
    def __init__(self):
        self.res_blocks = 2
        self.bottleneck = 16
        self.skip = True
        self.weight_norm = False
        self.coupling_bn = True

mask_config_value = 1.0
hps_value = HyperparameterConfig()

ARGS = {
    'mask_config': mask_config_value,
    'hps': hps_value
}
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 AbstractCoupling — 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 AbstractCoupling 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
ActNormImage Ran
ran on a synthesized input · constructed an object; no result was computed
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("d5af26f1bb10937d")
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, ActNormImage: 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)

# ActNormImage.__init__ takes one parameter: dim
# The class uses dim to create parameters of shape (dim, 1, 1)
# Looking at forward_flow and inverse_flow, they expect x to have shape (batch, dim, height, width)
# and they use x.shape[0] (batch), x.shape[2] (height), x.shape[3] (width)

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

ARGS = {
    'dim': 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 ActNormImage — 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 ActNormImage 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
simonwestberg/DD2412-Glow/model.py
pointer only (licence: NONE) · get_code("b774329417d3cb53")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, simonwestberg/DD2412-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.

AffineCoupling Ran
ran on a synthesized input · constructed an object; no result was computed
rhychen/Glow/models/glow.py
pointer only (licence: NONE) · get_code("fcdf1094082b5983")
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.

AffineCouplingLayer Ran
ran on a synthesized input
ssumin6/real_nvp/layers.py
pointer only (licence: NONE) · get_code("a68f3013f9d0faa8")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ssumin6/real_nvp 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.

ConditionalFlow Ran
ran on a synthesized input · constructed an object; no result was computed
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("b896483dfb367a1d")
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, ConditionalFlow: 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 SimpleConditionalFlow(nn.Module):
    """Minimal concrete implementation of ConditionalFlow for testing"""
    def __init__(self):
        super(SimpleConditionalFlow, self).__init__()
        self.device = torch.device('cpu')
        self.linear = nn.Linear(4, 4)
    
    def to(self, device):
        super(SimpleConditionalFlow, self).to(device)
        self.device = device
        return self
    
    def forward_flow(self, x, condition=None):
        """Returns transformed x and log jacobian determinant"""
        z = self.linear(x)
        log_jac = torch.zeros(x.shape[0], device=x.device)
        return z, log_jac
    
    def inverse_flow(self, x, condition=None):
        """Returns inverse transformed x and log jacobian determinant"""
        z = self.linear(x)
        log_jac = torch.zeros(x.shape[0], device=x.device)
        return z, log_jac
    
    def data_init(self, x, condition=None):
        return self.forward_flow(x, condition)[0]

# Create test data
x_test = torch.randn(2, 4)
condition_test = torch.randn(2, 3)

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 ConditionalFlow — 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 ConditionalFlow 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_zeros Ran
ran on a synthesized input · constructed an object; no result was computed
simonwestberg/DD2412-Glow/model.py
pointer only (licence: NONE) · get_code("2e844d999d1172c3")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, simonwestberg/DD2412-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.

Conv2dZeroInit Ran
ran on a synthesized input · constructed an object; no result was computed
rhychen/Glow/models/glow.py
pointer only (licence: NONE) · get_code("666bcaaa2d8effd9")
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.

CouplingLayer Ran
ran on a synthesized input · constructed an object; no result was computed
wjy5446/pytorch-Real-NVP/network.py
pointer only (licence: NONE) · get_code("3d011f7cef5a2e7b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, wjy5446/pytorch-Real-NVP 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.

CouplingLayer Ran
ran on a synthesized input · constructed an object; no result was computed
e-hulten/real-nvp-2d/rnvp.py
pointer only (licence: NONE) · get_code("0eaa0501f292c8c6")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, e-hulten/real-nvp-2d 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.

