SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2504.13752 · 2025

Learning to Attribute with Attention

arXiv · PDF · Open in the Atlas

Code that ran

3 of 3 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, 1 violated it and 0 have no contract verdict; 2 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/2504.13752.json · badge: /api/ran/2504.13752.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
madrylab/at2 canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
FunctionStatusWhere it lives
filter_fn Ran
ran, violated the contract we drafted
madrylab/at2/scripts/train_context_at2.py
code served (permissive licence) · get_code("29aa585e05dc1492")
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 bool, 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

# Construct a valid example dict that will be passed to filter_fn
example = {
    'category': 'summarization',
    'context': 'This is a sample context string that is well under 20000 characters in length.'
}

# Define ARGS mapping the parameter to the value
ARGS = {
    'example': example
}
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 bool, 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_position_ids_and_attention_mask Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
madrylab/at2/at2/attribution/attention.py
code served (permissive licence) · get_code("a9f502939a0ac7a1")
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 as ch
import torch.nn as nn

ch.manual_seed(0)

class MockModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.dtype = ch.float32

model = MockModel()

batch_size = 2
seq_len = 5
hidden_dim = 4

hidden_states = (ch.randn(batch_size, seq_len, hidden_dim, dtype=ch.float32),)

ARGS = {
    "model": model,
    "hidden_states": hidden_states,
}
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
infer_model_type Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
madrylab/at2/at2/attribution/attention.py
code served (permissive licence) · get_code("1463e0c52d2adb27")
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 str. 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 transformers import PreTrainedModel
from transformers.configuration_utils import PretrainedConfig

torch.manual_seed(0)

# Create a minimal mock model with a name_or_path attribute
class MockModel:
    def __init__(self, name_or_path):
        self.name_or_path = name_or_path

# Test with a model name containing 'llama'
model = MockModel(name_or_path="meta-llama/Llama-2-7b")

ARGS = {
    'model': model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 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 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.

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

Given a sequence of tokens generated by a language model, we may want to identify the preceding tokens that influence the model to generate this sequence. Performing such token attribution is expensive; a common approach is to ablate preceding tokens and directly measure their effects. To reduce the cost of token attribution, we revisit attention weights as a heuristic for how a language model uses previous tokens. Naive approaches to attribute model behavior with attention (e.g., averaging attention weights across attention heads to estimate a token's influence) have been found to be unreliable. To attain faithful attributions, we propose treating the attention weights of different attention heads as features. This way, we can learn how to effectively leverage attention weights for attribution (using signal from ablations). Our resulting method, Attribution with Attention (AT2), reliably performs on par with approaches that involve many ablations, while being significantly more efficient. To showcase the utility of AT2, we use it to prune less important parts of a provided context in a question answering setting, improving answer quality. We provide code for AT2 at https://github.com/MadryLab/AT2 .

For agents

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

get_harvested_code_for_paper("2504.13752")
get_code_for_paper("2504.13752")
have("2504.13752")

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

curl https://syntology.ai/api/ran/2504.13752.json

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

[![Syntology run record](https://syntology.ai/api/ran/2504.13752.svg)](https://syntology.ai/paper/2504.13752)

Connect an agent — have() is free.