SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2601.07293 · 2026

Inference-Time Scaling for Visual AutoRegressive modeling by Searching Representative Samples

Siyu Li, Xiumei Wang, Weidong Tang, Xinyan Wan

arXiv · PDF · Open in the Atlas

Code that ran

7 of 13 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; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 7 failures of our instrument; 6 no recorded run), measured 2026-11-22; 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/2601.07293.json · badge: /api/ran/2601.07293.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
WD7ang/VAR-Scaling canonical (classifier label)
classifier basis: census
7 ran · 6 no recorded run
FunctionStatusWhere it lives
Normalize Ran
ran; our contract draft was wrong, not the code
WD7ang/VAR-Scaling/VAR-main/models/basic_vae.py
code served (permissive licence) · get_code("9fcdaa6e423e8aa7")
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 GroupNorm. 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

torch.manual_seed(0)

# The Normalize function takes in_channels and num_groups (with default 32)
# It returns a GroupNorm layer

# in_channels: number of channels for the GroupNorm layer
# num_groups: number of groups to separate the channels into (default 32)

# We need to provide valid arguments where num_groups divides in_channels evenly
# Let's use in_channels=64 so it's divisible by the default num_groups=32

ARGS = {
    'in_channels': 64,
    'num_groups': 32
}
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 GroupNorm — a GroupNorm output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a GroupNorm 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
allgather Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
WD7ang/VAR-Scaling/VAR-main/dist.py
code served (permissive licence) · get_code("e847ee15f089dee6")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type, 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.distributed as tdist

torch.manual_seed(0)

# Create a small tensor for testing
# The function checks t.is_cuda and uses it in all_gather and torch.cat
# Shape: (2, 4) - small batch with 4 dimensions
tensor_arg = torch.randn(2, 4)

# Define ARGS dict with all parameters
ARGS = {
    "t": tensor_arg,
    "cat": 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 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
drop_path Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
WD7ang/VAR-Scaling/VAR-main/models/helpers.py
code served (permissive licence) · get_code("971ae8d2d8313e30")
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) and returned Tensor, ndim 4, dtype family float, shape (2, 4, 3, 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)

# Create a small tensor for testing
# Shape: (batch_size=2, channels=4, height=3, width=3) - 4D tensor
x = torch.randn(2, 4, 3, 3)

ARGS = {
    "x": x,
    "drop_prob": 0.5,
    "training": True,
    "scale_by_keep": True,
}
Setup block drafted by
a language model, Claude Haiku 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, 4, 3, 3) — shape (2, 4, 3, 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
gumbel_softmax_with_rng Ran
ran; our contract draft was wrong, not the code
WD7ang/VAR-Scaling/VAR-main/models/helpers.py
code served (permissive licence) · get_code("e543491ea064c733")
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
import torch.nn.functional as F

torch.manual_seed(0)

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

# Create an RNG generator
rng = torch.Generator()
rng.manual_seed(42)

ARGS = {
    "logits": logits,
    "tau": 1.0,
    "hard": False,
    "eps": 1e-10,
    "dim": -1,
    "rng": rng
}
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
nonlinearity Ran
ran; our contract draft was wrong, not the code
WD7ang/VAR-Scaling/VAR-main/models/basic_vae.py
code served (permissive licence) · get_code("3137073275f8c21a")
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) 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 single parameter 'x' which is used with:
# - multiplication operator (*)
# - torch.sigmoid() function
# This means x should be a tensor

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 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 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
normalize_01_into_pm1 Ran
ran; our contract draft was wrong, not the code
WD7ang/VAR-Scaling/VAR-main/utils/data.py
code served (permissive licence) · get_code("eabecb1bb23906fe")
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 normalize_01_into_pm1 takes a single parameter 'x'
# and performs: x.add(x).add_(-1)
# This means x must be a tensor with .add() method
# The function doubles x and subtracts 1, converting [0,1] range to [-1,1]

