Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
graphs: list<item: struct<slug: string, scan: string, transcoder_list: list<item: null>, prompt_tokens: list (... 619 chars omitted)
child 0, item: struct<slug: string, scan: string, transcoder_list: list<item: null>, prompt_tokens: list<item: stri (... 607 chars omitted)
child 0, slug: string
child 1, scan: string
child 2, transcoder_list: list<item: null>
child 0, item: null
child 3, prompt_tokens: list<item: string>
child 0, item: string
child 4, prompt: string
child 5, node_threshold: double
child 6, schema_version: int64
child 7, node_counts: struct<base_features: int64, adapter_features: int64, error_nodes: int64>
child 0, base_features: int64
child 1, adapter_features: int64
child 2, error_nodes: int64
child 8, comparison: struct<base_slug: string, adapter_slug: string, base_scan: string, adapter_scan: string, base_featur (... 338 chars omitted)
child 0, base_slug: string
child 1, adapter_slug: string
child 2, base_scan: string
child 3, adapter_scan: string
child 4, base_feature_scan: string
child 5, adapter_feature_scan: string
child 6, node_shapes: struct<base: string, adapter: string, shared: string>
child 0, base: string
child 1, adapter: string
child 2, shared: string
child 7, node_counts: struct<base_features: int64, adapter_features: int6
...
child 59, harm_034__pos09: string
child 60, harm_035__pos00: string
child 61, harm_035__pos01: string
child 62, harm_035__pos02: string
child 63, harm_035__pos03: string
child 64, harm_035__pos04: string
child 65, harm_035__pos05: string
child 66, harm_035__pos06: string
child 67, harm_035__pos07: string
child 68, harm_035__pos08: string
child 69, harm_035__pos09: string
child 70, harm_094__pos00: string
child 71, harm_094__pos01: string
child 72, harm_094__pos02: string
child 73, harm_094__pos03: string
child 74, harm_094__pos04: string
child 75, harm_094__pos05: string
child 76, harm_094__pos06: string
child 77, harm_094__pos07: string
child 78, harm_094__pos08: string
child 79, harm_094__pos09: string
later_todo: list<item: string>
child 0, item: string
base_model: string
run_name: string
overlay_graph_dir: string
compact_overlay_graph_paths: list<item: string>
child 0, item: string
prompt_format: string
gemmascope_config_path: string
overlay_options: struct<hide_direct_embedding_logit_links: bool, compact_max_base_feature_nodes: int64, compact_max_b (... 64 chars omitted)
child 0, hide_direct_embedding_logit_links: bool
child 1, compact_max_base_feature_nodes: int64
child 2, compact_max_base_error_nodes: int64
child 3, compact_max_adapter_feature_nodes: null
output_dir: string
compact_overlay_graph_dir: string
adapter_graph_dir: string
to
{'adapter_checkpoint': Value('string'), 'base_model': Value('string'), 'prompt_tokenizer_model': Value('string'), 'prompts': Value('string'), 'run_name': Value('string'), 'prompt_format': Value('string'), 'base_attribution_targets': Value('string'), 'output_dir': Value('string'), 'base_graph_dir': Value('string'), 'adapter_graph_dir': Value('string'), 'overlay_graph_dir': Value('string'), 'overlay_graph_paths': List(Value('string')), 'compact_overlay_graph_dir': Value('string'), 'compact_overlay_graph_paths': List(Value('string')), 'gemmascope_config_path': Value('string'), 'gemmascope': {'repo': Value('string'), 'width': Value('string'), 'requested_l0': Value('string'), 'l0_match': Value('string'), 'l0_by_layer': List(Value('string')), 'n_layers': Value('int64'), 'scan': Value('string')}, 'adapter_features': {'feature_data_path': Value('string'), 'feature_output_dir': Value('null'), 'scan': Value('string'), 'features_dir': Value('null')}, 'base_features': {'feature_data_path': Value('string'), 'scan': Value('string'), 'features_dir': Value('null')}, 'overlay_options': {'hide_direct_embedding_logit_links': Value('bool'), 'compact_max_base_feature_nodes': Value('int64'), 'compact_max_base_error_nodes': Value('int64'), 'compact_max_adapter_feature_nodes': Value('null')}, 'results': {'base': {'benign_egg__pos00': Value('string'), 'benign_egg__pos01': Value('string'), 'benign_egg__pos02': Value('string'), 'benign_egg__pos03': Value('string'), 'benign_egg__pos04': Value('string'),
...
