SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.07588 · 2026

Unified Biomolecular Trajectory Generation via Pretrained Variational Bridge

Yang Liu, Wenbing Huang, Ziyang Yu

arXiv · PDF · Open in the Atlas

Code that ran

15 of 21 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (10 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 10 have no contract verdict; 4 failures of our instrument; 6 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/2602.07588.json · badge: /api/ran/2602.07588.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
yaledeus/PVB community
Papers with Code archive
15 ran · 6 no recorded run
FunctionStatusWhere it lives
CosineCutoff Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("443e8e4b7a689a29")
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, CosineCutoff: 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
from torch import nn
import math

torch.manual_seed(0)

class CosineCutoff(nn.Module):

    def __init__(self, cutoff_lower=0.0, cutoff_upper=5.0):
        super(CosineCutoff, self).__init__()
        self.cutoff_lower = cutoff_lower
        self.cutoff_upper = cutoff_upper

    def forward(self, distances: torch.Tensor) -> torch.Tensor:
        if self.cutoff_lower > 0:
            cutoffs = 0.5 * (torch.cos(math.pi * (2 * (distances - self.cutoff_lower) / (self.cutoff_upper - self.cutoff_lower) + 1.0)) + 1.0)
            cutoffs = cutoffs * (distances < self.cutoff_upper)
            cutoffs = cutoffs * (distances > self.cutoff_lower)
            return cutoffs
        else:
            cutoffs = 0.5 * (torch.cos(distances * math.pi / self.cutoff_upper) + 1.0)
            cutoffs = cutoffs * (distances < self.cutoff_upper)
            return cutoffs


# Create valid test arguments
cutoff_lower = 0.5
cutoff_upper = 5.0

ARGS = {
    'cutoff_lower': cutoff_lower,
    'cutoff_upper': cutoff_upper
}
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 CosineCutoff — 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 CosineCutoff 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
ExpNormalSmearing Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("78a08041969c2349")
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, ExpNormalSmearing: 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
from torch import nn
import math

torch.manual_seed(0)

ARGS = {
    "cutoff_lower": 0.0,
    "cutoff_upper": 5.0,
    "num_rbf": 50,
    "trainable": True,
    "dtype": torch.float32,
}
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 ExpNormalSmearing — 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 ExpNormalSmearing 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
GaussianSmearing Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("9b06397336dacf66")
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, GaussianSmearing: 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
from torch import nn

torch.manual_seed(0)

ARGS = {
    'cutoff_lower': 0.0,
    'cutoff_upper': 5.0,
    'num_rbf': 50,
    'trainable': True,
    'dtype': torch.float32,
}
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 GaussianSmearing — 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 GaussianSmearing 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
MLP Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("399a9aa98224f17a")
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, MLP: 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
from torch import nn
import torch.nn.functional as F

torch.manual_seed(0)

class ShiftedSoftplus(nn.Module):
    """Applies the ShiftedSoftplus function :math:`\\text{ShiftedSoftplus}(x) = \\frac{1}{\\beta} *
    \\log(1 + \\exp(\\beta * x))-\\log(2)` element-wise.

    SoftPlus is a smooth approximation to the ReLU function and can be used
    to constrain the output of a machine to always be positive.
    """

    def __init__(self):
        super(ShiftedSoftplus, self).__init__()
        self.shift = torch.log(torch.tensor(2.0)).item()

    def forward(self, x):
        return F.softplus(x) - self.shift

act_class_mapping = {'ssp': ShiftedSoftplus, 'silu': nn.SiLU, 'tanh': nn.Tanh, 'sigmoid': nn.Sigmoid}

class MLP(nn.Module):
    """A simple multi-layer perceptron with a given number of layers and hidden channels.

    The simplest MLP has no hidden layers and is composed of two linear layers with a non-linear activation function in between:

    .. math::

        \\text{MLP}(x) = \\text{Linear}_o(\\text{act}(\\text{Linear}_i(x)))

    Where :math:`\\text{Linear}_i` has input size :math:`\\text{in_channels}` and output size :math:`\\text{hidden_channels}` and :math:`\\text{Linear}_o` has input size :math:`\\text{hidden_channels}` and output size :math:`\\text{out_channels}`.


