OpenSAE/SAEBase-Evaluation

SAEBase Evaluation

Versioned measurement records for sparse autoencoder research. The original 48-site software verification tables and the later 500-prompt Research Lab experiment are separate evidence populations.

A unified research dataset for evaluating sparse autoencoders, tracing feature effects, and studying representation structure in language models.

SAEBase Evaluation brings multiple forms of evidence into one Hugging Face dataset: model–SAE compatibility, reconstruction and sparsity metrics, feature interventions, controlled inputs, and research measurements from SAR, RSS, and Input Localization. Each record is designed to connect a numerical result to a specific model, SAE checkpoint, layer, input population, and experimental protocol.

Explore the data · SAEBase Research Lab · GitHub · v0.4.0

At a glance

Collection	Scale	What it helps you study
Model & SAE evaluation	18 model–SAE configurations, 48 layer sites	Sparsity, reconstruction, intervention response, checkpoint compatibility
Historical benchmark	64 held-out controlled prompt groups per site	Directly reproducible historical metrics and reliability records
Research Lab input cohort	10 sources × 50 parent groups, 200 discovery / 300 evaluation	Source-pinned input coverage and independent evaluation splits
Feature and intervention records	Up to 300 deep-site feature cards; six intervention sites	Feature activity and recorded intervention effects when underlying sample records are published
SAE Additivity Residuals (SAR)	1,500 reported evaluated prompt groups across K16/K8/K4	How well individual feature effects predict their joint downstream effects
Redistribution Sensitivity Score (RSS)	Six sites, 3,000 site–parent groups, 138,000 reported logical input occurrences	How prompt transformations redistribute SAE feature activations
Input Localization	Two six-factor designs × 64 states; one measured real-model grid	How feature responses depend on controlled input factors

The historical 64-group benchmark and the newer 500-group Research Lab cohort are separate experiments, not interchangeable samples. Research Lab aggregate counts and downloadable per-sample files are distinguished in the corresponding subsets.

Start in three lines

from datasets import load_dataset

metrics = load_dataset("OpenSAE/SAEBase-Evaluation", "metric_results", split="train")
print(metrics.select(range(3)).to_pandas().to_string(index=False))

Install: pip install -U datasets pandas.

The metric_results subset is the simplest entry point. One row = one metric measured at one model–SAE layer site. Its value column is a number, metric_id identifies what was measured, n_valid gives the population denominator, and source_report_url points to the original experiment record.

Which subset should I use?

All configurations below live inside this single Dataset repository. The following table is populated from the configurations declared in the live dataset card when this documentation is uploaded.

Subset / configuration	Area	Contents
verified_sites	Registry	48 historically verified model/SAE layer sites
metric_results	Benchmark	Individual L0, reconstruction, explained-variance and ablation-logit metrics
research_lab_summary	Research overview	Measurements reported by the Research Lab executor
sar_cohorts	SAR	K16/K8/K4 cohort counts and coalition configurations
rss_coverage	RSS	Distinct input coverage, selection/held-out counts and status
input_localization	Localization	Observed-support exact controlled-grid scope result
model_sae_registry	Registry	Model/SAE layer and hook registry
prompt_manifest	Inputs	500 frozen source-group identifiers and splits; no third-party prompt text
source_revisions	Inputs	Pinned upstream dataset revisions and source digests
controlled_inputs	Localization	Complete project-authored six-factor input texts
model_sae_profiles	Registry	One row per model–SAE configuration
source_split_counts	Inputs	Input counts by source and discovery/evaluation split
controlled_designs	Localization	Design identities and 64-state experiment definitions
historical_site_metrics	Benchmark	One wide metric row per verified historical site
historical_full_metrics	Benchmark	Detailed historical metric rows, including the identity intervention
historical_acceptance_probes	Validation	Individual real-checkpoint probe results
checkpoint_identities	Registry	Recorded checkpoint digests, hooks and runtime identity
historical_model_aggregates	Benchmark	Per-model/SAE summaries across selected historical layers
gpu_sar_per_prompt	SAR	Measured prompt-level SAR values and response/checksum references
gpu_rss_per_setting	RSS	Measured setting-level RSS and held-out persistence records
gpu_feature_cards	Features	Evidence-linked feature activity cards
gpu_intervention_pairs	Interventions	Real clean/edited model-output comparisons
gpu_site_capture_ledger	Validation	Per-site capture status and recorded model/checkpoint hashes
gpu_new_checkpoint_identities	Registry	Verified new-run identities distinct from historical ones
gpu_localization_records	Localization	Controlled-grid metric and numerical diagnostic records

Some supporting JSON and CSV files are also browsable from the Files tab. The available downloadable research resources are listed below when detected in the repository.

