SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.05027 · NeurIPS · 2024

Optimizing Automatic Differentiation with Deep Reinforcement Learning

Jamie Lohoff, Emre Neftci

arXiv · PDF · Open in the Atlas

Code that ran

12 of 29 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; 7 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 4 failures of our instrument; 17 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/2406.05027.json · badge: /api/ran/2406.05027.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
jamielohoff/graphax canonical (classifier label)
classifier basis: census
11 ran · 12 no recorded run
jamielohoff/alphagrad community
Papers with Code archive
1 ran · 5 no recorded run
FunctionStatusWhere it lives
add_elemental_rule Ran
ran, honoured the contract we drafted
jamielohoff/graphax/src/graphax/primitives/math.py
code served (permissive licence) · get_code("1e3e284050f7aa84")
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

torch.manual_seed(0)

# The function add_elemental_rule takes two parameters x and y
# and returns a tuple (1.0, 1.0)
# The function body doesn't actually use x or y, just returns constants
# But we need to provide valid arguments anyway

# Create small tensors for x and y
x = torch.randn(2, 4)
y = torch.randn(2, 4)

ARGS = {
    'x': x,
    'y': y
}
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 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
eye_like Ran
ran; our contract draft was wrong, not the code
jamielohoff/graphax/src/graphax/sparse/utils.py
code served (permissive licence) · get_code("f493bdb09a28f07c")
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 ArrayImpl, ndim 4, dtype family float, shape (2, 3, 2, 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 jax.numpy as jnp
from functools import reduce

# Test case 1: shape with out_len where dimensions match
# shape = (2, 3, 2, 3) means out_shape = (2, 3) and primal_shape = (2, 3)
ARGS = {
    "shape": (2, 3, 2, 3),
    "out_len": 2,
}
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 ArrayImpl, ndim 4, dtype family float, shape (2, 3, 2, 3) — shape (2, 3, 2, 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
eye_like_copy Ran
ran; our contract draft was wrong, not the code
jamielohoff/graphax/src/graphax/sparse/utils.py
code served (permissive licence) · get_code("3643f85316c3c4ad")
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 ArrayImpl, ndim 4, dtype family float, shape (3, 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 jax.numpy as jnp
from functools import reduce

# Create a small identity matrix to be reused
iota = jnp.eye(5)

# Test case 1: shape where primal_shape == out_shape
shape_1 = (3, 4, 3, 4)
out_len_1 = 2

# Test case 2: shape where dimensions match partially (Kronecker delta case)
shape_2 = (2, 3, 2, 4)
out_len_2 = 2

ARGS = {
    "shape": shape_1,
    "out_len": out_len_1,
    "iota": iota,
}
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 ArrayImpl, ndim 4, dtype family float, shape (3, 4, 3, 4) — shape (3, 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
get_aval_shape Ran
ran, honoured the contract we drafted
jamielohoff/graphax/src/graphax/primitives/base.py
code served (permissive licence) · get_code("9ced58452ae48749")
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 numpy as np
import torch

torch.manual_seed(0)

# Test case 1: numpy array input
array_val = np.array([[1, 2, 3], [4, 5, 6]])

# Test case 2: non-numpy input (e.g., int, list, dict, etc.)
scalar_val = 42

ARGS = {
    "val": array_val
}
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 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
get_ndim Ran
ran, honoured the contract we drafted
jamielohoff/graphax/src/graphax/primitives/base.py
code served (permissive licence) · get_code("0c5adab6e8148d63")
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 jax
import jax.numpy as jnp
import numpy as np

torch.manual_seed(0)

# Test with different types that the function checks for
scalar_float = 3.14
scalar_int = 42
torch_tensor = torch.randn(2, 4)
numpy_array = np.random.randn(3, 5)
jax_array = jnp.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0], [7.0, 8.0]])

# The function accepts arr parameter and checks:
# - if it's a float, int, or TypedFloat -> return 0
# - else return arr.ndim

ARGS = {
    "arr": torch_tensor
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, 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
get_shape Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jamielohoff/alphagrad/src/alphagrad/vertexgame/core.py
code served (permissive licence) · get_code("27157b30c8cb1b83")
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 jax
import jax.numpy as jnp

# The function accesses edges.shape[2] and edges.shape[1]
# So edges must be a 3D array
# Shape structure: (batch_or_leading_dim, num_v + num_i + 1, num_v)
# Let's use:
# - shape[0] = 2 (batch dimension)
# - shape[1] = 8 (this is num_v + num_i + 1, so if num_v=3, num_i=4, then 3+4+1=8)
# - shape[2] = 3 (num_v)

edges = jnp.zeros((2, 8, 3))

ARGS = {
    "edges": edges
}
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 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
get_shape Ran
ran, honoured the contract we drafted
jamielohoff/graphax/src/graphax/primitives/base.py
code served (permissive licence) · get_code("7cd542d5c5d2721c")
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 jax
import jax.numpy as jnp
import numpy as np

def get_shape(arr):
    if isinstance(arr, (float, int, jax._src.literals.TypedFloat)):
        return ()
    else:
        return arr.shape

