SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2106.02584 · NeurIPS · 2021

Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning

Clare Lyle, Yarin Gal, Tom Rainforth, Jannik Kossen, Neil Band, Aidan Gomez

arXiv · PDF · Open in the Atlas

Code that ran

22 of 30 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (4 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 1 violated it and 16 have no contract verdict; 3 failures of our instrument; 8 no recorded run); 1 of them was also run on the host, not in the jail, by the peer-agreement lane that set its level, 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/2106.02584.json · badge: /api/ran/2106.02584.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
oatml/non-parametric-transformers canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
oatml-markslab/proteinnpt community (classifier label)
classifier basis not recorded on the link
16 ran · 3 no recorded run
OATML/Non-Parametric-Transformers marked official in Papers with Code
Papers with Code archive
5 ran · 5 no recorded run
FunctionStatusWhere it lives
count_parameters Ran
ran, honoured the contract we drafted
oatml/non-parametric-transformers/npt/utils/train_utils.py
code served (permissive licence) · get_code("0e18add484f1a90c")
How it ran

Ran in Syntology's jail on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. The comparison behind this sample's level is a different run: the peer-agreement lane ran it again on the host, not in the jail (report dated 2026-09-04T22:32:05Z). 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)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

model = SimpleModel()

ARGS = {
    "model": model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
Args Ran
ran on a synthesized input · constructed an object; no result was computed
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("a94fc028457ce3f8")
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 an instance of its own class, Args: the run constructed the object and did not call it, so no result was computed. 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)

# The Args class accepts a 'data' parameter which is a dict
# The __getattr__ method accesses items from self.data
# We need to construct a valid dict to pass as the 'data' parameter

test_data = {
    'model': nn.Linear(4, 3),
    'learning_rate': 0.001,
    'epochs': 10,
    'batch_size': 2,
    'device': 'cpu',
    'loss_fn': nn.CrossEntropyLoss(),
    'optimizer': 'adam',
    'input_dim': 4,
    'output_dim': 3,
}

ARGS = {
    'data': test_data
}
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 Args — the run constructed an instance of its own class and did not call it; no shape 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
MAB Ran
ran on a synthesized input · constructed an object; no result was computed
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("7232573389ce2a5a")
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 an instance of its own class, MAB: the run constructed the object and did not call it, so no result was computed. 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)

class ConfigStub:
    def __init__(self):
        self.model_mix_heads = True
        self.model_num_heads = 2
        self.model_sep_res_embed = True
        self.model_att_block_layer_norm = True
        self.model_layer_norm_eps = 1e-12
        self.model_rff_depth = 1
        self.model_att_score_norm = 'softmax'
        self.model_pre_layer_norm = True
        self.viz_att_maps = False
        self.model_ablate_rff = False
        self.model_hidden_dropout_prob = 0.1
        self.model_att_score_dropout_prob = 0.1

ARGS = {
    'dim_Q': 3,
    'dim_KV': 4,
    'dim_emb': 8,
    'dim_out': 6,
    'c': ConfigStub(),
}
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 MAB — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MAB 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
PNPT_sample_training_points_inference Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/proteinnpt/data_processing.py
code served (permissive licence) · get_code("d2d1d1bc5e8b6b1d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family int, shape (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

# Set random seed for reproducibility
np.random.seed(0)

# Construct training_sequences dictionary
training_sequences = {
    'mutant_mutated_seq_pairs': [
        ('A1G', 'ACGTACGT'),
        ('A2T', 'ACGTACGT'),
        ('A3C', 'ACGTACGT'),
        ('A4G', 'ACGTACGT'),
        ('B1T', 'TGCATGCA'),
        ('B2G', 'TGCATGCA'),
    ],
    'fitness': [0.5, 0.6, 0.7, 0.8, 0.9, 1.0]
}

# Define parameters
sequences_sampling_method = 'distributed_target'
num_sampled_points = 3

# Create ARGS dictionary
ARGS = {
    'training_sequences': training_sequences,
    'sequences_sampling_method': sequences_sampling_method,
    'num_sampled_points': num_sampled_points
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (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
Print Ran
ran on a synthesized input · constructed an object; no result was computed
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("d6f128d3e349e966")
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 an instance of its own class, Print: the run constructed the object and did not call it, so no result was computed. 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)

# The Print module's forward method takes a tensor x and prints its shape
# Then returns x unchanged. It has no parameters to configure.
# We need to provide a dummy input tensor to test the forward pass.

