SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2309.03060 · NeurIPS · 2023

CoLA: Exploiting Compositional Structure for Automatic and Efficient Numerical Linear Algebra

Geoff Pleiss, Marc Finzi, Andrew Wilson, Andres Potapczynski

arXiv · PDF · Open in the Atlas

Code that ran

17 of 17 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; 1 honoured the contract we drafted, 4 violated it and 0 have no contract verdict; 12 failures of our instrument; 0 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/2309.03060.json · badge: /api/ran/2309.03060.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
wilson-labs/cola canonical (classifier label)
classifier basis: census
17 ran · 0 no recorded run
FunctionStatusWhere it lives
build_tqdm Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/utils/jax_tqdm.py
code served (permissive licence) · get_code("1222b814438ac04e")
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 typing
import jax
from jax.debug import callback
from jax.experimental import io_callback
from tqdm.auto import tqdm

# The function build_tqdm takes two parameters:
# 1. n: int - the number of iterations
# 2. message: typing.Optional[str] - an optional message for the progress bar

# Reading the function body:
# - n is used to create range(n), check iteration numbers against n, and format a string
# - message is used as a description for the tqdm bar (optional)

# Build arguments
ARGS = {
    "n": 100,
    "message": "Processing batches"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, tqdm — 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
check_valid_dtype Ran
ran, violated the contract we drafted
wilson-labs/cola/cola/backends/backends.py
code served (permissive licence) · get_code("1505bc3ca64e8748")
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 bool, 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 torch
import numpy as np

torch.manual_seed(0)

def dummy_alloc_fn(array, dtype):
    """A simple allocation function that converts array to the given dtype."""
    if isinstance(array, torch.Tensor):
        return array.to(dtype)
    elif isinstance(array, np.ndarray):
        return array.astype(dtype)
    else:
        raise TypeError(f"Unsupported array type: {type(array)}")

# Create a small test array
test_array = torch.randn(2, 4)

# Valid dtype for the test
test_dtype = torch.float32

ARGS = {
    "array": test_array,
    "dtype": test_dtype,
    "alloc_fn": dummy_alloc_fn
}
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 bool, 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
combine Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
wilson-labs/cola/cola/utils/custom_autodiff.py
code served (permissive licence) · get_code("e09ed6dad6d629b9")
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 list, ndim 3, 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 torch

torch.manual_seed(0)

# Create small torch tensors for torch_tensors iterable
torch_tensors = [
    torch.randn(2, 4),
    torch.randn(2, 4),
    torch.randn(2, 4)
]

# Create small non-torch tensors (using numpy or lists)
import numpy as np
non_torch_tensors = [
    np.array([[1, 2, 3, 4], [5, 6, 7, 8]]),
    np.array([[9, 10, 11, 12], [13, 14, 15, 16]]),
    np.array([[17, 18, 19, 20], [21, 22, 23, 24]])
]

# Create is_torch boolean selector list
is_torch = [True, False, True]

ARGS = {
    "torch_tensors": torch_tensors,
    "non_torch_tensors": non_torch_tensors,
    "is_torch": is_torch
}
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 list, ndim 3, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
eig Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/backends/np_fns.py
code served (permissive licence) · get_code("194698227eb24ab9")
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 numpy as np

# Create a small square matrix for eigenvalue decomposition
# Using distinct dimensions where appropriate
test_matrix = np.array([
    [4.0, 2.0],
    [1.0, 3.0]
], dtype=np.float64)

ARGS = {
    "array": test_matrix
}
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 — 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
is_xnp_array Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/ops/operator_base.py
code served (permissive licence) · get_code("79232c17bd3e80b9")
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 bool, 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 torch
import numpy as np

torch.manual_seed(0)

# Create a mock xnp object that behaves like numpy or torch
class MockXnp:
    def is_array(self, obj):
        # Return True for numpy arrays and torch tensors
        return isinstance(obj, (np.ndarray, torch.Tensor))