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
sample_with_top_k_top_p_ Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
WD7ang/VAR-Scaling/VAR-main/models/helpers.py
code served (permissive licence) · get_code("80371584448f2f6e")
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 int, shape (2, 3, 1) 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 logits tensor with shape (B, l, V) = (batch_size, sequence_length, vocab_size)
# Using distinct dimensions: B=2, l=3, V=4
logits_BlV = torch.randn(2, 3, 4)

# top_k parameter: filter to keep only top k logits
top_k = 2

# top_p parameter: nucleus sampling threshold
top_p = 0.9

# rng parameter: random number generator (None or torch.Generator)
rng = None

# num_samples parameter: number of samples to draw per position
num_samples = 1

ARGS = {
    "logits_BlV": logits_BlV,
    "top_k": top_k,
    "top_p": top_p,
    "rng": rng,
    "num_samples": num_samples,
}
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 int, shape (2, 3, 1) — shape (2, 3, 1) 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
allreduce Not yet run
unverified (harvested, no recorded run)
WD7ang/VAR-Scaling/VAR-main/dist.py
code served (permissive licence) · get_code("ff4c548fd07140d0")
How it ran

no invocation recorded.

build_dataset Not yet run
unverified (harvested, no recorded run)
WD7ang/VAR-Scaling/VAR-main/utils/data.py
code served (permissive licence) · get_code("8ff7703f731423de")
How it ran

no invocation recorded.

create_npz_from_sample_folder Not yet run
unverified (harvested, no recorded run)
WD7ang/VAR-Scaling/VAR-main/create_ngz.py
code served (permissive licence) · get_code("280dae1512c7b50f")
How it ran

no invocation recorded.

make_attn Not yet run
unverified (harvested, no recorded run)
WD7ang/VAR-Scaling/VAR-main/models/basic_vae.py
code served (permissive licence) · get_code("793c7e2489dd8dfc")
How it ran

no invocation recorded.

new_group Not yet run
unverified (harvested, no recorded run)
WD7ang/VAR-Scaling/VAR-main/dist.py
code served (permissive licence) · get_code("826ea3fc406a6a09")
How it ran

no invocation recorded.