# Create a small test tensor (batch_size=2, feature_dim=4)
test_input = torch.randn(2, 4)

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Print — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Print 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
ReshapeToFlat Ran
ran on a synthesized input
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("0ead572d1093cc14")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (1, 2, 12) 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

class ReshapeToFlat(nn.Module):
    """Reshapes a tensor of shape (N, D, E) to (1, N, D*E)."""

    def __init__(self):
        super(ReshapeToFlat, self).__init__()

    @staticmethod
    def forward(X):
        return X.reshape(1, X.size(0), -1)


INIT_ARGS = {}

ARGS = {
    'X': torch.randn(2, 4, 3)
}

if __name__ == '__main__':
    model = ReshapeToFlat(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Input shape: {ARGS['X'].shape}")
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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, 2, 12) — shape (1, 2, 12) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ReshapeToNested Ran
ran on a synthesized input · constructed an object; no result was computed
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("e6ffb4964d98205f")
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 an instance of its own class, ReshapeToNested: the run constructed the object and did not call it, so no result was computed. 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)

# ReshapeToNested.__init__ takes D as a parameter
# ReshapeToNested.forward takes X (a tensor) as input
# The forward method reshapes X from (1, N, D*E) to (N, D, E)
# So we need to construct a tensor with shape (1, N, D*E)

D = 4  # D parameter for the module
N = 2  # N dimension (batch size in the reshaped output)
E = 3  # E dimension (last dimension in the reshaped output)

# Create input tensor X with shape (1, N, D*E) = (1, 2, 12)
X = torch.randn(1, N, D * E)

ARGS = {
    "D": D,
}
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 ReshapeToNested — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ReshapeToNested 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
apc Ran
ran; our contract draft was wrong, not the code
oatml-markslab/proteinnpt/proteinnpt/utils/esm/modules.py
code served (permissive licence) · get_code("f6701ae421cbe0d4")
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, 4, 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 torch

torch.manual_seed(0)

# The function apc performs average product correction on a tensor
# It expects a tensor x with at least 2 dimensions
# The operations use sum on dimensions -1 and -2, so we need at least 2D
# Let's create a small 4D tensor (batch, sequence, dimension, dimension)
# to make the contact prediction use case clear

x = torch.randn(2, 3, 4, 5)

ARGS = {
    "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 4, dtype family float, shape (2, 3, 4, 5) — shape (2, 3, 4, 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
assign_group_stratified Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/cv_split.py
code served (permissive licence) · get_code("7fc8e20691370cda")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family int, shape (10,) 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

# The function takes a single parameter 'x' which is used with:
# - len(x) to get the number of items
# That's it - x just needs to be a sequence with a length

x = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])

ARGS = {
    'x': x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (10,) — shape (10,) 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_weight Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/msa_utils.py
code served (permissive licence) · get_code("7b4449eb96bfa4eb")
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 small test sequences
# seq: a single sequence (1D array)
seq = np.array([1.0, 0.0, 1.0, 0.5], dtype=np.float64)

# list_seq: a 2D array of multiple sequences
list_seq = np.array([
    [1.0, 0.0, 1.0, 0.5],
    [1.0, 0.0, 0.8, 0.6],
    [0.9, 0.1, 1.0, 0.4],
    [0.5, 0.5, 0.5, 0.5]
], dtype=np.float64)

# theta: similarity threshold (float between 0 and 1)
theta = 0.1

ARGS = {
    "seq": seq,
    "list_seq": list_seq,
    "theta": theta
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numba, 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
compute_weight_all Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/msa_utils.py
code served (permissive licence) · get_code("0d397ce1b8012be1")
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 (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 numpy as np

# Create a small MSA (Multiple Sequence Alignment) represented as binary/continuous vectors
# Each row is a sequence, represented as a feature vector
num_sequences = 5
sequence_length = 4

# Generate synthetic sequence data
# Sequences are represented as numpy arrays (e.g., position encoding or binary)
list_seq = np.random.rand(num_sequences, sequence_length).astype(np.float64)

