SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.25599 · 2026

PLMGH: What Matters in PLM-GNN Hybrids for Code Classification and Vulnerability Detection

Montréal Polytechnique, Canada Montréal, Mohamed Taoufik, Kaouthar El

arXiv · PDF · Open in the Atlas

Code that ran

4 of 5 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 2 failures of our instrument; 1 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/2604.25599.json · badge: /api/ran/2604.25599.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
PlayeerOne/PLMGH community
Papers with Code archive
4 ran · 1 no recorded run
FunctionStatusWhere it lives
Cls_Decoder Ran
ran on a synthesized input · constructed an object; no result was computed
PlayeerOne/PLMGH/model/gnn_classifier.py
code served (permissive licence) · get_code("7b62e9f0433f7918")
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, Cls_Decoder: 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
from torch_geometric.data import Data

torch.manual_seed(0)

# Create a minimal Data object with required attributes
# The forward method accesses input_graph.x and input_graph.batch
batch_size = 2
in_dim = 4
num_nodes = 6

x = torch.randn(num_nodes, in_dim)
batch = torch.tensor([0, 0, 0, 1, 1, 1])

input_graph = Data(x=x, batch=batch)

ARGS = {
    'in_dim': 4,
    'h_dim': 8,
    'out_dim': 250,
    'dropout': 0.25,
    'pooling_method': 'attentional'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torch_geometric — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Cls_Decoder — 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 Cls_Decoder 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
FeatureFusionLayer Ran
ran on a synthesized input · constructed an object; no result was computed
PlayeerOne/PLMGH/model/gnn_classifier.py
code served (permissive licence) · get_code("5ede0eb7bfb35614")
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, FeatureFusionLayer: 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
from torch_geometric.data import Data

torch.manual_seed(0)

# Create a minimal Data object with all required attributes
num_nodes = 2
node_type_idx = torch.randint(0, 250, (num_nodes,))
positional_feats = torch.randn(num_nodes, 32)
semantic_feats = torch.randn(num_nodes, 768)

g = Data()
g.node_type_idx = node_type_idx
g.positional_feats = positional_feats
g.semantic_feats = semantic_feats

ARGS = {
    'node_type_key': 'node_type_idx',
    'use_node_type': True,
    'pos_features_key': 'positional_feats',
    'use_pos_features': True,
    'semantic_features_key': 'semantic_feats',
    'use_semantic_features': True,
    'semantic_features_dim': 768,
    'pos_features_dim': 32,
    'node_type_dim': 256,
    'n_node_types': 250,
    'h_dim': 512,
    'out_dim': 256,
    'dropout': 0.25,
    'fusion_method': 'concat',
    'padding_idx': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torch_geometric — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type FeatureFusionLayer — 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 FeatureFusionLayer 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
_make_cls_metrics Ran
ran; our contract draft was wrong, not the code
PlayeerOne/PLMGH/model/gnn_classifier.py
code served (permissive licence) · get_code("922bef3f14cc2e77")
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 MetricCollection. 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 torchmetrics.classification as C
from torchmetrics import MetricCollection

torch.manual_seed(0)

# Test case 1: Binary classification with probabilities
ARGS = {
    'num_classes': 2,
    'average': None,
    'use_probs': True,
    'threshold': 0.5,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchmetrics — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MetricCollection — a MetricCollection output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MetricCollection output the recorded ndim 1 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
cross_entropy Ran
ran; our contract draft was wrong, not the code
PlayeerOne/PLMGH/model/gnn_classifier.py
code served (permissive licence) · get_code("64e83a2a64136d78")
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 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Test case 1: Binary classification with logits
# pred shape: (2, 1) for binary, target shape: (2,)
pred_binary = torch.randn(2, 1)
target_binary = torch.tensor([0, 1])

# Test case 2: Multiclass classification with logits
# pred shape: (2, 4) for 4 classes, target shape: (2,)
pred_multiclass = torch.randn(2, 4)
target_multiclass = torch.tensor([0, 2])

# Class weights for multiclass (4 classes)
class_weights = torch.tensor([1.0, 1.5, 0.8, 1.2])

# Pos weight for binary classification
pos_weight = torch.tensor([2.0])

# Use multiclass setup as the primary test case
ARGS = {
    'pred': pred_multiclass,
    'target': target_multiclass,
    'from_logits': True,
    'class_weights': class_weights,
    'pos_weight': None,
    'label_smoothing': 0.1,
    'bce_label_smoothing': 0.0,
    'reduction': 'mean'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchmetrics — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, 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
GNN_Classifier Not yet run
unverified (harvested, no recorded run)
PlayeerOne/PLMGH/model/gnn_classifier.py
code served (permissive licence) · get_code("a34994b2fba5fb9e")
How it ran

no invocation recorded.

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 5 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

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

Repositories linked to this paper

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

Abstract

Code understanding models increasingly rely on pretrained language models (PLMs) and graph neural networks (GNNs), which capture complementary semantic and structural information. We conduct a controlled empirical study of PLM→GNN hybrids for code classification and vulnerability detection tasks by systematically pairing three code-specialized PLMs with three foundational GNN architectures. We compare these hybrids against PLMonly and GNN-only baselines on Java250 and Devign, including an identifier-obfuscation setting. Across both tasks, hybrids consistently outperform GNN-only baselines and often improve ranking quality over frozen PLMs. On Devign, performance and robustness are more sensitive to the PLM feature source than to the GNN backbone. We also find that larger PLMs are not necessarily better feature extractors in this pipeline, and that the PLM choice has more impact than the GNN choice. Finally, we distill these findings into practical guidelines for PLM→GNN design choices in code classification and vulnerability detection.

For agents

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

get_harvested_code_for_paper("2604.25599")
get_code_for_paper("2604.25599")
have("2604.25599")

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

curl https://syntology.ai/api/ran/2604.25599.json

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

[![Syntology run record](https://syntology.ai/api/ran/2604.25599.svg)](https://syntology.ai/paper/2604.25599)

Connect an agent — have() is free.