SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.04185 · ICLR · 2025

Residual Stream Analysis with Multi-Layer SAEs

Laurence Aitchison, Tim Lawson, Lucy Farnik, Conor Houghton

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; 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/2409.04185.json · badge: /api/ran/2409.04185.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
tim-lawson/mlsae canonical (classifier label)
classifier basis: census
2 ran · 0 no recorded run
openai/sparse_autoencoder canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
LN Ran
ran; our contract draft was wrong, not the code
openai/sparse_autoencoder/sparse_autoencoder/model.py
code served (permissive licence) · get_code("e68981067b2bdefe")
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)

# Create a small tensor with shape (batch=2, features=4)
x = torch.randn(2, 4)

ARGS = {
    "x": x,
    "eps": 1e-05
}
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
layerwise Ran
ran; our contract draft was wrong, not the code
tim-lawson/mlsae/mlsae/metrics/layerwise.py
code served (permissive licence) · get_code("64b782e2d65d1c12")
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 partial. 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 functools import partial
from torchmetrics import Metric
import torch.nn as nn

torch.manual_seed(0)

# The layerwise function only takes n_layers as a parameter
# n_layers is used to create labels like ['layer_0', 'layer_1', ...]
# It returns a partial of LayerwiseWrapper

ARGS = {
    'n_layers': 4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jaxtyping, torch, torchmetrics — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type partial — a partial output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a partial 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
model_card_template Ran
ran; our contract draft was wrong, not the code
tim-lawson/mlsae/mlsae/model_card.py
code served (permissive licence) · get_code("8d5257ecf62287fb")
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
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "transformer": 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 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, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 later) and 2 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
LN 3 other papers
2 dated earlier, 1 later
2312.16682 (2023-12)
Some things are more CRINGE than others: Iterative Preference Optimization with the Pairwise Cringe Loss
2406.04093 (2024-06)
Scaling and evaluating sparse autoencoders
2411.07122 (2024-11)
SCAR: Sparse Conditioned Autoencoders for Concept Detection and Steering in LLMs

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

Sparse autoencoders (SAEs) are a promising approach to interpreting the internal representations of transformer language models. However, SAEs are usually trained separately on each transformer layer, making it difficult to use them to study how information flows across layers. To solve this problem, we introduce the multi-layer SAE (MLSAE): a single SAE trained on the residual stream activation vectors from every transformer layer. Given that the residual stream is understood to preserve information across layers, we expected MLSAE latents to 'switch on' at a token position and remain active at later layers. Interestingly, we find that individual latents are often active at a single layer for a given token or prompt, but the layer at which an individual latent is active may differ for different tokens or prompts. We quantify these phenomena by defining a distribution over layers and considering its variance. We find that the variance of the distributions of latent activations over layers is about two orders of magnitude greater when aggregating over tokens compared with a single token. For larger underlying models, the degree to which latents are active at multiple layers increases, which is consistent with the fact that the residual stream activation vectors at adjacent layers become more similar. Finally, we relax the assumption that the residual stream basis is the same at every layer by applying pre-trained tuned-lens transformations, but our findings remain qualitatively similar. Our results represent a new approach to understanding how representations change as they flow through transformers. We release our code to train and analyze MLSAEs at https://github.com/tim-lawson/mlsae.

For agents

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

get_harvested_code_for_paper("2409.04185")
get_code_for_paper("2409.04185")
have("2409.04185")

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

curl https://syntology.ai/api/ran/2409.04185.json

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

[![Syntology run record](https://syntology.ai/api/ran/2409.04185.svg)](https://syntology.ai/paper/2409.04185)

Connect an agent — have() is free.