SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.01008 · ICML · 2024

On Mechanistic Knowledge Localization in Text-to-Image Generative Models

Soheil Feizi, Samyadeep Basu, Priyatham Kattakinda, Vlad Morariu, Ryan Rossi, Varun Manjunatha, Keivan Rezaei, Cherry Zhao

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, 0 violated it and 0 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/2405.01008.json · badge: /api/ran/2405.01008.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
samyadeepbasu/locogen canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
FunctionStatusWhere it lives
get_module Ran
ran; our contract draft was wrong, not the code
samyadeepbasu/locogen/trace_attn_layer.py
code served (permissive licence) · get_code("8953f9ef90cf6afd")
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) and returned Linear. 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

torch.manual_seed(0)

# Create a simple model with named modules
model = nn.Sequential(
    nn.Linear(4, 8),
    nn.ReLU(),
    nn.Linear(8, 3)
)

# The function uses model.named_modules() to iterate through modules
# and compares the name 'n' with the 'name' parameter
# We need to provide a valid module name that exists in the model

# Get a valid module name from the model
valid_name = None
for n, m in model.named_modules():
    if n == "0":  # First Linear layer
        valid_name = n
        break

ARGS = {
    "model": model,
    "name": "0"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Linear — a Linear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Linear 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
high_level_layers Ran
ran; our contract draft was wrong, not the code
samyadeepbasu/locogen/loco_edit.py
code served (permissive licence) · get_code("a99640ca484b5055")
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) and returned list, 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 torch
import torch.nn as nn

torch.manual_seed(0)

class AttentionModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.to_k = nn.Linear(4, 4)
        self.to_v = nn.Linear(4, 4)

class UNetBlock(nn.Module):
    def __init__(self):
        super().__init__()
        self.attn2 = AttentionModule()

class SimpleUNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.down_blocks = nn.ModuleList([
            UNetBlock(),
            UNetBlock()
        ])
        self.mid_block = UNetBlock()
        self.up_blocks = nn.ModuleList([
            UNetBlock(),
            UNetBlock()
        ])

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.unet = SimpleUNet()

model = SimpleModel()

ARGS = {
    "model": model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list 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
init_latent Ran
ran; our contract draft was wrong, not the code
samyadeepbasu/locogen/trace_attn_layer.py
code served (permissive licence) · get_code("caadc5c4b7627e0c")
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) 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
import torch.nn as nn

torch.manual_seed(0)

# Build a minimal model with a unet that has in_channels and device
class MinimalUNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.in_channels = 4
        self.linear = nn.Linear(10, 10)  # dummy layer to have parameters
    
    def forward(self, x):
        return x

class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.unet = MinimalUNet()
        self.device = torch.device("cpu")
    
    def forward(self, x):
        return x

# Create instances
model = MinimalModel()
generator = torch.Generator()
generator.manual_seed(42)

# Parameters for init_latent
latent = None  # Will be generated inside the function
height = 32
width = 24
batch_size = 2

ARGS = {
    "latent": latent,
    "model": model,
    "height": height,
    "width": width,
    "generator": generator,
    "batch_size": batch_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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

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, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 1 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
get_module 5 other papers
2 dated earlier, 1 the same month, 2 later · 1 only as an assembly brick
2309.15098 (2023-09)
Attention Satisfies: A Constraint-Satisfaction Lens on Factual Errors of Language Models
2310.15213 (2023-10)
Function Vectors in Large Language Models
2405.15349 (2024-05)
Everything is Editable: Extend Knowledge Editing to Unstructured Data in Large Language Models
and 2 more not named here
init_latent 3 other papers
3 dated earlier
2305.15581 (2023-05)
Unsupervised Semantic Correspondence Using Stable Diffusion
2305.18729 (2023-05)
Real-World Image Variation by Aligning Diffusion Inversion Chain
2305.19164 (2023-05)
LANCE: Stress-testing Visual Models by Generating Language-guided Counterfactual Images

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

Identifying layers within text-to-image models which control visual attributes can facilitate efficient model editing through closed-form updates. Recent work, leveraging causal tracing show that early Stable-Diffusion variants confine knowledge primarily to the first layer of the CLIP text-encoder, while it diffuses throughout the UNet. Extending this framework, we observe that for recent models (e.g., SD-XL, Deep-Floyd), causal tracing fails in pinpointing localized knowledge, highlighting challenges in model editing. To address this issue, we introduce the concept of mechanistic localization in text-toimage models, where knowledge about various visual attributes (e.g., "style", "objects", "facts") can be mechanistically localized to a small fraction of layers in the UNet, thus facilitating efficient model editing. We localize knowledge using our method LOCOGEN which measures the direct effect of intermediate layers to output generation by performing interventions in the cross-attention layers of the UNet. We then employ LOCOEDIT, a fast closed-form editing method across popular open-source text-to-image models (including the latest SD-XL) and explore the possibilities of neuron-level model editing. Using mechanistic localization, our work offers a better view of successes and failures in localization-based textto-image model editing. Code will be available at https://github.com/samyadeepbasu/LocoGen.

For agents

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

get_harvested_code_for_paper("2405.01008")
get_code_for_paper("2405.01008")
have("2405.01008")

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

curl https://syntology.ai/api/ran/2405.01008.json

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

[![Syntology run record](https://syntology.ai/api/ran/2405.01008.svg)](https://syntology.ai/paper/2405.01008)

Connect an agent — have() is free.