SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2007.03898 · NeurIPS · 2020

NVAE: A Deep Hierarchical Variational Autoencoder

Jan Kautz, Arash Vahdat

arXiv · PDF · Open in the Atlas

Code that ran

26 of 41 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (15 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 18 have no contract verdict; 5 failures of our instrument; 15 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/2007.03898.json · badge: /api/ran/2007.03898.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
NVlabs/NVAE canonical (classifier label)
classifier basis: classifier_2025
9 ran · 14 no recorded run
SerezD/NVAE-from-scratch community
Papers with Code archive
14 ran · 1 no recorded run
oadonca/ANVAE community (classifier label)
classifier basis not recorded on the link
3 ran · 0 no recorded run
FunctionStatusWhere it lives
DecCombinerCell Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("e79daa47e7ea4581")
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, DecCombinerCell: 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
from torch.nn import Conv2d
from torch.nn.utils.parametrizations import weight_norm

torch.manual_seed(0)

class DecCombinerCell(nn.Module):

    def __init__(self, feature_channels: int, z_channels: int, out_channels: int):
        """
        combine feature + noise channels during decoding / generation
        """
        super().__init__()
        self.conv = weight_norm(Conv2d(feature_channels + z_channels, out_channels, kernel_size=1, bias=True))

    def forward(self, x, z):
        out = torch.cat([x, z], dim=1)
        out = self.conv(out)
        return out


ARGS = {
    'feature_channels': 3,
    'z_channels': 5,
    'out_channels': 7,
}
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 DecCombinerCell — 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 DecCombinerCell 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
DiscMixLogistic Ran
ran on a synthesized input · constructed an object; no result was computed
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("3b69906c3f162603")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

DiscMixLogistic Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("dc14f24bd29f3b0d")
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, DiscMixLogistic: 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 numpy as np

torch.manual_seed(0)

# The DiscMixLogistic class expects:
# params: torch.Tensor of shape (B, N_MIX + N_MIX * 3 * C, H, W)
# where N_MIX is the number of mixtures, C is img_channels (3)
# img_channels: int = 3
# num_bits: int = 8

# Let's construct valid parameters
batch_size = 2
img_channels = 3
num_bits = 8
height = 4
width = 5

# Number of mixtures - let's use 2
num_mixtures = 2

# params shape: (B, N_MIX + N_MIX * 3 * C, H, W)
# = (B, 2 + 2 * 3 * 3, H, W)
# = (B, 2 + 18, H, W)
# = (B, 20, H, W)
num_params_per_mixture = 1 + img_channels * 3
total_channels = num_mixtures * (1 + img_channels * 3)

params = torch.randn(batch_size, total_channels, height, width)

ARGS = {
    "params": params,
    "img_channels": 3,
    "num_bits": 8
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DiscMixLogistic — 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 DiscMixLogistic 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
EncCombinerCell Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("fd98ee5b850457e9")
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, EncCombinerCell: 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
from torch.nn import Conv2d
from torch.nn.utils.parametrizations import weight_norm

class EncCombinerCell(nn.Module):

    def __init__(self, in_channels: int, out_channels: int):
        super().__init__()
        self.conv = weight_norm(Conv2d(in_channels, out_channels, kernel_size=1, bias=True))

    def forward(self, x_enc, x_dec):
        """
        combine encoder ad decoder features
        """
        x_dec = self.conv(x_dec)
        out = x_enc + x_dec
        return out

torch.manual_seed(0)

ARGS = {
    'in_channels': 4,
    'out_channels': 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 EncCombinerCell — 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 EncCombinerCell 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
Global_Average_Pooling Ran
ran on a synthesized input
oadonca/ANVAE/modules.py
code served (permissive licence) · get_code("5f0c5cf7f773ac3b")
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 EagerTensor, ndim 2, dtype family float, shape (2, 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 tensorflow as tf
import numpy as np

# Create a small 4D tensor (batch_size=2, height=4, width=3, channels=5)
# GlobalAveragePooling2D expects input shape (batch, height, width, channels)
x = tf.constant(np.random.randn(2, 4, 3, 5).astype(np.float32))

ARGS = {
    'x': x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 2, dtype family float, shape (2, 5) — shape (2, 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: none by the harness; any seed is in the setup block
MaskedConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("ea5de3e39f202a06")
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, MaskedConv2d: 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 = {
    'mirror': False,
    'zero_diag': False,
    'in_channels': 3,
    'out_channels': 4,
    'kernel_size': 3,
    'stride': 1,
    'padding': 1,
    'dilation': 1,
    'groups': 1,
    'bias': True,
    'padding_mode': 'zeros',
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MaskedConv2d — 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 MaskedConv2d 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
NFBlock Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("2bf1bbe0e0e1f51c")
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, NFBlock: 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_z': 4,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type NFBlock — 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 NFBlock 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
NFCell Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("a82d266ee79c8089")
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, NFCell: 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 = {
    'num_z': 4,
    'mirror': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type NFCell — 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 NFCell 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
Normal Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("c9ab03bfda9403fb")
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, Normal: 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 numpy as np

torch.manual_seed(0)