09': Value('string'), 'harm_031__pos00': Value('string'), 'harm_031__pos01': Value('string'), 'harm_031__pos02': Value('string'), 'harm_031__pos03': Value('string'), 'harm_031__pos04': Value('string'), 'harm_031__pos05': Value('string'), 'harm_031__pos06': Value('string'), 'harm_031__pos07': Value('string'), 'harm_031__pos08': Value('string'), 'harm_031__pos09': Value('string'), 'harm_034__pos00': Value('string'), 'harm_034__pos01': Value('string'), 'harm_034__pos02': Value('string'), 'harm_034__pos03': Value('string'), 'harm_034__pos04': Value('string'), 'harm_034__pos05': Value('string'), 'harm_034__pos06': Value('string'), 'harm_034__pos07': Value('string'), 'harm_034__pos08': Value('string'), 'harm_034__pos09': Value('string'), 'harm_035__pos00': Value('string'), 'harm_035__pos01': Value('string'), 'harm_035__pos02': Value('string'), 'harm_035__pos03': Value('string'), 'harm_035__pos04': Value('string'), 'harm_035__pos05': Value('string'), 'harm_035__pos06': Value('string'), 'harm_035__pos07': Value('string'), 'harm_035__pos08': Value('string'), 'harm_035__pos09': Value('string'), 'harm_094__pos00': Value('string'), 'harm_094__pos01': Value('string'), 'harm_094__pos02': Value('string'), 'harm_094__pos03': Value('string'), 'harm_094__pos04': Value('string'), 'harm_094__pos05': Value('string'), 'harm_094__pos06': Value('string'), 'harm_094__pos07': Value('string'), 'harm_094__pos08': Value('string'), 'harm_094__pos09': Value('string')}}, 'later_todo': List(Value('string'))}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
graphs: list<item: struct<slug: string, scan: string, transcoder_list: list<item: null>, prompt_tokens: list (... 619 chars omitted)
child 0, item: struct<slug: string, scan: string, transcoder_list: list<item: null>, prompt_tokens: list<item: stri (... 607 chars omitted)
child 0, slug: string
child 1, scan: string
child 2, transcoder_list: list<item: null>
child 0, item: null
child 3, prompt_tokens: list<item: string>
child 0, item: string
child 4, prompt: string
child 5, node_threshold: double
child 6, schema_version: int64
child 7, node_counts: struct<base_features: int64, adapter_features: int64, error_nodes: int64>
child 0, base_features: int64
child 1, adapter_features: int64
child 2, error_nodes: int64
child 8, comparison: struct<base_slug: string, adapter_slug: string, base_scan: string, adapter_scan: string, base_featur (... 338 chars omitted)
child 0, base_slug: string
child 1, adapter_slug: string
child 2, base_scan: string
child 3, adapter_scan: string
child 4, base_feature_scan: string
child 5, adapter_feature_scan: string
child 6, node_shapes: struct<base: string, adapter: string, shared: string>
child 0, base: string
child 1, adapter: string
child 2, shared: string
child 7, node_counts: struct<base_features: int64, adapter_features: int6
...
child 59, harm_034__pos09: string
child 60, harm_035__pos00: string
child 61, harm_035__pos01: string
child 62, harm_035__pos02: string
child 63, harm_035__pos03: string
child 64, harm_035__pos04: string
child 65, harm_035__pos05: string
child 66, harm_035__pos06: string
child 67, harm_035__pos07: string
child 68, harm_035__pos08: string
child 69, harm_035__pos09: string
child 70, harm_094__pos00: string
child 71, harm_094__pos01: string
child 72, harm_094__pos02: string
child 73, harm_094__pos03: string
child 74, harm_094__pos04: string
child 75, harm_094__pos05: string
child 76, harm_094__pos06: string
child 77, harm_094__pos07: string
child 78, harm_094__pos08: string
child 79, harm_094__pos09: string
later_todo: list<item: string>
child 0, item: string
base_model: string
run_name: string
overlay_graph_dir: string
compact_overlay_graph_paths: list<item: string>
child 0, item: string
prompt_format: string
gemmascope_config_path: string
overlay_options: struct<hide_direct_embedding_logit_links: bool, compact_max_base_feature_nodes: int64, compact_max_b (... 64 chars omitted)
child 0, hide_direct_embedding_logit_links: bool
child 1, compact_max_base_feature_nodes: int64
child 2, compact_max_base_error_nodes: int64
child 3, compact_max_adapter_feature_nodes: null
output_dir: string
compact_overlay_graph_dir: string
adapter_graph_dir: string
to