# Set some sequences to be more similar to test the weight computation
list_seq[0] = np.array([0.8, 0.7, 0.6, 0.5])
list_seq[1] = np.array([0.81, 0.71, 0.61, 0.51])  # Similar to seq 0
list_seq[2] = np.array([0.1, 0.2, 0.3, 0.4])
list_seq[3] = np.array([0.11, 0.21, 0.31, 0.41])  # Similar to seq 2
list_seq[4] = np.array([0.5, 0.5, 0.5, 0.5])

# Similarity threshold (between 0 and 1)
# Lower theta means stricter similarity requirement
theta = 0.1

# Update frequency for progress bar
updated_freq = 2

ARGS = {
    'list_seq': list_seq,
    'theta': theta,
    'updated_freq': updated_freq
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numba, numpy, tqdm — 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 (5,) — shape (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
gelu Ran
ran; our contract draft was wrong, not the code
oatml-markslab/proteinnpt/proteinnpt/utils/esm/modules.py
code served (permissive licence) · get_code("2e7bd6c4ccd1ed68")
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 2, dtype family float, shape (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 math
import torch

torch.manual_seed(0)

# The gelu function takes a single parameter 'x' which is used in:
# - x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
# So x must be a tensor that supports arithmetic operations and torch.erf

# Create a small tensor for testing
x = torch.randn(2, 4)

ARGS = {
    '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 2, dtype family float, shape (2, 4) — shape (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
get_learning_rate Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/model_utils.py
code served (permissive licence) · get_code("7dabe8f85e5f4e16")
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 math

# Construct valid arguments for get_learning_rate function
ARGS = {
    'training_step': 500,
    'num_warmup_steps': 1000,
    'num_total_training_steps': 20000,
    'max_learning_rate': 0.0003,
    'min_learning_rate': 3e-05
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
get_parameter_names Ran
ran; our contract draft was wrong, not the code
oatml-markslab/proteinnpt/proteinnpt/utils/model_utils.py
code served (permissive licence) · get_code("a8991266f955dd35")
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)

# Build a simple model with nested structure and parameters
class SimpleChild(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.child1 = SimpleChild()
        self.child2 = nn.Linear(4, 2)
        self.register_parameter('direct_param', nn.Parameter(torch.randn(3)))
    
    def forward(self, x):
        return self.child2(self.child1(x))

# Create model instance
model = SimpleModel()

# Define forbidden layer types - we'll exclude BatchNorm layers for example
forbidden_layer_types = [nn.BatchNorm1d, nn.BatchNorm2d]

ARGS = {
    "model": model,
    "forbidden_layer_types": forbidden_layer_types,
}
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
get_slopes Ran
ran, honoured the contract we drafted
oatml-markslab/proteinnpt/proteinnpt/utils/tranception/model_pytorch.py
code served (permissive licence) · get_code("4305f4fdd8e54bc1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned 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 math

ARGS = {
    'n': 8,
    'mode': 'standard_alibi',
    'verbose': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
keep_singles Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/cv_split.py
code served (permissive licence) · get_code("ea7a41d551b7a7e5")
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 DataFrame. 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 pandas as pd

# Construct a minimal DataFrame with a 'mutant' column
# The function filters rows where the mutant column contains ':'
data = {
    'mutant': ['A1B', 'C2D:E3F', 'G4H', 'I5J:K6L:M7N'],
    'value': [10, 20, 30, 40]
}
DMS = pd.DataFrame(data)

ARGS = {
    'DMS': DMS,
    'mutant_column': 'mutant'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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
learning_rate_scheduler Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/model_utils.py
code served (permissive licence) · get_code("355bff3ffb0d66a7")
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 math

def get_learning_rate(training_step, num_warmup_steps=1000, num_total_training_steps=20000, max_learning_rate=0.0003, min_learning_rate=3e-05):
    """
    """
    if training_step <= num_warmup_steps:
        lr = max_learning_rate * training_step / num_warmup_steps
    elif training_step > num_total_training_steps:
        lr = min_learning_rate
    else:
        ratio_total_steps_post_warmup = (training_step - num_warmup_steps) / (num_total_training_steps - num_warmup_steps)
        cosine_scaler = 0.5 * (1.0 + math.cos(math.pi * ratio_total_steps_post_warmup))
        lr = min_learning_rate + cosine_scaler * (max_learning_rate - min_learning_rate)
    return lr

def learning_rate_scheduler(num_warmup_steps=1000, num_total_training_steps=20000, max_learning_rate=0.0003, min_learning_rate=3e-05):

    def get_lr(training_step):
        return get_learning_rate(training_step, num_warmup_steps, num_total_training_steps, max_learning_rate, min_learning_rate)
    return get_lr