# Create small tensors for mu and log_sigma
batch_size = 2
latent_dim = 4

mu = torch.randn(batch_size, latent_dim)
log_sigma = torch.randn(batch_size, latent_dim)
temp = 1.0

ARGS = {
    "mu": mu,
    "log_sigma": log_sigma,
    "temp": temp
}
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 Normal — 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 Normal 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
ResidualCellDecoder Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("21cbfdf47981afd2")
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, ResidualCellDecoder: 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_channels": 4,
    "out_channels": 8,
    "upsampling": True,
    "use_SE": True,
    "hidden_mul": 6
}
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 ResidualCellDecoder — 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 ResidualCellDecoder 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
ResidualCellEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("1d1460d11268fcbe")
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, ResidualCellEncoder: 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_channels": 4,
    "out_channels": 8,
    "downsampling": True,
    "use_SE": 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 ResidualCellEncoder — 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 ResidualCellEncoder 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
SE Ran
ran on a synthesized input · constructed an object; no result was computed
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("47f8add81ef809e7")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

SE Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("852119a74dcd01db")
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, SE: 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_channels": 16,
    "out_channels": 32,
}
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 SE — 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 SE 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
SkipDown Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("90b219a8fe64989e")
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, SkipDown: 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
from torch.nn import Conv2d
from torch.nn.utils.parametrizations import weight_norm

torch.manual_seed(0)

ARGS = {
    "in_channels": 3,
    "out_channels": 16,
    "stride": 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 SkipDown — 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 SkipDown 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
SkipUp Ran
ran on a synthesized input · constructed an object; no result was computed
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("a8aa4571eb4d2e83")
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, SkipUp: 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
from torch.nn import Conv2d
from torch.nn.utils.parametrizations import weight_norm

torch.manual_seed(0)

class SkipUp(nn.Module):

    def __init__(self, in_channels: int, out_channels: int, stride: int):
        super().__init__()
        self.conv = weight_norm(Conv2d(in_channels, out_channels, kernel_size=1, stride=stride))

    def forward(self, x):
        x = torch.nn.functional.interpolate(x, scale_factor=2, mode='bilinear', align_corners=True)
        return self.conv(x)

ARGS = {
    'in_channels': 3,
    'out_channels': 4,
    'stride': 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 SkipUp — 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 SkipUp 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
SyncBatchNormSwish Ran
ran on a synthesized input · constructed an object; no result was computed
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("b014ff059854d82e")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

batchNorm Ran
ran on a synthesized input
oadonca/ANVAE/modules.py
code served (permissive licence) · get_code("c5a084857e4bb9e0")
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 SymbolicTensor. 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 tensorflow as tf
import numpy as np

# Create a small batch of input data
# The function uses axes=[0, 1, 2] in tf.nn.moments, so we need at least 3 dimensions
# Shape: (batch=2, height=3, width=4, channels=5)
cur_input = tf.constant(np.random.randn(2, 3, 4, 5).astype(np.float32))

# Construct the layer_dict with cur_input
layer_dict = {
    'cur_input': cur_input
}

ARGS = {
    'layer_dict': layer_dict,
    'name': 'batch_norm'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SymbolicTensor — a SymbolicTensor output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_input_size Ran
ran, honoured the contract we drafted
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("4ca5ffe947c454b7")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

get_stride_for_cell_type Ran
ran, honoured the contract we drafted
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("417c413f2e89d954")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

groups_per_scale Ran
ran, honoured the contract we drafted
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("ec8681cb9635d8c2")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

gumbel_sampling Ran
ran; our contract draft was wrong, not the code
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("cb77f2d04693965c")
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)

# Create a small logits tensor: batch_size=2, num_classes=4
logits = torch.randn(2, 4)

ARGS = {
    "logits": logits,
    "temperature": 1.0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
mix_log_cdf_flow Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("9d358b42347bc9fc")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

normalize_weight_jit Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("509ea43acade44e9")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

one_hot Ran
ran; our contract draft was wrong, not the code
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("b95bc8179cca09d6")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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.