CouplingLayer Ran
ran on a synthesized input · constructed an object; no result was computed
ikostrikov/pytorch-flows/flows.py
code served (permissive licence) · get_code("34ffed75d4fb4931")
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)

# Create a mask for the coupling layer
# The mask alternates which inputs are transformed
num_inputs = 4
mask = torch.tensor([1.0, 0.0, 1.0, 0.0])

# Create conditional inputs (optional)
# If we want to test with conditional inputs, we need cond_inputs in forward
# But for the constructor, we just need num_cond_inputs parameter
num_cond_inputs = 3

# Hidden layer size
num_hidden = 8

# Activation function choices
s_act = 'tanh'
t_act = 'relu'

ARGS = {
    'num_inputs': num_inputs,
    'num_hidden': num_hidden,
    'mask': mask,
    'num_cond_inputs': num_cond_inputs,
    's_act': s_act,
    't_act': t_act
}
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
Flow Ran
ran on a synthesized input · constructed an object; no result was computed
li012589/NeuralRG/flow/rnvp.py
code served (permissive licence) · get_code("5f7fa08b89de722c")
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)

# Create a simple prior that has sample() and logProbability() methods
class SimplePrior:
    def sample(self, batchSize):
        return torch.randn(batchSize, 4)
    
    def logProbability(self, z):
        # Standard normal log probability
        return -0.5 * (z ** 2).sum(dim=1)

# Create a simple Flow subclass that implements inverse and forward
class SimpleFlow(nn.Module):
    def __init__(self, prior=None, name='Flow'):
        super(SimpleFlow, self).__init__()
        self.name = name
        self.prior = prior
        self.linear = nn.Linear(4, 4)
    
    def forward(self, x):
        # Returns (z, logp) where logp is the log determinant of Jacobian
        z = self.linear(x)
        logp = torch.zeros(x.shape[0])
        return z, logp
    
    def inverse(self, z):
        # Returns (x, logp_) where logp_ is the log determinant of Jacobian
        x = self.linear(z)
        logp_ = torch.zeros(z.shape[0])
        return x, logp_

prior = SimplePrior()

ARGS = {
    'prior': prior,
    'name': 'TestFlow'
}
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
Function_s_t Ran
ran on a synthesized input · constructed an object; no result was computed
wjy5446/pytorch-Real-NVP/network.py
pointer only (licence: NONE) · get_code("b6542dbfd17b1777")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, wjy5446/pytorch-Real-NVP 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.

InvertibleLayer Ran
ran on a synthesized input · constructed an object; no result was computed
hse-cs/probaforms/probaforms/models/realnvp.py
code served (permissive licence) · get_code("be475a4fed7464c2")
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, InvertibleLayer: 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)

# InvertibleLayer only takes var_size as a parameter in __init__
# The f() and g() methods take X and C as inputs but are not implemented (pass)
# So we just need to construct a valid var_size

var_size = 4

ARGS = {
    "var_size": var_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type InvertibleLayer — 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 InvertibleLayer 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
LinearBatchNorm Ran
ran on a synthesized input · constructed an object; no result was computed
ispamm/realnvp-demo-pytorch/rnvp.py
pointer only (licence: NONE) · get_code("e65c09e2e972e3c6")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ispamm/realnvp-demo-pytorch 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.

LinearCouplingLayer Ran
ran on a synthesized input · constructed an object; no result was computed
ispamm/realnvp-demo-pytorch/rnvp.py
pointer only (licence: NONE) · get_code("ee4d33b5da8d23fa")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ispamm/realnvp-demo-pytorch 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.

NN2 Ran
ran on a synthesized input
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("72ed3d4628c2fd6f")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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.

NormalizingFlow Ran
ran on a synthesized input · constructed an object; no result was computed
hse-cs/probaforms/probaforms/models/realnvp.py
code served (permissive licence) · get_code("53e531a482ac4dee")
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, NormalizingFlow: 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.distributions as dist

torch.manual_seed(0)

