SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.06525 · 2024

Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation

arXiv · PDF · Open in the Atlas

Code that ran

10 of 15 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; 8 failures of our instrument; 5 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/2406.06525.json · badge: /api/ran/2406.06525.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
foundationvision/llamagen canonical (classifier label)
classifier basis: census
10 ran · 5 no recorded run
FunctionStatusWhere it lives
build_imagenet Ran
ran; our contract draft was wrong, not the code
foundationvision/llamagen/dataset/imagenet.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("5cceaa9d0edbce06")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

build_imagenet_code Ran
ran; our contract draft was wrong, not the code
foundationvision/llamagen/dataset/imagenet.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("b6635606f35f6a1a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

build_t2i Ran
ran; our contract draft was wrong, not the code
foundationvision/llamagen/dataset/t2i.py
code served (permissive licence) · get_code("02cb09299c18ee5d")
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 Text2ImgDataset. 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 os
import json
import numpy as np
import torch
from torchvision import transforms
import tempfile
import shutil

torch.manual_seed(0)

# Create a temporary directory structure with minimal test data
temp_dir = tempfile.mkdtemp()
data_path = os.path.join(temp_dir, "data")
t5_feat_path = os.path.join(temp_dir, "t5_feats")
short_t5_feat_path = os.path.join(temp_dir, "short_t5_feats")

os.makedirs(data_path, exist_ok=True)
os.makedirs(t5_feat_path, exist_ok=True)
os.makedirs(short_t5_feat_path, exist_ok=True)

# Create a minimal JSONL file with image paths
jsonl_file = os.path.join(data_path, "test_data.jsonl")
code_dir_name = "code_001"
code_dir_path_full = os.path.join(t5_feat_path, code_dir_name)
os.makedirs(code_dir_path_full, exist_ok=True)
short_code_dir_path_full = os.path.join(short_t5_feat_path, code_dir_name)
os.makedirs(short_code_dir_path_full, exist_ok=True)

# Create dummy image file
from PIL import Image
dummy_img = Image.new('RGB', (256, 256), color='red')
img_file = os.path.join(temp_dir, "test_image.jpg")
dummy_img.save(img_file)

# Write JSONL data
with open(jsonl_file, 'w') as f:
    json.dump({"image_path": img_file}, f)
    f.write('\n')
    json.dump({"image_path": img_file}, f)
    f.write('\n')

# Create dummy t5 feature files
t5_feat_file_0 = os.path.join(code_dir_path_full, "0.npy")
t5_feat_file_1 = os.path.join(code_dir_path_full, "1.npy")
short_t5_feat_file_0 = os.path.join(short_code_dir_path_full, "0.npy")
short_t5_feat_file_1 = os.path.join(short_code_dir_path_full, "1.npy")

t5_feat_data = np.random.randn(1, 80, 2048).astype(np.float32)
np.save(t5_feat_file_0, t5_feat_data)
np.save(t5_feat_file_1, t5_feat_data)
np.save(short_t5_feat_file_0, t5_feat_data)
np.save(short_t5_feat_file_1, t5_feat_data)

# Create args class
class Args:
    def __init__(self):
        self.data_path = data_path
        self.t5_feat_path = t5_feat_path
        self.short_t5_feat_path = short_t5_feat_path
        self.image_size = 256
        self.downsample_size = 8

args = Args()

# Create a simple transform
transform = transforms.Compose([
    transforms.Resize((256, 256)),
    transforms.ToTensor(),
])

ARGS = {
    "args": args,
    "transform": transform,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Text2ImgDataset — a Text2ImgDataset 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
build_t2i_code Ran
ran; our contract draft was wrong, not the code
foundationvision/llamagen/dataset/t2i.py
code served (permissive licence) · get_code("ff9ed4675cb307aa")
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 Text2ImgDatasetCode. 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
from torch.utils.data import Dataset

torch.manual_seed(0)

class Args:
    def __init__(self):
        self.batch_size = 2
        self.num_workers = 0
        self.dataset_path = None
        self.text_encoder = None
        self.image_encoder = None
        self.vocab_size = 100
        self.max_length = 16
        self.image_size = 32
        self.num_classes = 10
        self.shuffle = True
        self.pin_memory = False

ARGS = {
    "args": 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 Text2ImgDatasetCode — a Text2ImgDatasetCode output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Text2ImgDatasetCode 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
build_t2i_image Ran
ran; our contract draft was wrong, not the code
foundationvision/llamagen/dataset/t2i.py
code served (permissive licence) · get_code("090d7b6b1260114e")
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 Text2ImgDatasetImg. 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 os
import json
import tempfile
from PIL import Image
import torch
from torchvision import transforms

torch.manual_seed(0)

