SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2011.10650 · 2020

Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images

arXiv · PDF · Open in the Atlas

Code that ran

8 of 9 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 6 failures of our instrument; 1 no recorded run), measured 2026-11-21; 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/2011.10650.json · badge: /api/ran/2011.10650.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
openai/vdvae canonical (classifier label)
classifier basis: classifier_2025
4 ran · 0 no recorded run
ericl122333/latent-vae reimplementation (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
vvvm23/vdvae reimplementation (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
copy not recorded community
not in the paper's declared repositories
1 ran · 0 no recorded run
EugenHotaj/pytorch-generative reimplementation (classifier label)
classifier basis: classifier_2025
0 ran · 1 no recorded run
FunctionStatusWhere it lives
add_vae_arguments Ran
ran; our contract draft was wrong, not the code
openai/vdvae/hps.py
code served (permissive licence) · get_code("a169b353d342ad7f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned ArgumentParser. 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 argparse

parser = argparse.ArgumentParser()

# Call the function to add all arguments
def add_vae_arguments(parser):
    parser.add_argument('--seed', type=int, default=0)
    parser.add_argument('--port', type=int, default=29500)
    parser.add_argument('--save_dir', type=str, default='./saved_models')
    parser.add_argument('--data_root', type=str, default='./')
    parser.add_argument('--desc', type=str, default='test')
    parser.add_argument('--hparam_sets', '--hps', type=str)
    parser.add_argument('--restore_path', type=str, default=None)
    parser.add_argument('--restore_ema_path', type=str, default=None)
    parser.add_argument('--restore_log_path', type=str, default=None)
    parser.add_argument('--restore_optimizer_path', type=str, default=None)
    parser.add_argument('--dataset', type=str, default='cifar10')
    parser.add_argument('--ema_rate', type=float, default=0.999)
    parser.add_argument('--enc_blocks', type=str, default=None)
    parser.add_argument('--dec_blocks', type=str, default=None)
    parser.add_argument('--zdim', type=int, default=16)
    parser.add_argument('--width', type=int, default=512)
    parser.add_argument('--custom_width_str', type=str, default='')
    parser.add_argument('--bottleneck_multiple', type=float, default=0.25)
    parser.add_argument('--no_bias_above', type=int, default=64)
    parser.add_argument('--scale_encblock', action='store_true')
    parser.add_argument('--test_eval', action='store_true')
    parser.add_argument('--warmup_iters', type=float, default=0)
    parser.add_argument('--num_mixtures', type=int, default=10)
    parser.add_argument('--grad_clip', type=float, default=200.0)
    parser.add_argument('--skip_threshold', type=float, default=400.0)
    parser.add_argument('--lr', type=float, default=0.00015)
    parser.add_argument('--lr_prior', type=float, default=0.00015)
    parser.add_argument('--wd', type=float, default=0.0)
    parser.add_argument('--wd_prior', type=float, default=0.0)
    parser.add_argument('--num_epochs', type=int, default=10000)
    parser.add_argument('--n_batch', type=int, default=32)
    parser.add_argument('--adam_beta1', type=float, default=0.9)
    parser.add_argument('--adam_beta2', type=float, default=0.9)
    parser.add_argument('--temperature', type=float, default=1.0)
    parser.add_argument('--iters_per_ckpt', type=int, default=25000)
    parser.add_argument('--iters_per_print', type=int, default=1000)
    parser.add_argument('--iters_per_save', type=int, default=10000)
    parser.add_argument('--iters_per_images', type=int, default=10000)
    parser.add_argument('--epochs_per_eval', type=int, default=10)
    parser.add_argument('--epochs_per_probe', type=int, default=None)
    parser.add_argument('--epochs_per_eval_save', type=int, default=20)
    parser.add_argument('--num_images_visualize', type=int, default=8)
    parser.add_argument('--num_variables_visualize', type=int, default=6)
    parser.add_argument('--num_temperatures_visualize', type=int, default=3)
    return parser

ARGS = {
    'parser': parser
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArgumentParser — a ArgumentParser output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ArgumentParser 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
draw_gaussian_diag_samples Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
vvvm23/vdvae/vae.py
pointer only (licence: NONE) · get_code("31e670f6fff8614d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, vvvm23/vdvae 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.

gaussian_analytical_kl Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
this paper's copy was not recorded; identical code first harvested from SerChirag/rs-imle
pointer only; the stored copy get_code serves, from SerChirag/rs-imle, is pointer only (licence: NONE) · get_code("f8752a89ed52495c")
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.