# Test case 1: scalar float
scalar_float = 3.14

# Test case 2: scalar int
scalar_int = 42

# Test case 3: JAX array
jax_array = jnp.array([[1.0, 2.0, 3.0, 4.0],
                        [5.0, 6.0, 7.0, 8.0]])

# Test case 4: numpy array
numpy_array = np.array([[1, 2, 3, 4],
                        [5, 6, 7, 8]])

ARGS = {
    "arr": jax_array
}
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, 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
iota_elemental_only Ran
ran, honoured the contract we drafted
jamielohoff/graphax/src/graphax/primitives/structural.py
code served (permissive licence) · get_code("13181dc7795c5de7")
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 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 torch.nn as nn

torch.manual_seed(0)

# Construct minimal valid arguments for iota_elemental_only
primal_out = torch.randn(2, 4)
primals = [torch.randn(2, 4), torch.randn(3, 5)]
params = {}

ARGS = {
    "primal_out": primal_out,
    "primals": primals,
    "params": params,
}
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 list, ndim 1, 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
sub_elemental_rule Ran
ran, honoured the contract we drafted
jamielohoff/graphax/src/graphax/primitives/math.py
code served (permissive licence) · get_code("c20b3a537fa9222a")
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

torch.manual_seed(0)

# The function sub_elemental_rule takes two parameters: x and y
# The function body simply returns a tuple (1.0, -1.0)
# No actual usage of x and y, so they can be any valid objects
# Using small tensors to match the requirement

x = torch.randn(2, 4)
y = torch.randn(2, 4)

ARGS = {
    "x": x,
    "y": y,
}
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 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
tree_allclose Ran
ran, violated the contract we drafted
jamielohoff/graphax/src/graphax/utils.py
code served (permissive licence) · get_code("3477609316330ba0")
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 ArrayImpl, 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 jax.numpy as jnp
import jax.tree_util as jtu

# Create two nested tree structures with JAX arrays that are close but not identical
tree1 = {
    'a': jnp.array([1.0, 2.0, 3.0, 4.0]),
    'b': {
        'c': jnp.array([[1.0, 2.0, 3.0],
                       [4.0, 5.0, 6.0]]),
        'd': jnp.array([7.0, 8.0])
    }
}

tree2 = {
    'a': jnp.array([1.00001, 2.00001, 3.00001, 4.00001]),
    'b': {
        'c': jnp.array([[1.00001, 2.00001, 3.00001],
                       [4.00001, 5.00001, 6.00001]]),
        'd': jnp.array([7.00001, 8.00001])
    }
}

ARGS = {
    'tree1': tree1,
    'tree2': tree2,
    'equal_nan': False
}
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 ArrayImpl, 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
with_type_promotion Ran
ran; our contract draft was wrong, not the code
jamielohoff/graphax/src/graphax/primitives/math.py
code served (permissive licence) · get_code("e86d867de0350e04")
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 jax
import jax.numpy as jnp
import jax.lax as lax
from typing import Callable

# Define a simple test function that takes operands and returns a tuple of arrays
def test_function(*operands, **params):
    """
    A simple function that takes operands (JAX arrays) and returns a tuple of arrays.
    This will be passed to with_type_promotion for type promotion.
    """
    # Return a tuple of results based on operands
    result1 = operands[0] + operands[1]
    result2 = operands[0] * operands[1]
    return (result1, result2)

# Create sample operands with different dtypes to test type promotion
operand1 = jnp.array([1, 2, 3, 4], dtype=jnp.float32)
operand2 = jnp.array([5, 6, 7, 8], dtype=jnp.int32)

# Define the ARGS dictionary
ARGS = {
    "fn": test_function
}
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 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
zeros_like Ran
ran, honoured the contract we drafted
jamielohoff/graphax/src/graphax/sparse/utils.py
code served (permissive licence) · get_code("7c57f2a75ae4fe9e")
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 ArrayImpl, ndim 4, dtype family float, shape (4, 3, 2, 5) 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.numpy as jnp
from jax._src.core import ShapedArray
from jax import ShapeDtypeStruct

# Create ShapedArray instances for invar and outvar
# ShapedArray requires an abstract value (aval) which has shape and dtype attributes
# We can use ShapeDtypeStruct or create a minimal class

class MinimalAval:
    def __init__(self, shape, dtype=jnp.float32):
        self.shape = shape
        self.dtype = dtype

class MinimalShapedArray:
    def __init__(self, shape, dtype=jnp.float32):
        self.aval = MinimalAval(shape, dtype)

# Create test ShapedArray instances with distinct sizes
invar = MinimalShapedArray((4, 3))  # 4x3 input shape
outvar = MinimalShapedArray((2, 5))  # 2x5 output shape

