SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2501.03575 · 2025

Cosmos World Foundation Model Platform for Physical AI

arXiv · PDF · Open in the Atlas

Code that ran

12 of 14 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, 1 violated it and 10 have no contract verdict; 1 failure of our instrument; 2 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/2501.03575.json · badge: /api/ran/2501.03575.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
nvidia-cosmos/cosmos-predict1 canonical (classifier label)
classifier basis not recorded on the link
7 ran · 0 no recorded run
nvidia/cosmos-tokenizer canonical (classifier label)
classifier basis not recorded on the link
3 ran · 0 no recorded run
nvlabs/tokenbench canonical (classifier label)
classifier basis not recorded on the link
2 ran · 2 no recorded run
FunctionStatusWhere it lives
PSNR Ran
ran on a synthesized input
nvlabs/tokenbench/token_bench/metrics_cli.py
code served (permissive licence) · get_code("2ec238b8692bb23a")
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 float, ndim 0, dtype family float, 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 numpy as np

# Create two small images in the range [0, 255]
# Using distinct dimensions: 4 height, 3 width, 2 channels
np.random.seed(0)
image1 = np.random.randint(0, 256, size=(4, 3, 2), dtype=np.uint8).astype(np.float32)
image2 = np.random.randint(0, 256, size=(4, 3, 2), dtype=np.uint8).astype(np.float32)

ARGS = {
    "input0": image1,
    "input1": image2,
}
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 float, ndim 0, dtype family float, 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
batch2time Ran
ran on a synthesized input
nvidia/cosmos-tokenizer/cosmos_tokenizer/modules/utils.py
code served (permissive licence) · get_code("cc3e6d9b4eee82e0")
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 5, dtype family float, shape (2, 4, 3, 5, 6) 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
from einops import pack, rearrange, unpack

torch.manual_seed(0)

# The function takes x (a tensor) and batch_size (an int)
# It reshapes x from shape (b*t, c, h, w) to (b, c, t, h, w)
# So we need to create a tensor with shape (b*t, c, h, w)

batch_size = 2
time_steps = 3
channels = 4
height = 5
width = 6

# Create tensor with shape (b*t, c, h, w) = (6, 4, 5, 6)
x = torch.randn(batch_size * time_steps, channels, height, width)

ARGS = {
    'x': x,
    'batch_size': batch_size,
}
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 Tensor, ndim 5, dtype family float, shape (2, 4, 3, 5, 6) — shape (2, 4, 3, 5, 6) 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
compute_llama3_ffn_hidden_dim Ran
ran on a synthesized input
nvidia-cosmos/cosmos-predict1/cosmos_predict1/autoregressive/modules/mlp.py
code served (permissive licence) · get_code("826dcfd43fab7282")
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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "dim": 768,
    "multiple_of": 256,
    "ffn_dim_multiplier": 1.3
}
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 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
create_norm Ran
ran on a synthesized input
nvidia-cosmos/cosmos-predict1/cosmos_predict1/autoregressive/modules/normalization.py
code served (permissive licence) · get_code("4431273b1abad22f")
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 RMSNorm. 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 = {
    "norm_type": "rmsnorm",
    "dim": 64,
    "eps": 1e-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 RMSNorm — a RMSNorm output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a RMSNorm 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
get_1d_sincos_pos_embed_from_grid Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
nvidia-cosmos/cosmos-predict1/cosmos_predict1/autoregressive/modules/embedding.py
code served (permissive licence) · get_code("e5947aba1d10885f")
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 ndarray, ndim 2, dtype family float, shape (5, 8) 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

# Create a simple 1D position array
# embed_dim must be even (checked by assert in function)
# pos should be a 1D array of positions

embed_dim = 8  # Small even number for embedding dimension
pos = np.array([0.0, 1.0, 2.0, 3.0, 4.0])  # 5 positions

ARGS = {
    'embed_dim': embed_dim,
    'pos': pos
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
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 2, dtype family float, shape (5, 8) — shape (5, 8) 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_vit_config Ran
ran on a synthesized input
nvidia-cosmos/cosmos-predict1/cosmos_predict1/autoregressive/networks/vit.py
code served (permissive licence) · get_code("aa4f611e6a714f05")
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 dict. 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
from typing import Any, Mapping

torch.manual_seed(0)

# The function get_vit_config takes a single parameter: model_name (str)
# It returns a Mapping[str, Any] containing ViT configuration parameters
# The function accepts 'pixtral-12b-vit' as a valid model name

ARGS = {
    'model_name': 'pixtral-12b-vit'
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
precompute_freqs_cis_2d Ran
ran, violated the contract we drafted
nvidia-cosmos/cosmos-predict1/cosmos_predict1/autoregressive/networks/vit.py
code served (permissive licence) · get_code("439497162c72f1dc")
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 (4, 3, 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)

ARGS = {
    "dim": 8,
    "height": 4,
    "width": 3,
    "theta": 10000.0,
}
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 3, dtype family float, shape (4, 3, 4) — shape (4, 3, 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
preprocess_single Ran
ran on a synthesized input
nvlabs/tokenbench/token_bench/fvd.py
code served (permissive licence) · get_code("7b03dd8d6c1862b8")
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 4, dtype family float, shape (3, 5, 224, 224) 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 math
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small video tensor with shape (c, t, h, w)
# c=3 (channels), t=8 (time/frames), h=256 (height), w=320 (width)
video = torch.randn(3, 8, 256, 320)