{'adapter_checkpoint': Value('string'), 'base_model': Value('string'), 'prompt_tokenizer_model': Value('string'), 'prompts': Value('string'), 'run_name': Value('string'), 'prompt_format': Value('string'), 'base_attribution_targets': Value('string'), 'output_dir': Value('string'), 'base_graph_dir': Value('string'), 'adapter_graph_dir': Value('string'), 'overlay_graph_dir': Value('string'), 'overlay_graph_paths': List(Value('string')), 'compact_overlay_graph_dir': Value('string'), 'compact_overlay_graph_paths': List(Value('string')), 'gemmascope_config_path': Value('string'), 'gemmascope': {'repo': Value('string'), 'width': Value('string'), 'requested_l0': Value('string'), 'l0_match': Value('string'), 'l0_by_layer': List(Value('string')), 'n_layers': Value('int64'), 'scan': Value('string')}, 'adapter_features': {'feature_data_path': Value('string'), 'feature_output_dir': Value('null'), 'scan': Value('string'), 'features_dir': Value('null')}, 'base_features': {'feature_data_path': Value('string'), 'scan': Value('string'), 'features_dir': Value('null')}, 'overlay_options': {'hide_direct_embedding_logit_links': Value('bool'), 'compact_max_base_feature_nodes': Value('int64'), 'compact_max_base_error_nodes': Value('int64'), 'compact_max_adapter_feature_nodes': Value('null')}, 'results': {'base': {'benign_egg__pos00': Value('string'), 'benign_egg__pos01': Value('string'), 'benign_egg__pos02': Value('string'), 'benign_egg__pos03': Value('string'), 'benign_egg__pos04': Value('string'),
...
09': Value('string'), 'harm_031__pos00': Value('string'), 'harm_031__pos01': Value('string'), 'harm_031__pos02': Value('string'), 'harm_031__pos03': Value('string'), 'harm_031__pos04': Value('string'), 'harm_031__pos05': Value('string'), 'harm_031__pos06': Value('string'), 'harm_031__pos07': Value('string'), 'harm_031__pos08': Value('string'), 'harm_031__pos09': Value('string'), 'harm_034__pos00': Value('string'), 'harm_034__pos01': Value('string'), 'harm_034__pos02': Value('string'), 'harm_034__pos03': Value('string'), 'harm_034__pos04': Value('string'), 'harm_034__pos05': Value('string'), 'harm_034__pos06': Value('string'), 'harm_034__pos07': Value('string'), 'harm_034__pos08': Value('string'), 'harm_034__pos09': Value('string'), 'harm_035__pos00': Value('string'), 'harm_035__pos01': Value('string'), 'harm_035__pos02': Value('string'), 'harm_035__pos03': Value('string'), 'harm_035__pos04': Value('string'), 'harm_035__pos05': Value('string'), 'harm_035__pos06': Value('string'), 'harm_035__pos07': Value('string'), 'harm_035__pos08': Value('string'), 'harm_035__pos09': Value('string'), 'harm_094__pos00': Value('string'), 'harm_094__pos01': Value('string'), 'harm_094__pos02': Value('string'), 'harm_094__pos03': Value('string'), 'harm_094__pos04': Value('string'), 'harm_094__pos05': Value('string'), 'harm_094__pos06': Value('string'), 'harm_094__pos07': Value('string'), 'harm_094__pos08': Value('string'), 'harm_094__pos09': Value('string')}}, 'later_todo': List(Value('string'))}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
2026.TA.tc8192_compliance_first10_graphs
Base-vs-adapter circuit-tracing comparison overlay graphs for the tc8192_compliance_first10 run.
Each overlay graph is one full-replacement attribution graph over a single prompt in which
MLP(x) = T_base(x) + T_adapter(x) + Err: base GemmaScope transcoder features (hexagon ⬢), sparse
transcoder-adapter features (circle ●) and real reconstruction-error nodes (triangle ▲) live in the
same graph, so the two models' circuits can be read position-by-position on identical tokens.
What this run is
The compliance contrast for the first-ten-token refusal build
(2026.TA.tc8192_refusal_first10_graphs): is the "refusal circuit" actually refusal-specific?
Same construction (a base-vs-adapter overlay at each of the first ten tokens of the adapter's own completion, deployed tc8192 adapter, chat template) but tracing the adapter's own COMPLIANT completion on 8 prompts it complies with: 3 benign + 5 harmful jailbreaks (harm_011 GPS-tracker; harm_031/034/035 Crimea/Holodomor/Agent-Orange disinfo; harm_094 huff-paint — the same disinfo/persuasion set the huge adapter jailbreaks on) = 80 graphs.
Result: diffing the top adapter features against the refusal build shows 12/13 recurring "refusal-opening" features also fire while complying — they are generic response-start features. Only L6/F4241 is refusal-specific (top-8 in ~33% of refusal graphs, 0 compliance graphs); inspect it here. A follow-up ablation (clamping L6/F4241 to 0) shows it intensifies emphasis rather than carrying the refusal decision.
