SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2106.07476 · ICML · 2021

Training Graph Neural Networks with 1000 Layers

Vladlen Koltun, Bernard Ghanem, Guohao Li, Matthias Üller

arXiv · PDF · Open in the Atlas

Code that ran

11 of 11 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 3 failures of our instrument; 0 no recorded run), 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/2106.07476.json · badge: /api/ran/2106.07476.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
lightaime/deep_gcns_torch canonical (classifier label)
classifier basis: census
11 ran · 0 no recorded run
FunctionStatusWhere it lives
act_layer Ran
ran; our contract draft was wrong, not the code
lightaime/deep_gcns_torch/gcn_lib/dense/torch_nn.py
code served (permissive licence) · get_code("81d31aad45664fa2")
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 ReLU. 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 = {
    'act': 'relu',
    'inplace': False,
    'neg_slope': 0.2,
    'n_prelu': 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 ReLU — a ReLU output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ReLU 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
act_layer Ran
ran on a synthesized input
lightaime/deep_gcns_torch/gcn_lib/sparse/torch_nn.py
code served (permissive licence) · get_code("6b65956f32ad8f13")
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 ReLU. 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 = {
    'act_type': 'relu',
    'inplace': False,
    'neg_slope': 0.2,
    'n_prelu': 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 ReLU — a ReLU output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ReLU 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
batched_index_select Ran
ran on a synthesized input
lightaime/deep_gcns_torch/gcn_lib/dense/torch_nn.py
code served (permissive licence) · get_code("c52260211d1aa99f")
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, 4, 3, 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)

# x should be shape (B, C, N, 1) as per docstring
# Let's use: batch_size=2, num_dims=4, num_vertices=3, and the last dim is 1
batch_size = 2
num_dims = 4
num_vertices = 3

x = torch.randn(batch_size, num_dims, num_vertices, 1)

# idx should be shape (B, N, k) as per docstring
# where k is the number of neighbors to fetch
k = 5
idx = torch.randint(0, num_vertices, (batch_size, num_vertices, k))

ARGS = {
    "x": x,
    "idx": idx,
}
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, 4, 3, 5) — shape (2, 4, 3, 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
dense_knn_matrix Ran
ran on a synthesized input
lightaime/deep_gcns_torch/gcn_lib/dense/torch_edge.py
code served (permissive licence) · get_code("7bc52d942ecf4430")
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 int, shape (2, 2, 3, 2) 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 input tensor with shape (batch_size, num_dims, num_points, 1)
# batch_size=2, num_dims=4, num_points=3, last dim=1
x = torch.randn(2, 4, 3, 1)

# Define k parameter
k = 2

ARGS = {
    "x": x,
    "k": k
}
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 int, shape (2, 2, 3, 2) — shape (2, 2, 3, 2) 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
knn_graph_matrix Ran
ran; our contract draft was wrong, not the code
lightaime/deep_gcns_torch/gcn_lib/sparse/torch_edge.py
code served (permissive licence) · get_code("c69d40b71d37d2d2")
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 int, shape (2, 30) 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 point cloud tensor
# Shape: (batch_size, num_points, num_dims)
# Using batch_size=2, num_points=5, num_dims=4
x = torch.randn(2, 5, 4)

# Create batch indices if needed
# batch parameter should indicate which batch each point belongs to
# For batch_size=2 with 5 points each, we have 10 total points
batch = torch.tensor([0, 0, 0, 0, 0, 1, 1, 1, 1, 1])

# Flatten x to (num_points_total, num_dims) format that knn_graph_matrix expects
# The function will reshape it internally based on batch
x_flat = x.reshape(-1, 4)

ARGS = {
    "x": x_flat,
    "k": 3,
    "batch": batch
}
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 int, shape (2, 30) — shape (2, 30) 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
knn_matrix Ran
ran on a synthesized input
lightaime/deep_gcns_torch/gcn_lib/sparse/torch_edge.py
code served (permissive licence) · get_code("2f0e8668c137d3aa")
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 int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct a small point cloud tensor
# Shape: (batch_size, num_points, num_dims)
x = torch.randn(2, 5, 4)

# k parameter for KNN
k = 3

# batch parameter (can be None or a tensor indicating batch boundaries)
batch = None

ARGS = {
    "x": x,
    "k": k,
    "batch": batch,
}
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 int — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_pretrained_models Ran
ran on a synthesized input
lightaime/deep_gcns_torch/utils/ckpt_util.py
code served (permissive licence) · get_code("3be91b595771f70b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import torch
import torch.nn as nn
from collections import OrderedDict
import numpy as np
import tempfile
import logging

logging.basicConfig(level=logging.INFO)

torch.manual_seed(0)

# Create a minimal model with state_dict
class SimpleModel(nn.Module):
    def __init__(self):
        super(SimpleModel, self).__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create a test model
model = SimpleModel()

# Create a checkpoint with the required structure
checkpoint = {
    'state_dict': model.state_dict(),
    'best_value': 0.95,
    'epoch': 5
}

# Create a temporary checkpoint file
with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.pth') as f:
    pretrained_model_path = f.name

torch.save(checkpoint, pretrained_model_path)

ARGS = {
    'model': model,
    'pretrained_model': pretrained_model_path,
    'phase': 'train',
    'ismax': 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 tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_pretrained_optimizer Ran
ran on a synthesized input
lightaime/deep_gcns_torch/utils/ckpt_util.py
code served (permissive licence) · get_code("c845b58421ba1367")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 os
import torch
import torch.nn as nn
from torch.optim import Adam
from torch.optim.lr_scheduler import StepLR
import tempfile

torch.manual_seed(0)