# Create test objects
numpy_array = np.array([1, 2, 3, 4], dtype=np.float32)
torch_tensor = torch.tensor([1.0, 2.0, 3.0, 4.0])
xnp_mock = MockXnp()

ARGS = {
    'obj': numpy_array,
    'xnp': xnp_mock
}
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 bool, 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
iterative_autograd Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/utils/custom_autodiff.py
code served (permissive licence) · get_code("07094ab2794b77be")
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 function. 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)

# Create a mock iterative_bwd function that serves as the VJP for the iterative solver
def mock_iterative_bwd(res, d_outputs, unflatten, *args, **kwargs):
    """
    Mock backward function for iterative solver.
    Takes (res, d_outputs, unflatten, *args, **kwargs) and returns gradient w.r.t. A
    """
    params, output = res
    # Return a gradient tensor with same shape as params
    dA = torch.ones_like(params)
    return (dA,)

# Create a mock iterative function decorated with iterative_autograd
def mock_iterative_fwd(A, *args, **kwargs):
    """
    Mock forward function that uses A as first argument.
    A is expected to be a tensor-like object.
    """
    # Do some computation with A
    result = A @ torch.ones(A.shape[-1], 4, dtype=A.dtype)
    return result

# Create a wrapper class that mimics torch tensor behavior with flatten/unflatten
class MockTensorWrapper:
    def __init__(self, tensor):
        self.tensor = tensor
        self.xnp = torch
    
    def flatten(self):
        """Returns (flattened_param, unflatten_fn)"""
        flat = self.tensor.flatten()
        
        def unflatten(params):
            return params.reshape(self.tensor.shape)
        
        return (flat, unflatten)
    
    def __getattr__(self, name):
        return getattr(self.tensor, name)

# Create mock A tensor (4x3 matrix for distinct dimensions)
A_tensor = torch.randn(4, 3, requires_grad=True)
A = MockTensorWrapper(A_tensor)

# Additional arguments for the iterative solver
extra_args = (torch.randn(3, 2),)
extra_kwargs = {"scale": 2.0}

ARGS = {
    "iterative_bwd": mock_iterative_bwd,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
loop_tqdm Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/utils/jax_tqdm.py
code served (permissive licence) · get_code("d551c022a49dc59a")
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 function. 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 typing
import jax
import jax.lax

# Define ARGS for loop_tqdm function
# loop_tqdm takes:
# - n: int (number of iterations)
# - message: typing.Optional[str] (optional message for progress bar)

ARGS = {
    "n": 10,
    "message": "Testing loop_tqdm"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
lstsq Ran
ran, violated the contract we drafted
wilson-labs/cola/cola/backends/np_fns.py
code served (permissive licence) · get_code("5dc26cca6fae567f")
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 float, shape (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 numpy as np

# Create a simple least squares problem
# A should be a 2D array (m x n) where m >= n
# b should be a 1D or 2D array (m,) or (m, k)

# Using distinct sizes: 5 rows, 3 columns for A, 5 rows for b
A = np.array([
    [1.0, 2.0, 3.0],
    [2.0, 3.0, 4.0],
    [3.0, 4.0, 5.0],
    [4.0, 5.0, 6.0],
    [5.0, 6.0, 7.0]
], dtype=np.float64)

b = np.array([1.0, 2.0, 3.0, 4.0, 5.0], dtype=np.float64)

ARGS = {
    "A": A,
    "b": b
}
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 float, shape (3,) — shape (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
maybe_get_dtype Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/ops/operator_base.py
code served (permissive licence) · get_code("12064a51bf1dd610")
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 dtype. 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

torch.manual_seed(0)

# The function tries to access obj.dtype and returns it, or returns None if AttributeError
# We need to test with an object that has dtype (like a tensor) and one that doesn't