# Define a mock invertible layer that implements f and g methods
class MockInvertibleLayer(nn.Module):
    def __init__(self, var_size, cond_size=None):
        super(MockInvertibleLayer, self).__init__()
        self.var_size = var_size
        self.cond_size = cond_size
        self.linear = nn.Linear(var_size, var_size)
        if cond_size is not None:
            self.cond_linear = nn.Linear(cond_size, var_size)
    
    def f(self, X, C):
        """Forward transformation"""
        # X: [batch_size, var_size]
        # C: [batch_size, cond_size] or None
        # Returns transformed X and log determinant (change of variables)
        transformed = self.linear(X)
        if C is not None:
            transformed = transformed + self.cond_linear(C)
        # Simple log determinant approximation
        log_det = torch.zeros(X.shape[0])
        return transformed, log_det
    
    def g(self, X, C):
        """Inverse transformation"""
        # X: [batch_size, var_size]
        # C: [batch_size, cond_size] or None
        # Returns inverse transformed X
        inverse_linear = nn.Linear(self.var_size, self.var_size)
        transformed = inverse_linear(X)
        if C is not None:
            transformed = transformed + self.cond_linear(C)
        return transformed

# Create mock layers
var_size = 4
cond_size = 3
batch_size = 2

layer1 = MockInvertibleLayer(var_size, cond_size)
layer2 = MockInvertibleLayer(var_size, cond_size)

layers = [layer1, layer2]

# Create a prior distribution (standard normal)
prior = dist.Normal(torch.zeros(var_size), torch.ones(var_size))

ARGS = {
    "layers": layers,
    "prior": prior,
}
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 NormalizingFlow — 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 NormalizingFlow 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
Permutation Ran
ran on a synthesized input · constructed an object; no result was computed
ispamm/realnvp-demo-pytorch/rnvp.py
pointer only (licence: NONE) · get_code("a0db7e5e5ab30955")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ispamm/realnvp-demo-pytorch 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.

RNVP Ran
ran on a synthesized input · constructed an object; no result was computed
li012589/NeuralRG/flow/rnvp.py
code served (permissive licence) · get_code("30e1477bc4b57f7d")
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, RNVP: 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)

# Simple prior with sample() and logProbability() methods
class SimplePrior:
    def sample(self, batchSize):
        return torch.randn(batchSize, 4)
    
    def logProbability(self, z):
        return -0.5 * (z ** 2).sum(dim=1)

# Simple neural network for s and t transformations
class SimpleNet(nn.Module):
    def __init__(self, input_dim, output_dim):
        super(SimpleNet, self).__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        return self.linear(x)

# Construct masks - alternating mask pattern for RNVP
dim = 4
num_layers = 2
maskList = []
for i in range(num_layers):
    mask = torch.zeros(dim)
    mask[i % 2::2] = 1
    maskList.append(mask)
maskList = torch.stack(maskList)

# Construct transformation networks (s and t networks)
tList = []
sList = []
for i in range(num_layers):
    tList.append(SimpleNet(dim, dim))
    sList.append(SimpleNet(dim, dim))

# Create prior
prior = SimplePrior()

ARGS = {
    'maskList': maskList,
    'tList': tList,
    'sList': sList,
    'prior': prior,
    'name': 'RNVP'
}
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 RNVP — 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 RNVP 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
RealNVP Ran
ran on a synthesized input · constructed an object; no result was computed
wjy5446/pytorch-Real-NVP/network.py
pointer only (licence: NONE) · get_code("9a35cdb9141d51ec")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, wjy5446/pytorch-Real-NVP 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.

RealNVP Ran
ran on a synthesized input · constructed an object; no result was computed
e-hulten/real-nvp-2d/rnvp.py
pointer only (licence: NONE) · get_code("22e52a581dc49567")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, e-hulten/real-nvp-2d 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.

ResidualBlock Ran
ran on a synthesized input · constructed an object; no result was computed
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("02d7de503f8f151c")
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, ResidualBlock: 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,
    "bottleneck": True,
    "weight_norm": 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 ResidualBlock — 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 ResidualBlock 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
ResidualModule Ran
ran on a synthesized input · constructed an object; no result was computed
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("fbccf5ed84113182")
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, ResidualModule: 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': 3,
    'dim': 4,
    'out_dim': 5,
    'res_blocks': 2,
    'bottleneck': True,
    'skip': True,
    'weight_norm': 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 ResidualModule — 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 ResidualModule 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
ScaleTranslate Ran
ran on a synthesized input · constructed an object; no result was computed
e-hulten/real-nvp-2d/rnvp.py
pointer only (licence: NONE) · get_code("d40249b744f13e2f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, e-hulten/real-nvp-2d 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 · constructed an object; no result was computed
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("a3d33c4dfab6325f")
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, Squeeze: 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 Flow(nn.Module):