ARGS = {
    "num_warmup_steps": 1000,
    "num_total_training_steps": 20000,
    "max_learning_rate": 0.0003,
    "min_learning_rate": 3e-05
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
one_hot_3D Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/msa_utils.py
code served (permissive licence) · get_code("52a00d097b2f46fd")
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 3, dtype family float, shape (2, 4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

ARGS = {
    "seq_keys": ["seq1", "seq2"],
    "seq_name_to_sequence": {
        "seq1": "ACGT",
        "seq2": "TGCA"
    },
    "alphabet": "ACGT",
    "seq_length": 4
}
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 3, dtype family float, shape (2, 4, 4) — shape (2, 4, 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
return_position_single Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/cv_split.py
code served (permissive licence) · get_code("4074372860cf12f3")
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

def return_position_single(mutation):
    """Note: Only works for single mutations"""
    position = mutation.split(':')[0][1:-1]
    return int(position)

# The function takes a mutation string and extracts the position
# The mutation string format appears to be: "X###:..." where X is a letter
# and ### is the position number. The function:
# 1. Splits on ':' and takes the first part
# 2. Removes the first and last character [1:-1]
# 3. Converts to int

mutation_string = "A123:G"

ARGS = {
    "mutation": mutation_string
}
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
split_data_based_on_test_fold_index Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/data_utils.py
code served (permissive licence) · get_code("72ad19ee550806af")
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 pandas as pd
import numpy as np

# Create a sample dataframe with fold assignments
np.random.seed(0)
n_samples = 100
fold_data = {
    'fold_modulo_5': np.random.randint(0, 5, n_samples),
    'feature_1': np.random.randn(n_samples),
    'feature_2': np.random.randn(n_samples),
}
dataframe = pd.DataFrame(fold_data)

ARGS = {
    'dataframe': dataframe,
    'fold_variable_name': 'fold_modulo_5',
    'test_fold_index': 0,
    'use_validation_set': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — 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
standardize Ran
ran on a synthesized input
oatml-markslab/proteinnpt/proteinnpt/utils/data_utils.py
code served (permissive licence) · get_code("74862b2cc41d3a95")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (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 numpy as np
import torch

torch.manual_seed(0)

# Create a small test array with known values
# Using shape (2, 4) for small batch and feature dimensions
x = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [5.0, 6.0, 7.0, 8.0]
], dtype=np.float32)

ARGS = {
    "x": x,
    "epsilon": 1e-10
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (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
symmetrize Ran
ran, violated the contract we drafted
oatml-markslab/proteinnpt/proteinnpt/utils/esm/modules.py
code served (permissive licence) · get_code("b385be9826c4a8b3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 3, dtype family float, shape (2, 4, 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)

# Create a small tensor with shape (2, 4, 4) for the final two dimensions to be square
# (required for transpose and addition to work meaningfully)
x = torch.randn(2, 4, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 (2, 4, 4) — shape (2, 4, 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
ImagePatcher Not yet run
unverified (harvested, no recorded run)
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("8d76f18f88c52502")
How it ran

no invocation recorded.

LinearImagePatcher Not yet run
unverified (harvested, no recorded run)
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("ce48ecd197be4b48")
How it ran

no invocation recorded.

MHSA Not yet run
unverified (harvested, no recorded run)
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("f3dd2c06a9c7eb93")
How it ran

no invocation recorded.

NPTModel Not yet run
unverified (harvested, no recorded run)
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("611506804344b302")
How it ran

no invocation recorded.

