SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.12733 · 2023

On the De-duplication of LAION-2B

arXiv · PDF · Open in the Atlas

Code that ran

5 of 5 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 5 have no contract verdict; 0 failures of our instrument; 0 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/2303.12733.json · badge: /api/ran/2303.12733.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
ryanwebster90/snip-dedup canonical (classifier label)
classifier basis not recorded on the link
5 ran · 0 no recorded run
FunctionStatusWhere it lives
abs_ind_to_feat_file Ran
ran on a synthesized input
ryanwebster90/snip-dedup/retrieve_dup_urls_demo.py
code served (permissive licence) · get_code("3084891315965afc")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct arguments for abs_ind_to_feat_file
# The function takes an absolute index and cumulative sizes, and optionally feature files
# It finds which "bucket" the absolute index falls into

# Create cumulative sizes array
# These represent cumulative counts, so should be strictly increasing
cum_sz = np.array([0, 10, 25, 40], dtype=np.int64)

# Create an absolute index that falls within the range
# Since cum_sz goes up to 40, let's use an index like 30
# argwhere(30 - cum_sz >= 0) will find indices where 30 >= cum_sz values
# This will match indices 0, 1, 2 (where cum_sz values are 0, 10, 25)
# The last_ind will be 2, ind_offset will be 25, local_ind will be 30 - 25 = 5
abs_ind = np.int64(30)

# Create feature files list - should have same length as cum_sz or at least
# have an entry at index last_ind (which will be 2)
feat_files = ["file_0.npy", "file_1.npy", "file_2.npy", "file_3.npy"]

ARGS = {
    "abs_ind": abs_ind,
    "cum_sz": cum_sz,
    "feat_files": feat_files
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple 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
compute_feats_for_chunk Ran
ran on a synthesized input
ryanwebster90/snip-dedup/snip_dedup/snip_compress.py
code served (permissive licence) · get_code("484eaf7129ca32c3")
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 ndarray, ndim 2, dtype family float, shape (10, 3) 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.nn as nn
import numpy as np

torch.manual_seed(0)

class SimpleNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

net = SimpleNet()
net.eval()

chunk = np.random.randn(10, 4).astype(np.float32)

ARGS = {
    "net": net,
    "chunk": chunk,
    "batch_size": 256
}
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 ndarray, ndim 2, dtype family float, shape (10, 3) — shape (10, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_cum_sz Ran
ran on a synthesized input
ryanwebster90/snip-dedup/retrieve_dup_urls_demo.py
code served (permissive licence) · get_code("9dfc01f8475874be")
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 ndarray, ndim 1, dtype family int, shape (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 numpy as np
import tempfile
import os

# Create temporary directory for test .npy files
tmpdir = tempfile.mkdtemp()

# Create small test numpy arrays and save them
arr1 = np.random.randn(5, 3).astype(np.float32)
arr2 = np.random.randn(7, 3).astype(np.float32)
arr3 = np.random.randn(2, 3).astype(np.float32)

feat_file_1 = os.path.join(tmpdir, 'feat1.npy')
feat_file_2 = os.path.join(tmpdir, 'feat2.npy')
feat_file_3 = os.path.join(tmpdir, 'feat3.npy')

np.save(feat_file_1, arr1)
np.save(feat_file_2, arr2)
np.save(feat_file_3, arr3)

feat_files = [feat_file_1, feat_file_2, feat_file_3]

ARGS = {
    'feat_files': feat_files
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (4,) — shape (4,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_emb Ran
ran on a synthesized input
ryanwebster90/snip-dedup/retrieve_dup_urls_demo.py
code served (permissive licence) · get_code("8e236fba907d13b4")
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 memmap. 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
import tempfile
import os

# Create a temporary .npy file with sample embedding data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "embeddings.npy")

# Create a small embedding matrix: shape (5, 4) - 5 embeddings of dimension 4
embedding_data = np.random.randn(5, 4).astype(np.float32)
np.save(temp_file, embedding_data)

# Define arguments for get_emb function
ARGS = {
    "ff": temp_file,
    "local_ind": 2  # Index into the first dimension of the array
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type memmap — a memmap output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a memmap output the recorded ndim 1 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
validate_parts Ran
ran on a synthesized input
ryanwebster90/snip-dedup/snip_dedup/_cli_helper.py
code served (permissive licence) · get_code("81481855587fca40")
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 tuple, ndim 1, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import sys

def validate_parts(parts):
    if type(parts) is int:
        sys.exit("Single value is not accepted for --parts as it's ambiguous between wanting only that exact part or that number of parts starting from 0. Please use a range instead like 0:2")
    parts_bounds = parts.split(':')
    try:
        parts_bounds = [int(part) for part in parts_bounds]
    except Exception:
        sys.exit(f'The parts pattern "{parts}" is not valid. It should be a valid range such as "0:2" or "14:42"')
    if len(parts_bounds) == 0:
        sys.exit('The --parts argument cannot be empty')
    elif len(parts_bounds) == 1:
        sys.exit("Single value is not accepted for --parts as it's ambiguous between wanting only that exact part or that number of parts starting from 0. Please use a range instead like 0:2")
    elif len(parts_bounds) > 2:
        sys.exit('Ranges with more than 2 parts, such as 0:2:14 are not valid for --parts. Please limit yourself with simple ranges such as 0:14')
    start_part, end_part = parts_bounds
    if start_part < 0 or end_part < 0:
        sys.exit('Only positive integers are allowed for --parts, such as 0:14')
    if end_part <= start_part:
        sys.exit('The --parts argument must be of the shape "s:e" with s < e, such as "0:1" or "14:42". The "e" bound is excluded.')
    return (start_part, end_part)

ARGS = {
    "parts": "0:10"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
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 tuple, ndim 1, dtype family int — a tuple 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

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 5 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.

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

Generative models, such as DALL-E, Midjourney, and Stable Diffusion, have societal implications that extend beyond the field of computer science. These models require large image databases like LAION-2B, which contain two billion images. At this scale, manual inspection is difficult and automated analysis is challenging. In addition, recent studies show that duplicated images pose copyright problems for models trained on LAION2B, which hinders its usability. This paper proposes an algorithmic chain that runs with modest compute, that compresses CLIP features to enable efficient duplicate detection, even for vast image volumes. Our approach demonstrates that roughly 700 million images, or about 30\%, of LAION-2B's images are likely duplicated. Our method also provides the histograms of duplication on this dataset, which we use to reveal more examples of verbatim copies by Stable Diffusion and further justify the approach. The current version of the de-duplicated set will be distributed online.

For agents

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

get_harvested_code_for_paper("2303.12733")
get_code_for_paper("2303.12733")
have("2303.12733")

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

curl https://syntology.ai/api/ran/2303.12733.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.12733.svg)](https://syntology.ai/paper/2303.12733)

Connect an agent — have() is free.