# Create a tensor with dtype attribute
tensor_obj = torch.randn(2, 4)

# Create a stand-in object without dtype attribute
class NoTypeObj:
    pass

no_type_obj = NoTypeObj()

# ARGS should map the parameter 'obj' to a value
# We'll use the tensor since it has dtype
ARGS = {
    "obj": tensor_obj
}
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 dtype — a dtype output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dtype 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
product Ran
ran, violated the contract we drafted
wilson-labs/cola/cola/ops/operators.py
code served (permissive licence) · get_code("18ed40676168eff9")
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
from functools import partial, reduce

torch.manual_seed(0)

# The function `product(c)` takes a single parameter `c`
# It uses reduce with a lambda that multiplies elements together
# So `c` should be an iterable of numbers that can be multiplied

# Test with a simple list of integers
c = [2, 3, 4, 5]

ARGS = {
    "c": c
}
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
scan_tqdm Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/utils/jax_tqdm.py
code served (permissive licence) · get_code("06b12e67297b8604")
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 function. 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 typing
import jax
import jax.numpy as jnp

# Setup for scan_tqdm function
# The function takes n (int) and message (Optional[str])

ARGS = {
    "n": 10,
    "message": "Test scan progress"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
sparse_csr Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/backends/jax_fns.py
code served (permissive licence) · get_code("c56a560ef1f1c286")
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 BCSR. 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 jax
import jax.numpy as jnp
import numpy as np
from jax.experimental.sparse import BCSR

# Create a small sparse matrix in CSR format
# Matrix shape: 3x4
# We'll create a simple sparse matrix with a few non-zero elements

# Row pointers: indicate where each row starts in the col_indices and data arrays
# For a 3x4 matrix with 3 non-zero elements
row_pointers = jnp.array([0, 1, 2, 3])

# Column indices: which column each non-zero element is in
col_indices = jnp.array([0, 2, 3])

# Data: the actual non-zero values
data = jnp.array([1.0, 2.0, 3.0])

# Shape of the sparse matrix
shape = (3, 4)

ARGS = {
    "row_pointers": row_pointers,
    "col_indices": col_indices,
    "data": data,
    "shape": shape,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type BCSR — a BCSR 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
sparse_csr Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/backends/torch_fns.py
code served (permissive licence) · get_code("624ef6c3eeae7741")
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. 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)

# For a sparse CSR tensor, we need:
# - row_pointers (crow_indices): shape [m+1] where m is number of rows
# - col_indices: shape [nnz] where nnz is number of non-zero elements
# - data (values): shape [nnz] with the non-zero values
# - shape: tuple (m, n) for the matrix dimensions

# Let's create a small 3x4 sparse matrix with 4 non-zero elements
m, n = 3, 4
nnz = 4

# row_pointers: indicates where each row starts in col_indices and data
# For 3 rows with 4 total non-zeros, a valid example is [0, 2, 3, 4]
row_pointers = torch.tensor([0, 2, 3, 4], dtype=torch.int64)

# col_indices: column index for each non-zero element
col_indices = torch.tensor([0, 2, 1, 3], dtype=torch.int64)

# data: the actual values
data = torch.tensor([1.0, 2.0, 3.0, 4.0], dtype=torch.float32)

# shape: dimensions of the matrix
shape = (m, n)

ARGS = {
    "row_pointers": row_pointers,
    "col_indices": col_indices,
    "data": data,
    "shape": shape,
}
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, shape (3, 4) — shape (3, 4) as a later run recorded it
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
svd Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
wilson-labs/cola/cola/backends/jax_fns.py
code served (permissive licence) · get_code("a9583780ceda1e46")
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 jax.numpy as jnp
import numpy as np

# Create a small matrix for SVD decomposition
# Using distinct dimensions (4x3) for clarity
A = jnp.array([
    [1.0, 2.0, 3.0],
    [4.0, 5.0, 6.0],
    [7.0, 8.0, 9.0],
    [10.0, 11.0, 12.0]
])