File	What it contains
prompt_manifest.csv	Frozen input hashes and group IDs
sources.csv	Original datasets and pinned revisions
sites.csv	Layer-site registry
controlled_inputs.csv	Project-authored 64-state designs
sar_cohorts.csv	SAR cohort summary
rss_coverage.csv	RSS coverage summary
localization.csv	Localization summary
sar_prompt_metrics.csv	Measured SAR prompt records
rss_group_settings.csv	Measured RSS group/setting records
feature_cards.csv	Feature-level evidence cards
intervention_pairs.csv	Observed intervention pairs
all_metric_results.csv	All source-pinned historical metric rows
acceptance_probes.csv	Checkpoint acceptance probes

Understanding the columns

Column	Meaning
model_id, sae_id, layer	Which pretrained language model and SAE dictionary were measured, and at which layer
site_id, hook_name	Exact model/SAE intervention location; check the schema before joining tables
metric_id, condition, value	Metric name, experimental edit/condition (if any), and numerical result
n_valid, status	Measurement denominator and whether the result is available or defined
sample_id, prompt_group_id	Input identity and the independent group to which variants belong
usage_split	Discovery or evaluation assignment; transformed versions inherit their parent group
source_commit, source_report_url, *_sha256	Code and data provenance and links for verification
protocol, k, M, n_fit, n_held	Scientific method and its coalition/sample configuration for SAR
selection_variant_ids, held_out_variant_ids	RSS variants used for selection and independent evaluation when raw records are present

A row is not always an independent observation. The statistical unit is the parent prompt group for natural-input studies and the complete 64-state grid for each controlled Localize design.

Practical examples

1. Find the most reconstructable layer in one SAE

from datasets import load_dataset

D = load_dataset("OpenSAE/SAEBase-Evaluation", "metric_results", split="train").to_pandas()
model = "gemma270m"
sae = "gemma270m_scope_resid"

layers = D[(D.model_id == model) & (D.sae_id == sae)]
metrics = layers.pivot_table(index="layer", columns="metric_id", values="value", aggfunc="first")
print(metrics[["core.l0", "core.reconstruction_mse", "core.explained_variance"]].sort_index())

Interpretation: smaller reconstruction MSE describes better recovery under that recorded activation scale; explained variance is normalized to a variance denominator. Compare layers within a compatible protocol before comparing arbitrary numbers from different models or hooks. See the complete benchmark example and the scripts in this card's companion ZIP.

2. Examine the effect of feature ablation

from datasets import load_dataset

D = load_dataset("OpenSAE/SAEBase-Evaluation", "metric_results", split="train").to_pandas()
ablations = D[(D.metric_id == "intervention.logit_delta_l2") & (D.condition == "feature-ablation")]
print(ablations[["model_id", "sae_id", "layer", "value", "n_valid"]].head(10))

Here, value is an L2 change in the model's output logits caused by the recorded edit. It indicates a downstream effect, not by itself a proof that the selected feature has a unique human-readable meaning.

3. Retrieve the 500-prompt source manifest

If your version of the Dataset includes the prompt_manifest subset:

from datasets import load_dataset

prompts = load_dataset("OpenSAE/SAEBase-Evaluation", "prompt_manifest", split="train").to_pandas()
print(prompts.groupby(["dataset_id", "usage_split"]).size().unstack(fill_value=0))

The manifest contains identifiers, grouping, upstream revisions and text digests, not necessarily the original copyrighted prompt text. This makes it useful for reproducing dataset selection, checking leakage and comparing runs on the same inputs.