soft_clamp Ran
ran; our contract draft was wrong, not the code
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("081e9e26fd1942f7")
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 takes a tensor x and a float n
# It performs: x.div(n).tanh_().mul(n)
# So x must be a torch.Tensor, and n is a float (default 5.0)

x = torch.randn(2, 4)
n = 5.0

ARGS = {
    "x": x,
    "n": n,
}
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
swish Ran
ran on a synthesized input
oadonca/ANVAE/modules.py
code served (permissive licence) · get_code("248788c293d8eccd")
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 EagerTensor, 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 tensorflow as tf
import numpy as np

# Create a small tensor for cur_input
cur_input = tf.constant([[1.0, -2.0, 3.0, -4.0],
                         [2.0, -1.0, 4.0, -3.0]], dtype=tf.float32)

# Create the layer_dict that the function expects
layer_dict = {
    'cur_input': cur_input
}

ARGS = {
    'layer_dict': layer_dict,
    'name': 'swish'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, 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
ARConv2d Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("57dee14e4512c612")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

ARInvertedResidual Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("213dfd5e4d31949b")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

AutoEncoder Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("ed8aab6b16f71a33")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

AutoEncoder Not yet run
unverified (harvested, no recorded run)
SerezD/NVAE-from-scratch/src/model.py
code served (permissive licence) · get_code("2b09442d4473a79b")
How it ran

no invocation recorded.

Cell Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("095d98d4a1f55bfd")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

CellAR Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("a0ce4d356a170530")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

Conv2D Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("a674092cc931b81d")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

DecCombinerCell Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("05cbbc54fd34ad83")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

EncCombinerCell Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("464aa3f0a27bcf6b")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

MixLogCDFParam Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("c4d8db6435b9c07a")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

Normal Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("277ed0e40f07a938")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

NormalDecoder Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("da4e99992612094c")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

PairedCellAR Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("04bf8ed49665f416")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

average_tensor Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("dd83a8a0b04c47fd")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

get_skip_connection Not yet run
unverified (harvested, no recorded run)
NVlabs/NVAE/model.py
pointer only (licence: NOASSERTION) · get_code("b8e68f7d505c301c")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NVlabs/NVAE (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. No invocation recorded.

Also present in other papers' code

Of 41 samples lifted from this paper's linked repositories, 41 carry a function identity and 0 do not. Of the 41 distinct functions with an identity, 6 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 6 later) and 35 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
get_stride_for_cell_type 2 other papers
2 later
2106.06819 (2021-06)
D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
2309.00073 (2023-09)
Diffusion Variational Autoencoder for Tackling Stochasticity in Multi-Step Regression Stock Price Prediction
Conv2D 1 other paper
1 later
2010.00654 (2020-10)
VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models
average_tensor 1 other paper
1 later
2010.00654 (2020-10)
VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models
get_input_size 1 other paper
1 later
2309.00073 (2023-09)
Diffusion Variational Autoencoder for Tackling Stochasticity in Multi-Step Regression Stock Price Prediction
get_skip_connection 1 other paper
1 later
2106.06819 (2021-06)
D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
normalize_weight_jit 1 other paper
1 later
2010.00654 (2020-10)
VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models

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

Normalizing flows, autoregressive models, variational autoencoders (VAEs), and deep energy-based models are among competing likelihood-based frameworks for deep generative learning. Among them, VAEs have the advantage of fast and tractable sampling and easy-to-access encoding networks. However, they are currently outperformed by other models such as normalizing flows and autoregressive models. While the majority of the research in VAEs is focused on the statistical challenges, we explore the orthogonal direction of carefully designing neural architectures for hierarchical VAEs. We propose Nouveau VAE (NVAE), a deep hierarchical VAE built for image generation using depth-wise separable convolutions and batch normalization. NVAE is equipped with a residual parameterization of Normal distributions and its training is stabilized by spectral regularization. We show that NVAE achieves state-of-the-art results among non-autoregressive likelihood-based models on the MNIST, CIFAR-10, CelebA 64, and CelebA HQ datasets and it provides a strong baseline on FFHQ. For example, on CIFAR-10, NVAE pushes the state-of-the-art from 2.98 to 2.91 bits per dimension, and it produces high-quality images on CelebA HQ as shown in Fig. 1. To the best of our knowledge, NVAE is the first successful VAE applied to natural images as large as 256×256 pixels. The source code is available at https://github.com/NVlabs/NVAE.

For agents

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

get_harvested_code_for_paper("2007.03898")
get_code_for_paper("2007.03898")
have("2007.03898")

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

curl https://syntology.ai/api/ran/2007.03898.json

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

[![Syntology run record](https://syntology.ai/api/ran/2007.03898.svg)](https://syntology.ai/paper/2007.03898)

Connect an agent — have() is free.