# Create temporary directories with test data
temp_dir = tempfile.mkdtemp()
lst_dir = os.path.join(temp_dir, 'lists')
face_lst_dir = os.path.join(temp_dir, 'face_lists')
img_dir = os.path.join(temp_dir, 'images')

os.makedirs(lst_dir)
os.makedirs(face_lst_dir)
os.makedirs(img_dir)

# Create dummy images
for i in range(2):
    img = Image.new('RGB', (4, 3), color=(i*50, i*100, i*150))
    img_path = os.path.join(img_dir, f'image_{i}.jpg')
    img.save(img_path)

# Create JSONL files with image paths
lst_file = os.path.join(lst_dir, 'dataset.jsonl')
with open(lst_file, 'w') as f:
    for i in range(2):
        img_path = os.path.join(img_dir, f'image_{i}.jpg')
        data = {'image_path': img_path}
        f.write(json.dumps(data) + '\n')

# Create face JSONL file
face_lst_file = os.path.join(face_lst_dir, 'faces_face.jsonl')
with open(face_lst_file, 'w') as f:
    for i in range(2):
        img_path = os.path.join(img_dir, f'image_{i}.jpg')
        data = {'image_path': img_path}
        f.write(json.dumps(data) + '\n')

# Create a simple transform
transform = transforms.Compose([
    transforms.Resize((4, 3)),
    transforms.ToTensor()
])

# Create args object with required attributes
class Args:
    def __init__(self, data_path, data_face_path):
        self.data_path = data_path
        self.data_face_path = data_face_path

args = Args(lst_dir, face_lst_dir)

ARGS = {
    'args': args,
    'transform': transform
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Text2ImgDatasetImg — a Text2ImgDatasetImg 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
center_crop_arr Ran
ran; our contract draft was wrong, not the code
foundationvision/llamagen/dataset/augmentation.py
code served (permissive licence) · get_code("1712a07966b542ee")
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 Image. 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 numpy as np
from PIL import Image

# Create a small test image
width, height = 256, 256
pil_image = Image.new('RGB', (width, height), color='red')

# Define arguments
ARGS = {
    'pil_image': pil_image,
    'image_size': 64
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
find_multiple Ran
ran, honoured the contract we drafted
foundationvision/llamagen/autoregressive/models/gpt.py
code served (permissive licence) · get_code("f6ff7671338c9c92")
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 int, ndim 0, dtype family int, 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)

# Analyze the function: find_multiple(n: int, k: int)
# The function takes two integers and returns an integer
# It checks if n is divisible by k, and if not, rounds n up to the next multiple of k

ARGS = {
    "n": 10,
    "k": 3
}
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 int, ndim 0, dtype family int, 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
precompute_freqs_cis_2d Ran
ran, honoured the contract we drafted
foundationvision/llamagen/autoregressive/models/gpt.py
code served (permissive licence) · get_code("7f53b2cdcd16fad9")
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 (18, 4, 2) 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 = {
    "grid_size": 4,
    "n_elem": 8,
    "base": 10000,
    "cls_token_num": 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 Tensor, ndim 3, dtype family float, shape (18, 4, 2) — shape (18, 4, 2) 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 Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
foundationvision/llamagen/autoregressive/models/generate.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("de8908ac0f302419")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

top_k_top_p_filtering Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
foundationvision/llamagen/autoregressive/models/generate.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("f2fea48028c7ae80")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

build_coco Not yet run
unverified (harvested, no recorded run)
foundationvision/llamagen/dataset/coco.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("a5dc3fcd09ebc843")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

build_openimage Not yet run
unverified (harvested, no recorded run)
foundationvision/llamagen/dataset/openimage.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("0a65aaf586d4f90b")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

logits_to_probs Not yet run
unverified (harvested, no recorded run)
foundationvision/llamagen/autoregressive/models/generate.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("9dac59261197935f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

precompute_freqs_cis Not yet run
unverified (harvested, no recorded run)
foundationvision/llamagen/autoregressive/models/gpt.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("0f0ff4e443413018")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

