Datasets:

siddharthmb
/

2026.TA.comply_vs_refuse_chat_hires_graphs

Tags:

Dataset Viewer

Split (1)

train

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
metadata: struct<slug: string, scan: string, transcoder_list: list<item: null>, prompt_tokens: list<item: stri (... 1033 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 (... 583 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: int64, error_nodes: int64>
          child 0, base_features: int64
          child 1, adapter_features: int64
          child 2, error_nodes: int64
      child 8, link_filter: struct<hide_direct_embedding_logit_links: bool, skipped_direct_embe
...
: string
  child 2, features_dir: null
results: struct<base: struct<harm_000: string, harm_012: string, harm_016: string, harm_017: string, harm_021 (... 297 chars omitted)
  child 0, base: struct<harm_000: string, harm_012: string, harm_016: string, harm_017: string, harm_021: string, har (... 86 chars omitted)
      child 0, harm_000: string
      child 1, harm_012: string
      child 2, harm_016: string
      child 3, harm_017: string
      child 4, harm_021: string
      child 5, harm_028: string
      child 6, harm_031: string
      child 7, harm_034: string
      child 8, harm_094: string
      child 9, harm_187: string
  child 1, adapter: struct<harm_000: string, harm_012: string, harm_016: string, harm_017: string, harm_021: string, har (... 86 chars omitted)
      child 0, harm_000: string
      child 1, harm_012: string
      child 2, harm_016: string
      child 3, harm_017: string
      child 4, harm_021: string
      child 5, harm_028: string
      child 6, harm_031: string
      child 7, harm_034: string
      child 8, harm_094: string
      child 9, harm_187: 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
overlay_graph_paths: list<item: string>
  child 0, item: 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': {'harm_000': Value('string'), 'harm_012': Value('string'), 'harm_016': Value('string'), 'harm_017': Value('string'), 'harm_021': Value('string'), 'harm_028': Value('string'), 'harm_031': Value('string'), 'harm_034': Value('string'), 'harm_094': Value('string'), 'harm_187': Value('string')}, 'adapter': {'harm_000': Value('string'), 'harm_012': Value('string'), 'harm_016': Value('string'), 'harm_017': Value('string'), 'harm_021': Value('string'), 'harm_028': Value('string'), 'harm_031': Value('string'), 'harm_034': Value('string'), 'harm_094': Value('string'), 'harm_187': 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
              metadata: struct<slug: string, scan: string, transcoder_list: list<item: null>, prompt_tokens: list<item: stri (... 1033 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 (... 583 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: int64, error_nodes: int64>
                        child 0, base_features: int64
                        child 1, adapter_features: int64
                        child 2, error_nodes: int64
                    child 8, link_filter: struct<hide_direct_embedding_logit_links: bool, skipped_direct_embe
              ...
              : string
                child 2, features_dir: null
              results: struct<base: struct<harm_000: string, harm_012: string, harm_016: string, harm_017: string, harm_021 (... 297 chars omitted)
                child 0, base: struct<harm_000: string, harm_012: string, harm_016: string, harm_017: string, harm_021: string, har (... 86 chars omitted)
                    child 0, harm_000: string
                    child 1, harm_012: string
                    child 2, harm_016: string
                    child 3, harm_017: string
                    child 4, harm_021: string
                    child 5, harm_028: string
                    child 6, harm_031: string
                    child 7, harm_034: string
                    child 8, harm_094: string
                    child 9, harm_187: string
                child 1, adapter: struct<harm_000: string, harm_012: string, harm_016: string, harm_017: string, harm_021: string, har (... 86 chars omitted)
                    child 0, harm_000: string
                    child 1, harm_012: string
                    child 2, harm_016: string
                    child 3, harm_017: string
                    child 4, harm_021: string
                    child 5, harm_028: string
                    child 6, harm_031: string
                    child 7, harm_034: string
                    child 8, harm_094: string
                    child 9, harm_187: 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
              overlay_graph_paths: list<item: string>
                child 0, item: 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': {'harm_000': Value('string'), 'harm_012': Value('string'), 'harm_016': Value('string'), 'harm_017': Value('string'), 'harm_021': Value('string'), 'harm_028': Value('string'), 'harm_031': Value('string'), 'harm_034': Value('string'), 'harm_094': Value('string'), 'harm_187': Value('string')}, 'adapter': {'harm_000': Value('string'), 'harm_012': Value('string'), 'harm_016': Value('string'), 'harm_017': Value('string'), 'harm_021': Value('string'), 'harm_028': Value('string'), 'harm_031': Value('string'), 'harm_034': Value('string'), 'harm_094': Value('string'), 'harm_187': Value('string')}}, 'later_todo': List(Value('string'))}
              because column names don't match