    Args:
        in_channels (int): Number of input features.
        out_channels (int): Number of output features.
        hidden_channels (int): Number of hidden features.
        activation (str): Activation function to use.
        num_hidden_layers (int, optional): Number of hidden layers. Defaults to 0.
        dtype (torch.dtype, optional): Data type to use. Defaults to torch.float32.
    """

    def __init__(self, in_channels, out_channels, hidden_channels, activation, num_hidden_layers=0, dtype=torch.float32):
        super(MLP, self).__init__()
        act_class = act_class_mapping[activation]
        self.act = act_class()
        self.layers = nn.Sequential()
        self.layers.append(nn.Linear(in_channels, hidden_channels, dtype=dtype))
        self.layers.append(self.act)
        for _ in range(num_hidden_layers):
            self.layers.append(nn.Linear(hidden_channels, hidden_channels, dtype=dtype))
            self.layers.append(self.act)
        self.layers.append(nn.Linear(hidden_channels, out_channels, dtype=dtype))

    def reset_parameters(self):
        for layer in self.layers:
            if isinstance(layer, nn.Linear):
                nn.init.xavier_uniform_(layer.weight)
                layer.bias.data.fill_(0)

    def forward(self, x):
        x = self.layers(x)
        return x


ARGS = {
    'in_channels': 4,
    'out_channels': 3,
    'hidden_channels': 8,
    'activation': 'silu',
    'num_hidden_layers': 2,
    'dtype': torch.float32,
}
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 MLP — 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 MLP 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
NeighborEmbedding Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("b594864adc680a61")
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, NeighborEmbedding: 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
from torch import nn
import math

torch.manual_seed(0)

# Small example tensors and parameters
num_nodes = 2
num_edges = 3
hidden_channels = 4
num_rbf = 5
cutoff_lower = 0.0
cutoff_upper = 5.0
max_z = 100
dtype = torch.float32

# Create input tensors
z = torch.tensor([1, 8], dtype=torch.long)  # Atomic numbers for H, O
x = torch.randn(num_nodes, 3, dtype=dtype)  # Node positions (3D coordinates)

# Edge index: [2, num_edges] - source and target node indices
edge_index = torch.tensor([
    [0, 0, 1],  # source nodes
    [1, 1, 0]   # target nodes
], dtype=torch.long)

# Edge weights (distances): [num_edges]
edge_weight = torch.tensor([1.5, 2.0, 1.5], dtype=dtype)

# Edge attributes (RBF features): [num_edges, num_rbf]
edge_attr = torch.randn(num_edges, num_rbf, dtype=dtype)

ARGS = {
    'hidden_channels': hidden_channels,
    'num_rbf': num_rbf,
    'cutoff_lower': cutoff_lower,
    'cutoff_upper': cutoff_upper,
    'max_z': max_z,
    'dtype': dtype,
}
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 NeighborEmbedding — 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 NeighborEmbedding 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
OptimizedDistance Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("54c2c64bd68403ef")
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, OptimizedDistance: 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
import numpy as np

torch.manual_seed(0)

# Create tiny test positions: 4 atoms in 3D space
pos = torch.randn(4, 3, dtype=torch.float32)

# Create batch tensor: which sample each atom belongs to
batch = torch.tensor([0, 0, 1, 1], dtype=torch.long)

# Create a valid periodic box that satisfies the requirements
# a[1] = a[2] = b[2] = 0
# a[0] >= 2*cutoff, b[1] >= 2*cutoff, c[2] >= 2*cutoff
# a[0] >= 2*b[0], a[0] >= 2*c[0], b[1] >= 2*c[1]
cutoff_upper = 5.0
cutoff_lower = 0.0
box_size = cutoff_upper * 3.0

box = torch.tensor([
    [box_size, 0.0, 0.0],
    [0.0, box_size, 0.0],
    [0.0, 0.0, box_size]
], dtype=torch.float32)

ARGS = {
    'cutoff_lower': cutoff_lower,
    'cutoff_upper': cutoff_upper,
    'max_num_pairs': -32,
    'return_vecs': False,
    'loop': False,
    'strategy': 'brute',
    'include_transpose': True,
    'resize_to_fit': True,
    'check_errors': True,
    'box': box,
    'long_edge_index': True
}
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 OptimizedDistance — 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 OptimizedDistance 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
ParamGVPFFNLayer Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("88dbb28d5b0f2288")
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, ParamGVPFFNLayer: 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
from torch import nn

torch.manual_seed(0)

# Small stand-in tensors for testing
# H: hidden state tensor, shape (batch_size, d_hidden)
# V: vector tensor, shape (batch_size, n_vectors, d_hidden)
# p: optional parameter tensor

batch_size = 2
d_hidden = 4
d_ffn = 8
d_output = 3
param_dim = 2

H = torch.randn(batch_size, d_hidden)
V = torch.randn(batch_size, 5, d_hidden)  # 5 vectors per batch
p = torch.randn(batch_size, param_dim)