random_crop_arr Not yet run
unverified (harvested, no recorded run)
foundationvision/llamagen/dataset/augmentation.py
pointer only (licence: MIT); the stored copy get_code serves, from ThisisBillhe/ZipAR, is pointer only (licence: NONE) · get_code("980fb099f82d8b84")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ThisisBillhe/ZipAR, and GitHub detected no licence in that repository 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.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 15 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 1 the same month, 15 later) and 0 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
center_crop_arr 37 other papers
7 dated earlier, 1 the same month, 29 later
2211.17091 (2022-11)
Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models
2212.09748 (2022-12)
Scalable Diffusion Models with Transformers
2311.09257 (2023-11)
UFOGen: You Forward Once Large Scale Text-to-Image Generation via Diffusion GANs
and 34 more not named here
find_multiple 13 other papers
12 later, 1 undated · 2 only as an assembly brick
2409.13689 (2024-09)
Temporally Aligned Audio for Video with Autoregression
2410.02705 (2024-10)
ControlAR: Controllable Image Generation with Autoregressive Models
2410.21264 (2024-10)
LARP: Tokenizing Videos with a Learned Autoregressive Generative Prior
and 10 more not named here
precompute_freqs_cis_2d 9 other papers
8 later, 1 undated · 2 only as an assembly brick
2409.13689 (2024-09)
Temporally Aligned Audio for Video with Autoregression
2410.02705 (2024-10)
ControlAR: Controllable Image Generation with Autoregressive Models
2411.15867 (2024-11)
PanoLlama: Generating Endless and Coherent Panoramas with Next-Token-Prediction LLMs
and 6 more not named here
top_k_top_p_filtering 8 other papers
1 dated earlier, 6 later, 1 undated
2106.13219 (2021-06)
Towards Understanding and Mitigating Social Biases in Language Models
2410.02705 (2024-10)
ControlAR: Controllable Image Generation with Autoregressive Models
2411.15867 (2024-11)
PanoLlama: Generating Endless and Coherent Panoramas with Next-Token-Prediction LLMs
and 5 more not named here
precompute_freqs_cis 7 other papers
6 later, 1 undated
2410.02705 (2024-10)
ControlAR: Controllable Image Generation with Autoregressive Models
2411.15867 (2024-11)
PanoLlama: Generating Endless and Coherent Panoramas with Next-Token-Prediction LLMs
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
and 4 more not named here
logits_to_probs 6 other papers
5 later, 1 undated
2410.02705 (2024-10)
ControlAR: Controllable Image Generation with Autoregressive Models
2411.15867 (2024-11)
PanoLlama: Generating Endless and Coherent Panoramas with Next-Token-Prediction LLMs
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
and 3 more not named here
build_imagenet 5 other papers
5 later
2410.02705 (2024-10)
ControlAR: Controllable Image Generation with Autoregressive Models
2412.03069 (2024-12)
TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
and 2 more not named here
sample 5 other papers
4 later, 1 undated
2411.15867 (2024-11)
PanoLlama: Generating Endless and Coherent Panoramas with Next-Token-Prediction LLMs
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
2505.14673 (2025-05)
Training-Free Watermarking for Autoregressive Image Generation
and 2 more not named here
build_openimage 4 other papers
4 later
2412.03069 (2024-12)
TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
2502.20313 (2025-02)
FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction
and 1 more not named here
random_crop_arr 4 other papers
4 later
2412.03069 (2024-12)
TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
2503.10568 (2025-03)
Autoregressive Image Generation with Randomized Parallel Decoding
and 1 more not named here
build_coco 3 other papers
3 later
2412.03069 (2024-12)
TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
2503.10568 (2025-03)
Autoregressive Image Generation with Randomized Parallel Decoding
build_imagenet_code 3 other papers
3 later
2412.04062 (2024-12)
ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
2505.14673 (2025-05)
Training-Free Watermarking for Autoregressive Image Generation
2505.23660 (2025-05)
D-AR: Diffusion via Autoregressive Models
build_t2i 1 other paper
1 later
2503.10568 (2025-03)
Autoregressive Image Generation with Randomized Parallel Decoding
build_t2i_code 1 other paper
1 later
2503.10568 (2025-03)
Autoregressive Image Generation with Randomized Parallel Decoding
build_t2i_image 1 other paper
1 later
2503.10568 (2025-03)
Autoregressive Image Generation with Randomized Parallel Decoding

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

We introduce LlamaGen, a new family of image generation models that apply original ``next-token prediction'' paradigm of large language models to visual generation domain. It is an affirmative answer to whether vanilla autoregressive models, e.g., Llama, without inductive biases on visual signals can achieve state-of-the-art image generation performance if scaling properly. We reexamine design spaces of image tokenizers, scalability properties of image generation models, and their training data quality. The outcome of this exploration consists of: (1) An image tokenizer with downsample ratio of 16, reconstruction quality of 0.94 rFID and codebook usage of 97% on ImageNet benchmark. (2) A series of class-conditional image generation models ranging from 111M to 3.1B parameters, achieving 2.18 FID on ImageNet 256x256 benchmarks, outperforming the popular diffusion models such as LDM, DiT. (3) A text-conditional image generation model with 775M parameters, from two-stage training on LAION-COCO and high aesthetics quality images, demonstrating competitive performance of visual quality and text alignment. (4) We verify the effectiveness of LLM serving frameworks in optimizing the inference speed of image generation models and achieve 326% - 414% speedup. We release all models and codes to facilitate open-source community of visual generation and multimodal foundation models.

For agents

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

get_harvested_code_for_paper("2406.06525")
get_code_for_paper("2406.06525")
have("2406.06525")

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

curl https://syntology.ai/api/ran/2406.06525.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.06525.svg)](https://syntology.ai/paper/2406.06525)

Connect an agent — have() is free.