ARGS = {
    "invar": invar,
    "outvar": outvar,
}
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 ArrayImpl, ndim 4, dtype family float, shape (4, 3, 2, 5) — shape (4, 3, 2, 5) 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
cross_country Not yet run
unverified (harvested, no recorded run)
jamielohoff/alphagrad/src/alphagrad/vertexgame/core.py
code served (permissive licence) · get_code("ff09a535363f2b4c")
How it ran

no invocation recorded.

device_put_elemental_rule Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/primitives/structural.py
code served (permissive licence) · get_code("d0e6738929346739")
How it ran

no invocation recorded.

dot_general_elemental_only Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/primitives/linalg.py
code served (permissive licence) · get_code("58031c16445a9519")
How it ran

no invocation recorded.

dot_general_elemental_rule Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/primitives/linalg.py
code served (permissive licence) · get_code("f909c31e94402c11")
How it ran

no invocation recorded.

get_fmas_jacprod Not yet run
unverified (harvested, no recorded run)
jamielohoff/alphagrad/src/alphagrad/vertexgame/core.py
code served (permissive licence) · get_code("61c62ae5257fa109")
How it ran

no invocation recorded.

iota_elemental_rule Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/primitives/structural.py
code served (permissive licence) · get_code("78d7b183cd04201e")
How it ran

no invocation recorded.

jacve Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/core.py
code served (permissive licence) · get_code("c6f1447ed053ebb2")
How it ran

no invocation recorded.

measure Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/performance_measurements/perf.py
code served (permissive licence) · get_code("f55fecb3286bee3e")
How it ran

no invocation recorded.

measure_execution_time_with_jax Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/performance_measurements/perf.py
code served (permissive licence) · get_code("2f45e8735e37c6fa")
How it ran

no invocation recorded.

reduce_sum_elemental_rule Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/primitives/reductions.py
code served (permissive licence) · get_code("a33b526df34622a8")
How it ran

no invocation recorded.

select_elemental_only Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/primitives/reductions.py
code served (permissive licence) · get_code("e087ddcd2fd02c8f")
How it ran

no invocation recorded.

select_elemental_rule Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/primitives/reductions.py
code served (permissive licence) · get_code("f12c8af8592424e6")
How it ran

no invocation recorded.

sparsity_fmas_map Not yet run
unverified (harvested, no recorded run)
jamielohoff/alphagrad/src/alphagrad/vertexgame/core.py
code served (permissive licence) · get_code("f794181f1d75bbc9")
How it ran

no invocation recorded.

sparsity_where Not yet run
unverified (harvested, no recorded run)
jamielohoff/alphagrad/src/alphagrad/vertexgame/core.py
code served (permissive licence) · get_code("7eb5241a707e9579")
How it ran

no invocation recorded.

unload_post_transforms Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/core.py
code served (permissive licence) · get_code("45119ffdcd9245c8")
How it ran

no invocation recorded.

unload_pre_transforms Not yet run
unverified (harvested, no recorded run)
jamielohoff/graphax/src/graphax/core.py
code served (permissive licence) · get_code("76547d6365d31f62")
How it ran

no invocation recorded.

vertex_eliminate Not yet run
unverified (harvested, no recorded run)
jamielohoff/alphagrad/src/alphagrad/vertexgame/core.py
code served (permissive licence) · get_code("c1b89909991b710c")
How it ran

no invocation recorded.

Also present in other papers' code

Of 29 samples lifted from this paper's linked repositories, 29 carry a function identity and 0 do not. Of the 29 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 29 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

Computing Jacobians with automatic differentiation is ubiquitous in many scientific domains such as machine learning, computational fluid dynamics, robotics, and finance. Even small savings in the number of computations or memory usage in Jacobian computations can already incur massive savings in energy consumption and runtime. While there exist many methods that allow for such savings, they generally trade computational efficiency for approximations of the exact Jacobian. In this paper, we present a novel method to optimize the number of necessary multiplications for Jacobian computation by leveraging deep reinforcement learning (RL) and a concept called cross-country elimination while still computing the exact Jacobian. Cross-country elimination is a framework for automatic differentiation that phrases Jacobian accumulation as ordered elimination of all vertices on the computational graph where every elimination incurs a certain computational cost. We formulate the search for the optimal elimination order that minimizes the number of necessary multiplications as a single player game which is played by an RL agent. We demonstrate that this method achieves up to 33% improvements over state-of-the-art methods on several relevant tasks taken from diverse domains. Furthermore, we show that these theoretical gains translate into actual runtime improvements by providing a cross-country elimination interpreter in JAX that can efficiently execute the obtained elimination orders. 38th Conference on Neural Information Processing Systems (NeurIPS 2024).

For agents

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

get_harvested_code_for_paper("2406.05027")
get_code_for_paper("2406.05027")
have("2406.05027")

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

curl https://syntology.ai/api/ran/2406.05027.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.05027.svg)](https://syntology.ai/paper/2406.05027)

Connect an agent — have() is free.