pil_loader Not yet run
unverified (harvested, no recorded run)
WD7ang/VAR-Scaling/VAR-main/utils/data.py
code served (permissive licence) · get_code("b1714d4e3b7ac3b9")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 12 are also present in other papers whose code we hold (12 with at least one paper dated earlier, 3 the same month, 5 later) and 1 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
nonlinearity 156 other papers
149 dated earlier, 1 the same month, 6 later · 25 only as an assembly brick
2104.07652 (2021-04)
Geometry-Free View Synthesis: Transformers and no 3D Priors
2106.00132 (2021-06)
On Fast Sampling of Diffusion Probabilistic Models
2106.06819 (2021-06)
D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
and 153 more not named here
drop_path 66 other papers
60 dated earlier, 1 the same month, 3 later, 2 undated · 15 only as an assembly brick
1906.01502 (2019-06)
How multilingual is Multilingual BERT?
2011.10566 (2020-11)
Exploring Simple Siamese Representation Learning
2103.03230 (2021-03)
Barlow Twins: Self-Supervised Learning via Redundancy Reduction
and 63 more not named here
Normalize 54 other papers
52 dated earlier, 1 the same month, 1 later · 3 only as an assembly brick
2204.06125 (2022-04)
Hierarchical Text-Conditional Image Generation with CLIP Latents
2208.13753 (2022-08)
Frido: Feature Pyramid Diffusion for Complex Scene Image Synthesis
2211.13221 (2022-11)
Latent Video Diffusion Models for High-Fidelity Long Video Generation
and 51 more not named here
allgather 8 other papers
8 dated earlier
2404.02905 (2024-04)
Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
2406.09750 (2024-06)
ControlVAR: Exploring Controllable Visual Autoregressive Modeling
2408.01181 (2024-08)
VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling
and 5 more not named here
allreduce 8 other papers
8 dated earlier
2404.02905 (2024-04)
Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
2406.09750 (2024-06)
ControlVAR: Exploring Controllable Visual Autoregressive Modeling
2408.01181 (2024-08)
VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling
and 5 more not named here
gumbel_softmax_with_rng 8 other papers
7 dated earlier, 1 later
2408.01181 (2024-08)
VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling
2410.10812 (2024-10)
HART: Efficient Visual Generation with Hybrid Autoregressive Transformer
2411.10433 (2024-11)
M-VAR: Decoupled Scale-wise Autoregressive Modeling for High-Quality Image Generation
and 5 more not named here
make_attn 8 other papers
8 dated earlier
2404.02905 (2024-04)
Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
2408.01181 (2024-08)
VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling
2410.10812 (2024-10)
HART: Efficient Visual Generation with Hybrid Autoregressive Transformer
and 5 more not named here
new_group 8 other papers
8 dated earlier
2404.02905 (2024-04)
Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
2406.09750 (2024-06)
ControlVAR: Exploring Controllable Visual Autoregressive Modeling
2408.01181 (2024-08)
VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling
and 5 more not named here
sample_with_top_k_top_p_ 8 other papers
7 dated earlier, 1 later
2408.01181 (2024-08)
VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling
2410.10812 (2024-10)
HART: Efficient Visual Generation with Hybrid Autoregressive Transformer
2411.10433 (2024-11)
M-VAR: Decoupled Scale-wise Autoregressive Modeling for High-Quality Image Generation
and 5 more not named here
normalize_01_into_pm1 4 other papers
4 dated earlier
2411.10433 (2024-11)
M-VAR: Decoupled Scale-wise Autoregressive Modeling for High-Quality Image Generation
2412.04431 (2024-12)
Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis
2507.11441 (2025-07)
Implementing Adaptations for Vision AutoRegressive Model
and 1 more not named here
build_dataset 2 other papers
2 dated earlier
2411.10433 (2024-11)
M-VAR: Decoupled Scale-wise Autoregressive Modeling for High-Quality Image Generation
2412.04431 (2024-12)
Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis
pil_loader 2 other papers
2 dated earlier
2411.10433 (2024-11)
M-VAR: Decoupled Scale-wise Autoregressive Modeling for High-Quality Image Generation
2412.04431 (2024-12)
Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis

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

While inference-time scaling has significantly enhanced generative quality in large language and diffusion models, its application to vector-quantized (VQ) visual autoregressive modeling (VAR) remains unexplored. We introduce VAR-Scaling, the first general framework for inference-time scaling in VAR, addressing the critical challenge of discrete latent spaces that prohibit continuous path search. We find that VAR scales exhibit two distinct pattern types: general patterns and specific patterns, where later-stage specific patterns conditionally optimize early-stage general patterns. To overcome the discrete latent space barrier in VQ models, we map sampling spaces to quasi-continuous feature spaces via kernel density estimation (KDE), where high-density samples approximate stable, high-quality solutions. This transformation enables effective navigation of sampling distributions. We propose a densityadaptive hybrid sampling strategy: Top-k sampling focuses on highdensity regions to preserve quality near distribution modes, while Randomk sampling explores low-density areas to maintain diversity and prevent premature convergence. Consequently, VAR-Scaling optimizes sample fidelity at critical scales to enhance output quality. Experiments in class-conditional and text-to-image evaluations demonstrate significant improvements in inference process. The code is available at https: //github.com/WD7ang/VAR-Scaling.

For agents

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

get_harvested_code_for_paper("2601.07293")
get_code_for_paper("2601.07293")
have("2601.07293")

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

curl https://syntology.ai/api/ran/2601.07293.json

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

[![Syntology run record](https://syntology.ai/api/ran/2601.07293.svg)](https://syntology.ai/paper/2601.07293)

Connect an agent — have() is free.