4. Work with SAR, RSS and Localize results

from datasets import load_dataset

repo = "OpenSAE/SAEBase-Evaluation"

# Study populations and the selected coalition configurations.
sar = load_dataset(repo, "sar_cohorts", split="train").to_pandas()
print(sar[["k", "m", "fit", "held", "successful_groups"]])

# Counts in these tables are measurements of coverage, not per-prompt effect sizes.
rss = load_dataset(repo, "rss_coverage", split="train").to_pandas()
print(rss[["parent_site_groups", "unique_site_input_occurrences"]])

loc = load_dataset(repo, "input_localization", split="train").to_pandas()
print(loc[["site_id", "states", "observed_support_features", "budget_at_90pct"]])

SAR: A coalition is a selected subset of SAE features. The original SAR estimator fits and evaluates an additive approximation to measured downstream responses. The published Research Lab run explicitly uses a decoder-mediated, residual-preserving intervention protocol, and separates K16, K8 and K4 cohorts. The sar_cohorts table reports group counts, not the numerical SAR residual for each prompt. When provided, gpu_sar_per_prompt contains the actual per-prompt estimates and checksums of their underlying response vectors.

RSS: The study compares changes in SAE activations under prompt transformations. rss_coverage gives logical site–input coverage, not independent observations or an empirical mean RSS value. When provided, gpu_rss_per_setting contains the support, direction-related summary and held-out rate for actual measured settings. Proposed transformations have not all been certified to preserve task meaning.

Localization: A six-factor experiment enumerates all (2^6 = 64) input states, then examines how much of the resulting feature variation can be represented by limited subsets of input factors. controlled_inputs contains project-authored states; input_localization is a separate real model result. The recorded budget_at_90pct follows the method's exact scope-cost definition; do not interpret it as simply a count of nonzero SAE dictionary features.

For richer per-record analysis, use these subsets if shown in the available-configurations table: historical_full_metrics, historical_acceptance_probes, gpu_sar_per_prompt, gpu_rss_per_setting, gpu_feature_cards, gpu_intervention_pairs, and gpu_localization_records. Aggregate summaries must not be treated as substitutes for unuploaded per-prompt measurements.

Research Lab experiments (v0.4.0, separate cohort)

SAEBase also reports a separate, frozen 500-prompt research cohort for feature activations, interventions, and further research analyses. Research Lab offers interactive browsing of recorded measurements, and the GitHub evidence summary documents the experiments and their scope.

This is not the same population as the 64-group-per-site v0.3.0 verification tables. Only files actually published in this dataset repository should be treated as downloadable dataset assets; a recorded GPU experiment does not imply its complete private raw archive is publicly released. Keep the cohorts and their source manifests separate.

How the dataset is organized

The collection connects different kinds of observations without merging incompatible protocols:

SAEBase-Evaluation (one dataset repository)
 ├── Historical evaluation      model/SAE/layer × metric, source-linked
 ├── Model and site registry    checkpoint identities, hooks, layers
 ├── Input manifests           source revisions, groups and splits
 ├── Controlled designs        64-state project-authored input grids
 ├── Feature & interventions   recorded codes/edits, where exported
 ├── SAR / RSS                  measured group/setting metrics, where exported
 └── Localization              controlled factor-scope evaluation

Use (model_id, sae_id, layer) to connect historical metrics to the newer registry. In particular, some older site_id values abbreviate the model/layer, while the Research Lab site IDs also include the SAE ID; do not assume they are directly equal as strings.

Data provenance: the historical Stage D reports are sourced from the exact verified checkpoint matrix and original raw run archives. The newer Research Lab results are documented in the v0.4.0 experiment report with the relevant source releases, model identities and measurement protocols.

Use it for your research

This dataset is designed for three practical workflows:

• Compare reconstruction, sparsity and measured feature effects across layers or SAE checkpoints while preserving protocol identities.
• Audit and reproduce an existing result by tracing a table row back to its source report, source code revision and checkpoint digest.
• Develop new interpretability analyses using SAR, RSS, controlled inputs and feature-level records, when the needed per-sample arrays are available in the dataset or in referenced experiment archives.

For live exploration or to try the related code, use SAEBase Research Lab. This dataset publishes research records and inputs, not pretrained SAE weight binaries or an online inference service.

Citation

If you use these records, cite the SAEBase software version and original model/SAE sources, along with the relevant SAR, RSS or Localize paper for the corresponding method. Repository and release history · SAEBase v0.4.0.

---

_{Maintained by OpenSAE. Results retain their original experimental context and upstream source identities.}