ARGS = {
    "video": video,
    "resolution": 224,
    "sequence_length": 5
}
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 4, dtype family float, shape (3, 5, 224, 224) — shape (3, 5, 224, 224) 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
reshape_for_broadcast Ran
ran on a synthesized input
nvidia-cosmos/cosmos-predict1/cosmos_predict1/autoregressive/networks/vit.py
code served (permissive licence) · get_code("7176ff743ab3af05")
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 (1, 3, 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)

# Construct valid arguments for reshape_for_broadcast
# The function expects:
# - freqs_cis: shape (x.shape[1], x.shape[-1])
# - x: input tensor with ndim >= 2

# Create input tensor x with shape (batch=2, seq_len=3, dim=4)
x = torch.randn(2, 3, 4)

# Create freqs_cis with shape matching (x.shape[1], x.shape[-1]) = (3, 4)
freqs_cis = torch.randn(3, 4, dtype=torch.complex64)

ARGS = {
    "freqs_cis": freqs_cis,
    "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 3, dtype family float, shape (1, 3, 4) — shape (1, 3, 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
scaled_dot_product_attention Ran
ran on a synthesized input
nvidia-cosmos/cosmos-predict1/cosmos_predict1/autoregressive/modules/attention.py
code served (permissive licence) · get_code("abc31b2f673701c7")
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 4, dtype family float, shape (2, 3, 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 math

torch.manual_seed(0)

# Tiny tensors for attention computation
# Shape: (batch_size, seq_len, num_heads, head_dim)
batch_size = 2
seq_len = 3
num_heads = 2
head_dim = 4

q = torch.randn(batch_size, num_heads, seq_len, head_dim)
k = torch.randn(batch_size, num_heads, seq_len, head_dim)
v = torch.randn(batch_size, num_heads, seq_len, head_dim)

# Optional mask tensor (causal mask shape)
mask = torch.triu(torch.ones(seq_len, seq_len) * float('-inf'), diagonal=1)

ARGS = {
    "q": q,
    "k": k,
    "v": v,
    "head_dim": head_dim,
    "mask": None,
    "is_causal": True,
    "dropout_p": 0.0,
}
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 4, dtype family float, shape (2, 3, 2, 4) — shape (2, 3, 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
space2batch Ran
ran on a synthesized input
nvidia/cosmos-tokenizer/cosmos_tokenizer/modules/utils.py
code served (permissive licence) · get_code("b48fa21d32c1e846")
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. 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 einops import pack, rearrange, unpack

torch.manual_seed(0)

# Create a small 5D tensor: (batch, channels, time, height, width)
x = torch.randn(2, 4, 3, 5, 6)

ARGS = {
    "x": x
}
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 tuple — a tuple 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
time2batch Ran
ran on a synthesized input
nvidia/cosmos-tokenizer/cosmos_tokenizer/modules/utils.py
code served (permissive licence) · get_code("e9de9d559844e9ae")
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. 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 einops import pack, rearrange, unpack

torch.manual_seed(0)

# Create a small 5D tensor: (batch=2, channels=4, time=3, height=5, width=6)
x = torch.randn(2, 4, 3, 5, 6)

ARGS = {
    'x': x
}
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 tuple — a tuple 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
LPIPS Not yet run
unverified (harvested, no recorded run)
nvlabs/tokenbench/token_bench/metrics_cli.py
code served (permissive licence) · get_code("ef1f7e1ebf6386e6")
How it ran

no invocation recorded.

SSIM Not yet run
unverified (harvested, no recorded run)
nvlabs/tokenbench/token_bench/metrics_cli.py
code served (permissive licence) · get_code("4ea7aadfca86cd09")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 3 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 2 later) and 11 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_1d_sincos_pos_embed_from_grid 42 other papers
22 dated earlier, 20 later · 22 only as an assembly brick
2212.09748 (2022-12)
Scalable Diffusion Models with Transformers
2303.14389 (2023-03)
Masked Diffusion Transformer is a Strong Image Synthesizer
2305.13311 (2023-05)
VDT: General-purpose Video Diffusion Transformers via Mask Modeling
and 39 more not named here
precompute_freqs_cis_2d 1 other paper
1 dated earlier
2310.06825 (2023-10)
Mistral 7B
scaled_dot_product_attention 1 other paper
1 later
2503.03751 (2025-03)
GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control

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

Physical AI needs to be trained digitally first. It needs a digital twin of itself, the policy model, and a digital twin of the world, the world model. In this paper, we present the Cosmos World Foundation Model Platform to help developers build customized world models for their Physical AI setups. We position a world foundation model as a general-purpose world model that can be fine-tuned into customized world models for downstream applications. Our platform covers a video curation pipeline, pre-trained world foundation models, examples of post-training of pre-trained world foundation models, and video tokenizers. To help Physical AI builders solve the most critical problems of our society, we make Cosmos open-source and our models open-weight with permissive licenses available via https://github.com/nvidia-cosmos/cosmos-predict1.

For agents

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

get_harvested_code_for_paper("2501.03575")
get_code_for_paper("2501.03575")
have("2501.03575")

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

curl https://syntology.ai/api/ran/2501.03575.json

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

[![Syntology run record](https://syntology.ai/api/ran/2501.03575.svg)](https://syntology.ai/paper/2501.03575)

Connect an agent — have() is free.