    def __init__(self):
        super(Flow, self).__init__()
        self.device = torch.device('cpu')

    def to(self, device):
        super(Flow, self).to(device)
        self.device = device

    def forward_flow(self, x):
        """ returns forward flow computation result and log of jacobian """
        raise NotImplementedError

    def inverse_flow(self, x):
        """ returns inverse flow computation result and log of jacobian  """
        raise NotImplementedError

    def data_init(self, x):
        return self.forward_flow(x)[0]

def squeeze__(image):
    b, c, h, w = image.shape
    image = image.reshape(b, c, h // 2, 2, w // 2, 2).permute(0, 1, 3, 5, 2, 4)
    return image.reshape(b, 4 * c, h // 2, w // 2)

def unsqueeze__(image):
    b, c, h, w = image.shape
    image = image.reshape(b, c // 4, 2, 2, h, w).permute(0, 1, 4, 2, 5, 3)
    return image.reshape(b, c // 4, 2 * h, 2 * w)

class Squeeze(Flow):

    def forward_flow(self, image):
        return (squeeze__(image), torch.zeros(image.shape[0], device=image.device))

    def inverse_flow(self, image):
        return (unsqueeze__(image), torch.zeros(image.shape[0], device=image.device))

# Create a sample image tensor: batch_size=2, channels=3, height=4, width=4
sample_image = torch.randn(2, 3, 4, 4)

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 Squeeze — 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 Squeeze 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
Unsqueeze Ran
ran on a synthesized input · constructed an object; no result was computed
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("681e06d7532d0edf")
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, Unsqueeze: 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 Unsqueeze(nn.Module):
    def __init__(self):
        super(Unsqueeze, self).__init__()
        self.device = torch.device('cpu')

    def to(self, device):
        super(Unsqueeze, self).to(device)
        self.device = device
        return self

    def forward_flow(self, image):
        b, c, h, w = image.shape
        image = image.reshape(b, c // 4, 2, 2, h, w).permute(0, 1, 4, 2, 5, 3)
        result = image.reshape(b, c // 4, 2 * h, 2 * w)
        return (result, torch.zeros(image.shape[0], device=image.device))

    def inverse_flow(self, image):
        b, c, h, w = image.shape
        image = image.reshape(b, c, h // 2, 2, w // 2, 2).permute(0, 1, 3, 5, 2, 4)
        result = image.reshape(b, 4 * c, h // 2, w // 2)
        return (result, torch.zeros(image.shape[0], device=image.device))

    def data_init(self, x):
        return self.forward_flow(x)[0]

unsqueeze_instance = Unsqueeze()

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 Unsqueeze — 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 Unsqueeze 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
WeightNormConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("355dc98275bca37b")
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, WeightNormConv2d: 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": 3,
    "out_dim": 16,
    "kernel_size": 3,
    "stride": 1,
    "padding": 1,
    "bias": True,
    "weight_norm": True,
    "scale": 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 WeightNormConv2d — 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 WeightNormConv2d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
__get_channelwise_mask Ran
ran; our contract draft was wrong, not the code
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("3f4a65a29736ed2e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Analyze the function:
# - shape: used as indexing argument, needs to be a tuple/list of ints
# - device: used in torch.zeros(), should be a torch device
# - dtype: used in torch.zeros(), should be a torch dtype

ARGS = {
    "shape": (4, 3),
    "device": torch.device("cpu"),
    "dtype": torch.float32,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (4, 3) — shape (4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
__get_checkerboard_mask Ran
ran; our contract draft was wrong, not the code
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("60ea456ee4a19edf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    "shape": (2, 3, 4),
    "device": "cpu",
    "dtype": torch.float32
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 4) — shape (2, 3, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
actnorm Ran
ran on a synthesized input
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("c793226ebfc2c109")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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.

affine_coupling Ran
ran on a synthesized input
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("cd3b1c121c287a2d")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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.