# Create a minimal model
class TinyModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

model = TinyModel()

# Create optimizer and scheduler
optimizer = Adam(model.parameters(), lr=0.001)
scheduler = StepLR(optimizer, step_size=1, gamma=0.1)

# Create a temporary checkpoint file with optimizer and scheduler state
checkpoint = {
    'optimizer_state_dict': optimizer.state_dict(),
    'scheduler_state_dict': scheduler.state_dict(),
}

# Write checkpoint to a temporary file
with tempfile.NamedTemporaryFile(mode='wb', delete=False, suffix='.pt') as f:
    torch.save(checkpoint, f.name)
    checkpoint_path = f.name

# Prepare arguments
ARGS = {
    'pretrained_model': checkpoint_path,
    'optimizer': optimizer,
    'scheduler': scheduler,
    'lr': 0.001,
    'use_ckpt_lr': True,
}
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 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
norm_layer Ran
ran; our contract draft was wrong, not the code
lightaime/deep_gcns_torch/gcn_lib/dense/torch_nn.py
pointer only (licence: MIT); the stored copy get_code serves, from gswycf/signgraph, is pointer only (licence: NONE) · get_code("09e6ddb17c0adf28")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from gswycf/signgraph, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

norm_layer Ran
ran on a synthesized input
lightaime/deep_gcns_torch/gcn_lib/sparse/torch_nn.py
code served (permissive licence) · get_code("34c1fc07af3538f8")
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 BatchNorm1d. 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)

# The function takes two parameters: norm_type and nc
# norm_type: a string specifying the normalization type ('batch', 'layer', or 'instance')
# nc: an integer specifying the number of channels/features

ARGS = {
    'norm_type': 'batch',
    'nc': 4
}
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 BatchNorm1d — a BatchNorm1d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a BatchNorm1d 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
pairwise_distance Ran
ran on a synthesized input
lightaime/deep_gcns_torch/gcn_lib/dense/torch_edge.py
code served (permissive licence) · get_code("bc7cc66599be9207")
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 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 tensor with shape (batch_size, num_points, num_dims)
# Using distinct sizes: batch_size=2, num_points=4, num_dims=3
x = torch.randn(2, 4, 3)

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 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

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 8 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 0 the same month, 8 later) and 3 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
norm_layer 5 other papers
1 dated earlier, 4 later
2010.09891 (2020-10)
Robust Optimization as Data Augmentation for Large-scale Graphs
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction
2206.00272 (2022-06)
Vision GNN: An Image is Worth Graph of Nodes
and 2 more not named here
act_layer 3 other papers
1 dated earlier, 2 later
2010.09891 (2020-10)
Robust Optimization as Data Augmentation for Large-scale Graphs
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction
2304.07221 (2023-04)
Instance-aware Dynamic Prompt Tuning for Pre-trained Point Cloud Models
pairwise_distance 3 other papers
1 dated earlier, 2 later
2010.09891 (2020-10)
Robust Optimization as Data Augmentation for Large-scale Graphs
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction
aaai_28607 (2024 (year only))
Enhance Sketch Recognition's Explainability via Semantic Component-Level Parsing
knn_graph_matrix 2 other papers
1 dated earlier, 1 later
2010.09891 (2020-10)
Robust Optimization as Data Augmentation for Large-scale Graphs
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction
load_pretrained_models 2 other papers
1 dated earlier, 1 later
2010.09891 (2020-10)
Robust Optimization as Data Augmentation for Large-scale Graphs
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction
load_pretrained_optimizer 2 other papers
1 dated earlier, 1 later
2010.09891 (2020-10)
Robust Optimization as Data Augmentation for Large-scale Graphs
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction
act_layer 1 other paper
1 later
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction
batched_index_select 1 other paper
1 later
2111.00064 (2021-11)
Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction

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

Deep graph neural networks (GNNs) have achieved excellent results on various tasks on increasingly large graph datasets with millions of nodes and edges. However, memory complexity has become a major obstacle when training deep GNNs for practical applications due to the immense number of nodes, edges, and intermediate activations. To improve the scalability of GNNs, prior works propose smart graph sampling or partitioning strategies to train GNNs with a smaller set of nodes or sub-graphs. In this work, we study reversible connections, group convolutions, weight tying, and equilibrium models to advance the memory and parameter efficiency of GNNs. We find that reversible connections in combination with deep network architectures enable the training of overparameterized GNNs that significantly outperform existing methods on multiple datasets. Our models RevGNN-Deep (1001 layers with 80 channels each) and RevGNN-Wide (448 layers with 224 channels each) were both trained on a single commodity GPU and achieve an ROC-AUC of 87.74 ± 0.13 and 88.24 ± 0.15 on the ogbn-proteins dataset. To the best of our knowledge, RevGNN-Deep is the deepest GNN in the literature by one order of magnitude.

For agents

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

get_harvested_code_for_paper("2106.07476")
get_code_for_paper("2106.07476")
have("2106.07476")

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

curl https://syntology.ai/api/ran/2106.07476.json

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

[![Syntology run record](https://syntology.ai/api/ran/2106.07476.svg)](https://syntology.ai/paper/2106.07476)

Connect an agent — have() is free.