get_evenly_spaced_indices Ran
ran, honoured the contract we drafted
ericl122333/latent-vae/vae/model.py
pointer only (licence: NONE) · get_code("2ec4359e4ba42b26")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ericl122333/latent-vae 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.

get_resolutions Ran
ran, honoured the contract we drafted
ericl122333/latent-vae/vae/model.py
pointer only (licence: NONE) · get_code("070ddf6b065864d3")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ericl122333/latent-vae 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.

get_width_settings Ran
ran; our contract draft was wrong, not the code
openai/vdvae/vae.py
code served (permissive licence) · get_code("9419abf027f0115d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned defaultdict. 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 collections import defaultdict

# The function takes two parameters:
# 1. width: an integer representing a default width value
# 2. s: a string that can be None or a comma-separated list of "key:value" pairs

# Test case 1: s is None or empty
width = 64
s = None

# Test case 2: s has some mappings
# Let's use a non-None s to test the parsing logic
width = 128
s = "0:256,1:512,3:1024"

ARGS = {
    "width": width,
    "s": s
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type defaultdict — a defaultdict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a defaultdict 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
pad_channels Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
openai/vdvae/vae.py
code served (permissive licence) · get_code("dd095f22dc8304f0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned Tensor, ndim 4, dtype family float, shape (2, 6, 5, 7) 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 4D tensor with shape (batch, channels, height, width)
# Using distinct dimensions: batch=2, channels=3, height=5, width=7
t = torch.randn(2, 3, 5, 7)

# width parameter should be >= d2 (channels dimension) to make sense
# Let's use width=6 which is larger than the current channel count of 3
width = 6

ARGS = {
    't': t,
    'width': width
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
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, 6, 5, 7) — shape (2, 6, 5, 7) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_layer_string Ran
ran; our contract draft was wrong, not the code
openai/vdvae/vae.py
code served (permissive licence) · get_code("95d3df9fe1921426")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned list, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

ARGS = {
    's': '64x2,128m4,256d2,512'
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, 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
reproduce Not yet run
unverified (harvested, no recorded run)
EugenHotaj/pytorch-generative/pytorch_generative/models/vae/vd_vae.py
code served (permissive licence) · get_code("6e32683806d3201d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 distinct functions with an identity, 5 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 5 later) and 4 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
gaussian_analytical_kl 1 other paper
1 later
2106.01798 (2021-06)
Implicit MLE: Backpropagating Through Discrete Exponential Family Distributions
get_evenly_spaced_indices 1 other paper
1 later
2303.13714 (2023-03)
High Fidelity Image Synthesis With Deep VAEs In Latent Space
get_resolutions 1 other paper
1 later
2303.13714 (2023-03)
High Fidelity Image Synthesis With Deep VAEs In Latent Space
get_width_settings 1 other paper
1 later
2201.12414 (2022-01)
Posterior Matching for Arbitrary Conditioning
parse_layer_string 1 other paper
1 later
2201.12414 (2022-01)
Posterior Matching for Arbitrary Conditioning

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 2026-11-21.

Repositories linked to this paper

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

Abstract

We present a hierarchical VAE that, for the first time, generates samples quickly while outperforming the PixelCNN in log-likelihood on all natural image benchmarks. We begin by observing that, in theory, VAEs can actually represent autoregressive models, as well as faster, better models if they exist, when made sufficiently deep. Despite this, autoregressive models have historically outperformed VAEs in log-likelihood. We test if insufficient depth explains why by scaling a VAE to greater stochastic depth than previously explored and evaluating it CIFAR-10, ImageNet, and FFHQ. In comparison to the PixelCNN, these very deep VAEs achieve higher likelihoods, use fewer parameters, generate samples thousands of times faster, and are more easily applied to high-resolution images. Qualitative studies suggest this is because the VAE learns efficient hierarchical visual representations. We release our source code and models at https://github.com/openai/vdvae.

For agents

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

get_harvested_code_for_paper("2011.10650")
get_code_for_paper("2011.10650")
have("2011.10650")

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

curl https://syntology.ai/api/ran/2011.10650.json

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

[![Syntology run record](https://syntology.ai/api/ran/2011.10650.svg)](https://syntology.ai/paper/2011.10650)

Connect an agent — have() is free.