affine_coupling_layer Ran
ran on a synthesized input · constructed an object; no result was computed
CompPhysVienna/paper_NF_for_rare_events/conditional_bg/network/coupling_layers.py
code served (permissive licence) · get_code("94f135340bbbc29f")
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, affine_coupling_layer: 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)

# Define a minimal transform network that satisfies the interface
class MinimalTransformNetwork(nn.Module):
    def __init__(self, input_dim, output_dim, condition_dim, activation=nn.ReLU):
        super(MinimalTransformNetwork, self).__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim
        self.condition_dim = condition_dim
        self.activation = activation
        
        # Simple linear network
        combined_dim = input_dim + (condition_dim if condition_dim else 0)
        self.net = nn.Sequential(
            nn.Linear(combined_dim, 8),
            activation(),
            nn.Linear(8, output_dim)
        )
    
    def forward(self, x, condition=None):
        if condition is not None:
            x = torch.cat([x, condition], dim=-1)
        return self.net(x)

# Create dimension parameters
dimension = 4
condition_dim = 3
batch_size = 2

# Create masks - alternating pattern for coupling
mask1 = torch.tensor([0, 1, 0, 1], dtype=torch.float32)
mask2 = torch.tensor([1, 0, 1, 0], dtype=torch.float32)
masks = torch.stack([mask1, mask2])

# Transform network class
transform_network = MinimalTransformNetwork

# Transform network parameters
transform_network_parameters = {
    'activation': nn.ReLU
}

# Build ARGS dictionary
ARGS = {
    'dimension': dimension,
    'masks': masks,
    'transform_network': transform_network,
    'transform_network_parameters': transform_network_parameters,
    'condition_dim': condition_dim,
    'clamp_scale': 1.0,
    'scale_tanh': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type affine_coupling_layer — 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 affine_coupling_layer 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
bits_per_dim Ran
ran on a synthesized input
chrischute/real-nvp/util/optim_util.py
code served (permissive licence) · get_code("54dc276d7472ea04")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create a small input tensor with batch size 2 and spatial dimensions
x = torch.randn(2, 4, 3)

# Create a scalar negative log-likelihood loss tensor
nll = torch.tensor(2.5)

ARGS = {
    "x": x,
    "nll": nll,
}
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 Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
checkNan Ran
ran; our contract draft was wrong, not the code
li012589/NeuralRG/flow/rnvp.py
code served (permissive licence) · get_code("299ba404a9f96d42")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function checkNan takes a tensor x and asserts that it contains no NaN values
# It returns the tensor if the assertion passes
# We need to provide a tensor with no NaN values

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
checkerboard_mask Ran
ran on a synthesized input
chrischute/real-nvp/util/array_util.py
code served (permissive licence) · get_code("45a2481cb9317453")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (1, 1, 4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    "height": 4,
    "width": 3,
    "reverse": False,
    "dtype": torch.float32,
    "device": None,
    "requires_grad": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (1, 1, 4, 3) — shape (1, 1, 4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
default_initializer Ran
ran; our contract draft was wrong, not the code
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("a256d12412fc4a42")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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.

entry_stop_gradients Ran
ran, honoured the contract we drafted
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("137d28e6764f74e4")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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.

flatten_sum Ran
ran; our contract draft was wrong, not the code
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("42a8363a66214fb0")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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.

gen_network Ran
ran; our contract draft was wrong, not the code
hse-cs/probaforms/probaforms/models/realnvp.py
code served (permissive licence) · get_code("8f1ce335784ce477")
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 Sequential. 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_inputs': 5,
    'n_outputs': 3,
    'hidden': (10, 8),
    'activation': 'tanh'
}
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 Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential 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
get_mask Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("a9af99f9dc6f88d5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Test case 1: checkerboard pattern with 3D shape
shape_1 = (2, 3, 4)
pattern_1 = 'checkerboard'
device_1 = 'cpu'
dtype_1 = torch.float

# Test case 2: channelwise pattern with 2D shape
shape_2 = (4, 3)
pattern_2 = 'channelwise'
device_2 = 'cpu'
dtype_2 = torch.float32

