SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2108.12284 · EMNLP · 2021

The Devil is in the Detail: Simple Tricks Improve Systematic Generalization of Transformers

Jürgen Schmidhuber, Kazuki Irie, Róbert Csordás

arXiv · PDF · Open in the Atlas

Code that ran

1 of 2 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 0 have no contract verdict; 1 failure of our instrument; 1 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/2108.12284.json · badge: /api/ran/2108.12284.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
robertcsordas/transformer_generalization canonical (classifier label)
classifier basis: census
1 ran · 1 no recorded run
FunctionStatusWhere it lives
TransformerEncoderWithLayer Ran
ran; our contract draft was wrong, not the code
robertcsordas/transformer_generalization/layers/transformer/transformer.py
code served (permissive licence) · get_code("8bbe7d0623460d7d")
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 torch
import torch.nn as nn
import torch.nn.functional as F
from typing import Optional, Callable

torch.manual_seed(0)

# Define the missing classes that are used in the function
class AttentionMask:
    def __init__(self, attn_mask, key_padding_mask):
        self.attn_mask = attn_mask
        self.key_padding_mask = key_padding_mask

class MultiHeadAttention(nn.Module):
    def __init__(self, d_model, nhead, dropout=0.1):
        super().__init__()
        self.d_model = d_model
        self.nhead = nhead
        self.dropout = dropout
        self.linear_q = nn.Linear(d_model, d_model)
        self.linear_k = nn.Linear(d_model, d_model)
        self.linear_v = nn.Linear(d_model, d_model)
        self.linear_out = nn.Linear(d_model, d_model)
    
    def forward(self, query, key, attention_mask):
        # Simple stub implementation
        return query

class TransformerEncoderLayer(nn.Module):
    def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1, activation: Callable[[torch.Tensor], torch.Tensor]=F.relu):
        super(TransformerEncoderLayer, self).__init__()
        self.self_attn = MultiHeadAttention(d_model, nhead, dropout=dropout)
        self.linear1 = nn.Linear(d_model, dim_feedforward)
        self.dropout = nn.Dropout(dropout)
        self.linear2 = nn.Linear(dim_feedforward, d_model)
        self.norm1 = nn.LayerNorm(d_model)
        self.norm2 = nn.LayerNorm(d_model)
        self.dropout1 = nn.Dropout(dropout)
        self.dropout2 = nn.Dropout(dropout)
        self.activation = activation
        self.reset_parameters()

    def forward(self, src: torch.Tensor, mask: Optional[torch.Tensor]=None) -> torch.Tensor:
        src2 = self.self_attn(src, src, AttentionMask(mask, None))
        src = src + self.dropout1(src2)
        src = self.norm1(src)
        src2 = self.linear2(self.dropout(self.activation(self.linear1(src))))
        src = src + self.dropout2(src2)
        src = self.norm2(src)
        return src

    def reset_parameters(self):
        nn.init.xavier_uniform_(self.linear1.weight, gain=nn.init.calculate_gain('relu') if self.activation is F.relu else 1.0)
        nn.init.xavier_uniform_(self.linear2.weight)

class TransformerEncoder(nn.Module):
    def __init__(self, layer, n_layers: int, *args, **kwargs):
        super().__init__()
        self.layers = nn.ModuleList([layer(*args, **kwargs) for _ in range(n_layers)])

    def forward(self, data: torch.Tensor, *args, **kwargs):
        for l in self.layers:
            data = l(data, *args, **kwargs)
        return data

def TransformerEncoderWithLayer(layer=TransformerEncoderLayer):
    return lambda *args, **kwargs: TransformerEncoder(layer, *args, **kwargs)

# Setup arguments for TransformerEncoderWithLayer
ARGS = {
    'layer': TransformerEncoderLayer
}
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 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
TransformerDecoderWithLayer Not yet run
unverified (harvested, no recorded run)
robertcsordas/transformer_generalization/layers/transformer/transformer.py
code served (permissive licence) · get_code("e865d7a05203dc75")
How it ran

no invocation recorded.

Also present in other papers' code

Of 2 samples lifted from this paper's linked repositories, 2 carry a function identity and 0 do not. Of the 2 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 1 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
TransformerEncoderWithLayer 1 other paper
1 later
2310.19089 (2023-10)
Pushdown Layers: Encoding Recursive Structure in Transformer Language Models

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

Recently, many datasets have been proposed to test the systematic generalization ability of neural networks. The companion baseline Transformers, typically trained with default hyper-parameters from standard tasks, are shown to fail dramatically. Here we demonstrate that by revisiting model configurations as basic as scaling of embeddings, early stopping, relative positional embedding, and Universal Transformer variants, we can drastically improve the performance of Transformers on systematic generalization. We report improvements on five popular datasets: SCAN, CFQ, PCFG, COGS, and Mathematics dataset. Our models improve accuracy from 50% to 85% on the PCFG productivity split, and from 35% to 81% on COGS. On SCAN, relative positional embedding largely mitigates the EOS decision problem (Newman et al., 2020), yielding 100% accuracy on the length split with a cutoff at 26. Importantly, performance differences between these models are typically invisible on the IID data split. This calls for proper generalization validation sets for developing neural networks that generalize systematically. We publicly release the code to reproduce our results 1 .

For agents

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

get_harvested_code_for_paper("2108.12284")
get_code_for_paper("2108.12284")
have("2108.12284")

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

curl https://syntology.ai/api/ran/2108.12284.json

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

[![Syntology run record](https://syntology.ai/api/ran/2108.12284.svg)](https://syntology.ai/paper/2108.12284)

Connect an agent — have() is free.