# Define activation function
act_fn = nn.SiLU()

ARGS = {
    "d_hidden": d_hidden,
    "d_ffn": d_ffn,
    "act_fn": act_fn,
    "d_output": d_output,
    "vector_act": "none",
    "vector_norm": False,
    "param_dim": param_dim,
}

# Test data for forward pass
TEST_FORWARD_ARGS = {
    "H": H,
    "V": V,
    "p": p,
}
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 ParamGVPFFNLayer — 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 ParamGVPFFNLayer 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
SO3LayerNorm Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("0e13725b367dee5d")
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, SO3LayerNorm: 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
from torch import nn

torch.manual_seed(0)

# SO3LayerNorm expects:
# - dim: int parameter for which dimension to compute norm over

ARGS = {
    "dim": 1
}
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 SO3LayerNorm — 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 SO3LayerNorm 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
ShiftedSoftplus Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("52c667a05695e6eb")
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, ShiftedSoftplus: 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
from torch import nn

torch.manual_seed(0)

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 ShiftedSoftplus — 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 ShiftedSoftplus 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
StochasticInterpolantMatcher Ran
ran on a synthesized input · constructed an object; no result was computed
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("918089d623f37c06")
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, StochasticInterpolantMatcher: 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 numpy as np

torch.manual_seed(0)

# Create small test tensors
batch_size = 2
dim = 4

x0 = torch.randn(batch_size, dim)
x1 = torch.randn(batch_size, dim)
xt = torch.randn(batch_size, dim)

# Time tensor - batch of scalars in [0, 1]
t = torch.tensor([0.3, 0.7])

# Scalar sigma for diffusion
sigma = torch.tensor(0.5)

# Velocity field (x1 - x0)
vel = x1 - x0

# Denoiser/noise field
den = torch.randn(batch_size, dim)

ARGS = {
}
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 StochasticInterpolantMatcher — 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 StochasticInterpolantMatcher 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
_construct_knn_graph Ran
ran; our contract draft was wrong, not the code
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("d7712884f7817823")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct small coordinate tensor (N=4 nodes, 3D coordinates)
X = torch.randn(4, 3, dtype=torch.float32)

# Construct rule_mats: list of (N, N) boolean adjacency matrices
# These represent valid edges after filtering
rule_mat1 = torch.tensor([
    [False, True, True, False],
    [True, False, True, True],
    [True, True, False, True],
    [False, True, True, False]
], dtype=torch.bool)

rule_mat2 = torch.tensor([
    [False, True, False, False],
    [True, False, True, True],
    [False, True, False, True],
    [False, True, True, False]
], dtype=torch.bool)

rule_mats = [rule_mat1, rule_mat2]

# Number of neighbors to keep
k_neighbors = 2

# Reverse flag
reverse = False

# Undirected flag
undirected = False

ARGS = {
    'X': X,
    'rule_mats': rule_mats,
    'k_neighbors': k_neighbors,
    'reverse': reverse,
    'undirected': undirected
}
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 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
_topk Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("3383b300e62cef36")
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 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 a small input tensor with shape (batch=2, features=4)
input_tensor = torch.randn(2, 4)

# Set k to a value smaller than the feature dimension (4)
# This will trigger the else branch with repetition logic
k_value = 6

# dim parameter - using 1 to sort along the feature dimension
dim_value = 1

# largest parameter - True to get top-k values
largest_value = True

ARGS = {
    "input": input_tensor,
    "k": k_value,
    "dim": dim_value,
    "largest": largest_value,
}
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 tuple, ndim 3, 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
construct_edges Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("38e7dd80e9b900cb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Tiny test setup for construct_edges
N = 4  # number of nodes
Z = torch.randint(1, 8, (N,))  # atomic numbers
X = torch.randn(N, 3)  # coordinates (N, 3)
bid = torch.tensor([0, 0, 1, 1])  # batch IDs
mask = torch.tensor([1, 1, 0, 0])  # mask (1 for ligand, 0 for protein)
bond_index = torch.tensor([[0, 1], [1, 2]]).t().contiguous()  # (2, num_bonds)