# Use the first test case for ARGS
ARGS = {
    'shape': shape_1,
    'pattern': pattern_1,
    'device': device_1,
    'dtype': dtype_1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 4) — shape (2, 3, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_norm_layer Ran
ran on a synthesized input
chrischute/real-nvp/util/norm_util.py
code served (permissive licence) · get_code("94af3072059ccb7c")
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 partial. 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 functools
import torch.nn as nn

torch_manual_seed_code = "import torch; torch.manual_seed(0)"

ARGS = {
    'norm_type': 'instance'
}
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 partial — a partial output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a partial output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_param_groups Ran
ran on a synthesized input
chrischute/real-nvp/util/norm_util.py
code served (permissive licence) · get_code("e478f33fab38d758")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple neural network with some parameters that have the norm_suffix
class SimpleNet(nn.Module):
    def __init__(self):
        super(SimpleNet, self).__init__()
        # Create a linear layer
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
        
        # Manually add a parameter with the norm_suffix 'weight_g'
        # This simulates a WeightNorm layer
        self.register_parameter('weight_g', nn.Parameter(torch.randn(2)))
        
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

net = SimpleNet()

ARGS = {
    'net': net,
    'weight_decay': 0.01,
    'norm_suffix': 'weight_g',
    'verbose': 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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
int_shape Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from openai/weightnorm
pointer only · get_code("3ba6aec873833b39")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. 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.

log_standard_Gaussian Ran
ran, honoured the contract we drafted
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("678f84c2e99d3379")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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.

squeeze_2x2 Ran
ran on a synthesized input
chrischute/real-nvp/util/array_util.py
code served (permissive licence) · get_code("4c2fbb62a7ee51a7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 16, 2, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small tensor for the forward pass (squeeze operation)
# Shape: (B=2, C=4, H=4, W=6) - channels divisible by 4, spatial dims even
x_forward = torch.randn(2, 4, 4, 6)

# Create a small tensor for the reverse pass (unsqueeze operation)
# Shape: (B=2, C=16, H=2, W=3) - channels divisible by 4
x_reverse = torch.randn(2, 16, 2, 3)

ARGS = {
    "x": x_forward,
    "reverse": False,
    "alt_order": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 16, 2, 3) — shape (2, 16, 2, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
squeeze__ Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("314875cfe97c786f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 16, 3, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function squeeze__ expects:
# - image: a 4D tensor with shape (b, c, h, w)
# The function reshapes it by:
# 1. Reshaping to (b, c, h//2, 2, w//2, 2)
# 2. Permuting to (0, 1, 3, 5, 2, 4)
# 3. Reshaping to (b, 4*c, h//2, w//2)
# So h and w must be even numbers

# Create a small test image tensor
# batch=2, channels=4, height=6, width=8
image = torch.randn(2, 4, 6, 8)

ARGS = {
    "image": image
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 16, 3, 4) — shape (2, 16, 3, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
unsqueeze__ Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mahkons/flows/RealNVP/RealNVP.py
code served (permissive licence) · get_code("e24beea03138f40b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 1, 6, 10) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a test image tensor with shape (b, c, h, w)
# The function expects c to be divisible by 4
# Using batch=2, channels=4, height=3, width=5
image = torch.randn(2, 4, 3, 5)

ARGS = {
    "image": image
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 1, 6, 10) — shape (2, 1, 6, 10) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Basic Not yet run
unverified (harvested, no recorded run)
sshish/NF/NF.py
pointer only (licence: NONE) · get_code("ccc805cf7aa6b406")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, sshish/NF 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. No invocation recorded.

CDF_quadratic Not yet run
unverified (harvested, no recorded run)
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("32da9e5b5ad57d81")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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. No invocation recorded.

ChannelwiseAdditiveCoupling Not yet run
unverified (harvested, no recorded run)
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("2905171d5a4f77df")
How it ran

no invocation recorded.

ChannelwiseAffineCoupling Not yet run
unverified (harvested, no recorded run)
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("81950ec1ad22027a")
How it ran

no invocation recorded.

ChannelwiseCoupling Not yet run
unverified (harvested, no recorded run)
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("349f39c743f5c526")
How it ran

no invocation recorded.