Findings / write-ups (repo): my_notes/08-10-26/refusal_token_tracing/compare_refusal_compliance.md,
my_notes/08-10-26/refusal_token_tracing/refusal_token_tracing.pdf, my_notes/08-10-26/ablation_results.md.
Serve script: sh/visualize graphs/visualize 08-10-26 hf.sh (#3b, port 8060).
Serve it
uv run --extra viz python -m analysis.attribution.serve_comparison_graphs \
--graph_file_dir siddharthmb/2026.TA.tc8192_compliance_first10_graphs:overlay --port 8060
Swap :overlay for :overlay_compact for the pruned view. The snapshot is downloaded and cached on
first run (HF_HOME); node-click feature examples stream from the feature collections linked below.
Contents
| path | files | size |
|---|---|---|
overlay/ |
80 | 22.9 GB |
overlay_compact/ |
80 | 56.9 MB |
prompts/ |
81 | 29.6 KB |
The one-sided base/ and adapter/ graph dirs from the same run are not uploaded (the overlay
graphs contain both sides); they remain on the cluster at /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10 and are re-derivable with the
reproduction command below.
Provenance
| field | value |
|---|---|
| adapter checkpoint | siddharthmb/2026.TA.gemma2_2b_tc8192_decb_l1w0.001_tarbb_lb2.0_ln1_dr20000_lr8e-04_bs4_sl14793860 |
| base model | google/gemma-2-2b |
| adapter feature examples | siddharthmb/2026.TA.features_2026.TA.gemma2_2b_tc8192_decb_l1w0.001_tarbb_lb2.0_ln1_dr20000_lr_he94e9602bafa |
| base feature examples | siddharthmb/2026.TA.features_gemma-2-2b_gemmascope_width_16k_average_l0_76_ms100000_ml1024_tk2_h12ad59325ffd |
| prompt set | /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10/prompts |
| prompt format | chat |
| base attribution targets | union_adapter_top |
| GemmaScope | google/gemma-scope-2b-pt-transcoders / width_16k / average_l0_76 (per-layer L0s in gemmascope_config.yaml) |
| overlay options | {"hide_direct_embedding_logit_links": true, "compact_max_base_feature_nodes": 64, "compact_max_base_error_nodes": 0, "compact_max_adapter_feature_nodes": null} |
Feature collections (dataset-side artifacts baked into the graphs):
siddharthmb/2026.TA.features_2026.TA.gemma2_2b_tc8192_decb_l1w0.001_tarbb_lb2.0_ln1_dr20000_lr_he94e9602bafasiddharthmb/2026.TA.features_gemma-2-2b_gemmascope_width_16k_average_l0_76_ms100000_ml1024_tk2_h12ad59325ffd
Reproduce
Exact build invocation (from the run's Slurm log):
uv run --no-sync --extra viz python -m analysis.attribution.run_base_adapter_comparison --adapter_checkpoint siddharthmb/2026.TA.gemma2_2b_tc8192_decb_l1w0.001_tarbb_lb2.0_ln1_dr20000_lr8e-04_bs4_sl14793860 --base_model google/gemma-2-2b --prompts /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10/prompts --prompt_format chat --run_name tc8192_compliance_first10 --output_dir /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10 --gemmascope_width width_16k --gemmascope_l0 average_l0_76 --gemmascope_l0_match nearest --feature_data_path siddharthmb/2026.TA.features_2026.TA.gemma2_2b_tc8192_decb_l1w0.001_tarbb_lb2.0_ln1_dr20000_lr_he94e9602bafa --base_feature_data_path siddharthmb/2026.TA.features_gemma-2-2b_gemmascope_width_16k_average_l0_76_ms100000_ml1024_tk2_h12ad59325ffd --continuation_max_new_tokens 0
Build log on the cluster: /juice2/u/siddharth/transcoder-adapters/logs/attribution/16760250.out
Re-star the jailbreak/compliance prompts in the dropdown after any rebuild (idempotent, auto-detects the set from the baked adapter continuations):
uv run python -m analysis.evals.classify_adapter_compliance \
--graph_dir /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10/overlay --star --also_star_dir /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10/overlay_compact
Uploaded with:
uv run python -m analysis.attribution.upload_comparison_graphs --graph_dir /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10
Cluster paths
- graph dir:
/nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10 - manifest:
/nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/tc8192_compliance_first10/comparison-manifest.json(also in this repo) - build log:
/juice2/u/siddharth/transcoder-adapters/logs/attribution/16760250.out
- Downloads last month
- 236