ARGS = {
    "Z": Z,
    "X": X,
    "bid": bid,
    "mask": mask,
    "bond_index": bond_index,
    "cutoff_lower": 0.0,
    "cutoff_upper": 5.0,
    "cutoff_H": 3.5,
    "k_neighbors": 2,
}
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 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
edge_inclusion_mask Ran
ran, honoured the contract we drafted
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("1b9130ea3059f877")
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 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 torch

torch.manual_seed(0)

# edge_inclusion_mask takes two parameters: edge_index and bond_index
# Both are used with .max() and indexing operations [0] and [1]
# They represent edge connections as 2 x N tensors (source and target node indices)

# Create edge_index: 2 x 3 tensor representing 3 edges
edge_index = torch.tensor([
    [0, 1, 2],  # source nodes
    [1, 2, 0]   # target nodes
], dtype=torch.long)

# Create bond_index: 2 x 2 tensor representing 2 bonds (subset of edges)
bond_index = torch.tensor([
    [0, 1],  # source nodes
    [1, 2]   # target nodes
], dtype=torch.long)

ARGS = {
    "edge_index": edge_index,
    "bond_index": bond_index
}
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 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
scatter Ran
ran; our contract draft was wrong, not the code
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("ef4e8a8441121db8")
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 (3, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create test tensors
src = torch.randn(2, 4)  # Source tensor: batch=2, features=4
index = torch.tensor([[0, 1, 2, 0], [1, 2, 0, 1]], dtype=torch.long)  # Index tensor with same shape as src
dim = 0
dim_size = 3
reduce = 'sum'

ARGS = {
    'src': src,
    'index': index,
    'dim': dim,
    'dim_size': dim_size,
    'reduce': reduce,
}
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 (3, 4) — shape (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
EquivariantMultiHeadAttention Not yet run
unverified (harvested, no recorded run)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("6b94fbc30cae5c34")
How it ran

no invocation recorded.

TorchMD_VQ_ET Not yet run
unverified (harvested, no recorded run)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("ef6cf86632e97573")
How it ran

no invocation recorded.

_broadcast Not yet run
unverified (harvested, no recorded run)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("7c50644e7481fffa")
How it ran

no invocation recorded.

_init_linear_ Not yet run
unverified (harvested, no recorded run)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("8e64d43fb37434b3")
How it ran

no invocation recorded.

dyVAE Not yet run
unverified (harvested, no recorded run)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("6cc7fdca5fe5e062")
How it ran

no invocation recorded.

sequential_and Not yet run
unverified (harvested, no recorded run)
yaledeus/PVB/module/model.py
code served (permissive licence) · get_code("0f29a3d2ad9f4458")
How it ran

no invocation recorded.

Also present in other papers' code

Of 21 samples lifted from this paper's linked repositories, 21 carry a function identity and 0 do not. Of the 21 distinct functions with an identity, 11 are also present in other papers whose code we hold (11 with at least one paper dated earlier, 0 the same month, 0 later) and 10 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
CosineCutoff
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
ExpNormalSmearing
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
GaussianSmearing
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
ParamGVPFFNLayer
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
SO3LayerNorm
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
ShiftedSoftplus
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
_broadcast
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
_init_linear_
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
_topk
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
scatter
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules
sequential_and
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2506.03157 (2025-06)
UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules

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

Molecular Dynamics (MD) simulations provide a fundamental tool for characterizing molecular behavior at full atomic resolution, but their applicability is severely constrained by the computational cost. To address this, a surge of deep generative models has recently emerged to learn dynamics at coarsened timesteps for efficient trajectory generation, yet they either generalize poorly across systems or, due to limited molecular diversity of trajectory data, fail to fully exploit structural information to improve generative fidelity. Here, we present the Pretrained Variational Bridge (PVB) in an encoder-decoder fashion, which maps the initial structure into a noised latent space and transports it toward stage-specific targets through augmented bridge matching. This unifies training on both single-structure and paired trajectory data, enabling consistent use of cross-domain structural knowledge across training stages. Moreover, for protein-ligand complexes, we further introduce a reinforcement learning-based optimization via adjoint matching that speeds progression toward the holo state, which supports efficient post-optimization of docking poses. Experiments on proteins and protein-ligand complexes demonstrate that PVB faithfully reproduces thermodynamic and kinetic observables from MD while delivering stable and efficient generative dynamics.

For agents

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

get_harvested_code_for_paper("2602.07588")
get_code_for_paper("2602.07588")
have("2602.07588")

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

curl https://syntology.ai/api/ran/2602.07588.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.07588.svg)](https://syntology.ai/paper/2602.07588)

Connect an agent — have() is free.