Need 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.comply_vs_refuse_chat_hires_graphs

Base-vs-adapter circuit-tracing comparison overlay graphs for the comply_vs_refuse_chat 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 clean comply-vs-refuse contrast — the main result. Native chat template, huge (tc16384) adapter, 12288 feature nodes (hi-res). 10 prompts: 4 confident jailbreaks (★ starred: harm_031/034 Crimea/Holodomor disinfo, harm_094 drug-persuasion, harm_187 cyber) vs 6 diverse confident refusals.

Both sides of the contrast are in-distribution (chat is the adapter's training template) and confident, which is what makes this the trustworthy split — it is the build used to locate the adapter's refusal/compliance decision features. Contrast it with the template-flip builds (2026.TA.flip_huge_chat_hires_graphs / ..._plain_hires_graphs), which are deliberately NOT clean.

Findings / write-up (repo): my_notes/08-10-26/refusal_compliance_feature_audit.md. Serve script: sh/visualize graphs/visualize 08-10-26 hf.sh (#4, port 8057).

Serve it

uv run --extra viz python -m analysis.attribution.serve_comparison_graphs \
  --graph_file_dir siddharthmb/2026.TA.comply_vs_refuse_chat_hires_graphs:overlay --port 8057

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.

path	files	size
`overlay/`	11	1.8 GB
`overlay_compact/`	11	48.8 MB

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/comply_vs_refuse_chat_hires and are re-derivable with the reproduction command below.

Provenance

field	value
adapter checkpoint	`siddharthmb/2026.TA.gemma2_2b_huge_tc16384_decb_l1w0.0003_norm_sch_tarbb_lb2.0_ln1.0_dr500000_lr2e-04_bs8_sl`
base model	`google/gemma-2-2b`
adapter feature examples	`siddharthmb/2026.TA.features_2026.TA.gemma2_2b_huge_tc16384_decb_l1w0.0003_norm_sch_tarbb_lb2_hff15229f5fe1`
base feature examples	`siddharthmb/2026.TA.features_gemma-2-2b_gemmascope_width_16k_average_l0_76_ms100000_ml1024_tk1_hf83e96d574d2`
prompt set	`/nlp/u/siddharth/transcoder-adapters/experiments/interesting_queries/results/strict_compliance_refusal/comply_vs_refuse_chat`
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):

Reproduce

Exact build invocation (from the run's Slurm log):

uv run --extra viz python -m analysis.attribution.run_base_adapter_comparison --adapter_checkpoint siddharthmb/2026.TA.gemma2_2b_huge_tc16384_decb_l1w0.0003_norm_sch_tarbb_lb2.0_ln1.0_dr500000_lr2e-04_bs8_sl --base_model google/gemma-2-2b --gemmascope_width width_16k --gemmascope_l0 average_l0_76 --base_feature_data_path siddharthmb/2026.TA.features_gemma-2-2b_gemmascope_width_16k_average_l0_76_ms100000_ml1024_tk1_hf83e96d574d2 --feature_data_path siddharthmb/2026.TA.features_2026.TA.gemma2_2b_huge_tc16384_decb_l1w0.0003_norm_sch_tarbb_lb2_hff15229f5fe1 --prompts /nlp/u/siddharth/transcoder-adapters/experiments/interesting_queries/results/strict_compliance_refusal/comply_vs_refuse_chat --prompt_format chat --max_feature_nodes 12288 --max_n_logits 10 --run_name comply_vs_refuse_chat --output_dir /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/comply_vs_refuse_chat_hires

Build log on the cluster: /juice2/u/siddharth/transcoder-adapters/slurm-16723869.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/comply_vs_refuse_chat_hires/overlay --star --also_star_dir /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/comply_vs_refuse_chat_hires/overlay_compact

Uploaded with:

uv run python -m analysis.attribution.upload_comparison_graphs --graph_dir /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/comply_vs_refuse_chat_hires

Cluster paths

graph dir: /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/comply_vs_refuse_chat_hires
manifest: /nlp/scr/siddharth/transcoder-adapters/base_adapter_comparisons/comply_vs_refuse_chat_hires/comparison-manifest.json (also in this repo)
build log: /juice2/u/siddharth/transcoder-adapters/slurm-16723869.out

Downloads last month: 56

Total file size:

1.93 GB