ARGS = {
    "A": A,
    "full_matrices": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — 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
to_np Ran
ran, violated the contract we drafted
wilson-labs/cola/cola/backends/jax_fns.py
code served (permissive licence) · get_code("0381ba678b73491b")
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 float, 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 jax
from jax import numpy as jnp

# Create a simple JAX array to pass to to_np
test_array = jnp.array([1.0, 2.0, 3.0, 4.0], dtype=jnp.float32)

# Define arguments for to_np function
ARGS = {
    'array': test_array
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — 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 float, 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
while_loop Ran
ran, honoured the contract we drafted
wilson-labs/cola/cola/utils/torch_tqdm.py
code served (permissive licence) · get_code("200e115745cfc25b")
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)

# cond_fun: takes a value and returns a boolean
# body_fun: takes a value and returns a new value
# init_val: initial value to start with

# Simple example: count up from 0 to 5
def cond_fun(val):
    """Continue while val < 5"""
    return val < 5

def body_fun(val):
    """Increment val by 1"""
    return val + 1

init_val = 0

ARGS = {
    "cond_fun": cond_fun,
    "body_fun": body_fun,
    "init_val": init_val,
}
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
while_loop_winfo Ran
ran; our contract draft was wrong, not the code
wilson-labs/cola/cola/utils/torch_tqdm.py
code served (permissive licence) · get_code("f71a8d09e632c999")
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 torch
import numpy as np
from torch import nn

torch.manual_seed(0)

# Define a simple iterative state that will converge
class SimpleState:
    def __init__(self, value):
        self.value = value
    
    def __repr__(self):
        return f"SimpleState({self.value})"

# Define the error function - should decrease over iterations
def errorfn(state):
    return float(state.value)

# Define condition function - continue while error > tolerance
def cond_fun(state):
    return state.value > 1e-4

# Define body function - decreases the state value
def body_fun(state):
    state.value *= 0.8  # Reduce by 20% each iteration
    return state

# Initial state with high error
init_val = SimpleState(1.0)

# Tolerance level
tol = 1e-4

# Maximum iterations
max_iters = 50

# Every n iterations update progress
every = 1

# Description for progress bar
desc = 'Test while loop'

# Whether to show progress bar
pbar = False

# Additional kwargs
kwargs = {}

ARGS = {
    'errorfn': errorfn,
    'tol': tol,
    'max_iters': max_iters,
    'every': every,
    'desc': desc,
    'pbar': pbar,
    'kwargs': kwargs
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, tqdm — 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

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 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 17 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

Many areas of machine learning and science involve large linear algebra problems, such as eigendecompositions, solving linear systems, computing matrix exponentials, and trace estimation. The matrices involved often have Kronecker, convolutional, block diagonal, sum, or product structure. In this paper, we propose a simple but general framework for large-scale linear algebra problems in machine learning, named CoLA (Compositional Linear Algebra). By combining a linear operator abstraction with compositional dispatch rules, CoLA automatically constructs memory and runtime efficient numerical algorithms. Moreover, CoLA provides memory efficient automatic differentiation, low precision computation, and GPU acceleration in both JAX and PyTorch, while also accommodating new objects, operations, and rules in downstream packages via multiple dispatch. CoLA can accelerate many algebraic operations, while making it easy to prototype matrix structures and algorithms, providing an appealing drop-in tool for virtually any computational effort that requires linear algebra. We showcase its efficacy across a broad range of applications, including partial differential equations, Gaussian processes, equivariant model construction, and unsupervised learning.

For agents

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

get_harvested_code_for_paper("2309.03060")
get_code_for_paper("2309.03060")
have("2309.03060")

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

curl https://syntology.ai/api/ran/2309.03060.json

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

[![Syntology run record](https://syntology.ai/api/ran/2309.03060.svg)](https://syntology.ai/paper/2309.03060)

Connect an agent — have() is free.