CheckerboardAdditiveCoupling Not yet run
unverified (harvested, no recorded run)
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("00838e83ac1117ce")
How it ran

no invocation recorded.

CheckerboardAffineCoupling Not yet run
unverified (harvested, no recorded run)
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("d77fd2f51f8dae00")
How it ran

no invocation recorded.

CheckerboardCoupling Not yet run
unverified (harvested, no recorded run)
fmu2/realNVP/realnvp.py
code served (permissive licence) · get_code("f5044d42673aa07f")
How it ran

no invocation recorded.

CouplingLayer Not yet run
unverified (harvested, no recorded run)
sshish/NF/NF.py
pointer only (licence: NONE) · get_code("de669ff9f15b36d1")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, sshish/NF 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. No invocation recorded.

CouplingLayer Not yet run
unverified (harvested, no recorded run)
taesung89/real-nvp/real_nvp/nn.py
pointer only (licence: NONE) · get_code("f62eb0295869e616")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, taesung89/real-nvp 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. No invocation recorded.

IM_rNVP Not yet run
unverified (harvested, no recorded run)
xlwan/KRnet/BR_lib/BR_model.py
pointer only (licence: GPL-3.0) · get_code("3c41f44f3a1ddce7")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, xlwan/KRnet, 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. No invocation recorded.

Layer Not yet run
unverified (harvested, no recorded run)
taesung89/real-nvp/real_nvp/nn.py
pointer only (licence: NONE) · get_code("0e7224db7dc1693f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, taesung89/real-nvp 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. No invocation recorded.

Showing the first 60. An agent gets every one from get_harvested_code_for_paper("1605.08803").

Also present in other papers' code

Of 70 samples lifted from this paper's linked repositories, 70 carry a function identity and 0 do not. Of the 70 distinct functions with an identity, 6 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 6 later) and 64 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
int_shape
brick of this paper's assembly: helper
4 other papers
1 dated earlier, 3 later
1602.07868 (2016-02)
Weight Normalization: A Simple Reparameterization to Accelerate Training of Deep Neural Networks
1702.08658 (2017-02)
Towards Deeper Understanding of Variational Autoencoding Models
1807.10589 (2018-07)
Diverse feature visualizations reveal invariances in early layers of deep neural networks
and 1 more not named here
AffineCoupling
brick of this paper's assembly: entry
1 other paper
1 later · 1 only as an assembly brick
1807.03039 (2018-07)
Glow: Generative Flow with Invertible 1x1 Convolutions
ConditionalFlow
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
1705.07057 (2017-05)
Masked Autoregressive Flow for Density Estimation
Conv2dZeroInit
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
1807.03039 (2018-07)
Glow: Generative Flow with Invertible 1x1 Convolutions
bits_per_dim 1 other paper
1 later
2106.03357 (2021-06)
Evaluating State-of-the-Art Classification Models Against Bayes Optimality
entry_stop_gradients
brick of this paper's assembly: helper
1 other paper
1 later
2112.14038 (2021-12)
DAS-PINNs: A deep adaptive sampling method for solving high-dimensional partial differential equations

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

More are linked than shown here.

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

Abstract

Unsupervised learning of probabilistic models is a central yet challenging problem in machine learning. Specifically, designing models with tractable learning, sampling, inference and evaluation is crucial in solving this task. We extend the space of such models using real-valued non-volume preserving (real NVP) transformations, a set of powerful, stably invertible, and learnable transformations, resulting in an unsupervised learning algorithm with exact log-likelihood computation, exact and efficient sampling, exact and efficient inference of latent variables, and an interpretable latent space. We demonstrate its ability to model natural images on four datasets through sampling, log-likelihood evaluation, and latent variable manipulations.

For agents

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

get_harvested_code_for_paper("1605.08803")
get_code_for_paper("1605.08803")
have("1605.08803")

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

curl https://syntology.ai/api/ran/1605.08803.json

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

[![Syntology run record](https://syntology.ai/api/ran/1605.08803.svg)](https://syntology.ai/paper/1605.08803)

Connect an agent — have() is free.