SaveAttMaps Not yet run
unverified (harvested, no recorded run)
OATML/Non-Parametric-Transformers/npt/model/npt.py
code served (permissive licence) · get_code("67c82e2cb5a619f5")
How it ran

no invocation recorded.

get_tranception_tokenizer Not yet run
unverified (harvested, no recorded run)
oatml-markslab/proteinnpt/proteinnpt/utils/tranception/model_pytorch.py
code served (permissive licence) · get_code("e20cd5502a6c2414")
How it ran

no invocation recorded.

nanmean Not yet run
unverified (harvested, no recorded run)
oatml-markslab/proteinnpt/proteinnpt/utils/tranception/model_pytorch.py
code served (permissive licence) · get_code("5574012ba4af04ff")
How it ran

no invocation recorded.

process_batch Not yet run
unverified (harvested, no recorded run)
oatml-markslab/proteinnpt/proteinnpt/baselines/data_processing.py
code served (permissive licence) · get_code("c5e4288eb459c27d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 30 samples lifted from this paper's linked repositories, 30 carry a function identity and 0 do not. Of the 30 distinct functions with an identity, 7 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 2 the same month, 7 later) and 23 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
gelu 198 other papers
77 dated earlier, 4 the same month, 113 later, 4 same year, order not determinable · 32 only as an assembly brick
1408.5882 (2014-08)
Convolutional Neural Networks for Sentence Classification
1810.04805 (2018-10)
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
1901.10125 (2019-01)
Glyce: Glyph-vectors for Chinese Character Representations
and 195 more not named here
count_parameters 180 other papers
36 dated earlier, 7 the same month, 137 later · 8 only as an assembly brick
1705.08415 (2017-05)
Supervised Community Detection with Line Graph Neural Networks
1707.01926 (2017-07)
Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting
1808.09367 (2018-08)
Deriving Machine Attention from Human Rationales
and 177 more not named here
symmetrize 5 other papers
5 later
2206.06583 (2022-06)
Exploring evolution-aware &amp; -free protein language models as protein function predictors
2406.10391 (2024-06)
BEACON: Benchmark for Comprehensive RNA Tasks and Language Models
2408.11363 (2024-08)
ProteinGPT: Multimodal LLM for Protein Property Prediction and Structure Understanding
and 2 more not named here
apc 4 other papers
4 later
2206.06583 (2022-06)
Exploring evolution-aware &amp; -free protein language models as protein function predictors
2408.11363 (2024-08)
ProteinGPT: Multimodal LLM for Protein Property Prediction and Structure Understanding
2605.11189 (2026-05)
Deep Learning for Protein Complex Prediction and Design by
and 1 more not named here
get_parameter_names 4 other papers
4 later
2405.18540 (2024-05)
Learning diverse attacks on large language models for robust red-teaming and safety tuning
2407.09533 (2024-07)
Video Occupancy Models
2408.16090 (2024-08)
EPO: Hierarchical LLM Agents with Environment Preference Optimization
and 1 more not named here
nanmean 2 other papers
2 later
2205.13760 (2022-05)
Tranception: protein fitness prediction with autoregressive transformers and inference-time retrieval
2212.07328 (2022-12)
MODELING MULTIMODAL ALEATORIC UNCERTAINTY IN SEGMENTATION WITH MIXTURE OF STOCHASTIC EXPERTS
get_slopes 1 other paper
1 later
2205.13760 (2022-05)
Tranception: protein fitness prediction with autoregressive transformers and inference-time retrieval

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We challenge a common assumption underlying most supervised deep learning: that a model makes a prediction depending only on its parameters and the features of a single input. To this end, we introduce a general-purpose deep learning architecture that takes as input the entire dataset instead of processing one datapoint at a time. Our approach uses self-attention to reason about relationships between datapoints explicitly, which can be seen as realizing non-parametric models using parametric attention mechanisms. However, unlike conventional non-parametric models, we let the model learn end-to-end from the data how to make use of other datapoints for prediction. Empirically, our models solve cross-datapoint lookup and complex reasoning tasks unsolvable by traditional deep learning models. We show highly competitive results on tabular data, early results on CIFAR-10, and give insight into how the model makes use of the interactions between points.

For agents

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

get_harvested_code_for_paper("2106.02584")
get_code_for_paper("2106.02584")
have("2106.02584")

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

curl https://syntology.ai/api/ran/2106.02584.json

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

[![Syntology run record](https://syntology.ai/api/ran/2106.02584.svg)](https://syntology.ai/paper/2106.02584)

Connect an agent — have() is free.