Dataset Viewer
Duplicate
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
name: string
source: string
split_layer: int64
hook: string
n_layer: int64
n_head: int64
d_model: int64
vocab_size: int64
n_positions: int64
prepend_bos: bool
bos_token_id: int64
top_k: int64
quantization: string
files: struct<embeddings: string, part_a: string, part_b: string>
  child 0, embeddings: string
  child 1, part_a: string
  child 2, part_b: string
external_data: struct<part_b: list<item: string>>
  child 0, part_b: list<item: string>
      child 0, item: string
bytes: struct<part_a.onnx: int64, part_b.onnx: int64, embeddings.f16.bin: int64>
  child 0, part_a.onnx: int64
  child 1, part_b.onnx: int64
  child 2, embeddings.f16.bin: int64
sha256: struct<part_a.onnx: string, part_b.onnx: string, embeddings.f16.bin: string>
  child 0, part_a.onnx: string
  child 1, part_b.onnx: string
  child 2, embeddings.f16.bin: string
verification: struct<fp32: struct<prompts: int64, positions: int64, torch_vs_hf_max_abs: double, onnx_vs_hf_max_ab (... 655 chars omitted)
  child 0, fp32: struct<prompts: int64, positions: int64, torch_vs_hf_max_abs: double, onnx_vs_hf_max_abs: double, kl (... 266 chars omitted)
      child 0, prompts: int64
      child 1, positions: int64
      child 2, torch_vs_hf_max_abs: double
      child 3, onnx_vs_hf_max_abs: double
      child 4, kl_mean: double
      child 5, kl_max: double
      child 6, top1_agreement: double
      child 7, argmax_match: bool
      child 8, greedy: struct<prompts: int64, tokens: int64, identical: int64, mean_matching_prefix
...
, 21015: string
  child 241, 21141: string
  child 242, 21282: string
  child 243, 21309: string
  child 244, 21327: string
  child 245, 21387: string
  child 246, 21439: string
  child 247, 21452: string
  child 248, 21525: string
  child 249, 21539: string
  child 250, 21571: string
  child 251, 21687: string
  child 252, 21717: string
  child 253, 21718: string
  child 254, 21782: string
  child 255, 21842: string
  child 256, 21849: string
  child 257, 21952: string
  child 258, 22021: string
  child 259, 22029: string
  child 260, 22097: string
  child 261, 22134: string
  child 262, 22141: string
  child 263, 22328: string
  child 264, 22371: string
  child 265, 22422: string
  child 266, 22434: string
  child 267, 22486: string
  child 268, 22543: string
  child 269, 22614: string
  child 270, 22710: string
  child 271, 22730: string
  child 272, 22749: string
  child 273, 22771: string
  child 274, 22792: string
  child 275, 22845: string
  child 276, 23023: string
  child 277, 23096: string
  child 278, 23283: string
  child 279, 23335: string
  child 280, 23387: string
  child 281, 23409: string
  child 282, 23494: string
  child 283, 23525: string
  child 284, 23724: string
  child 285, 23780: string
  child 286, 24008: string
  child 287, 24045: string
  child 288, 24114: string
  child 289, 24191: string
  child 290, 24276: string
  child 291, 24336: string
  child 292, 24380: string
  child 293, 24454: string
  child 294, 24470: string
  child 295, 24471: string
to
{'format': Value('int64'), 'model': Value('string'), 'prompt': Value('string'), 'maxNewTokens': Value('int64'), 'steer': {'vectorId': Value('string'), 'strength': Value('float64'), 'label': Value('string')}, 'labels': {'144': Value('string'), '196': Value('string'), '344': Value('string'), '612': Value('string'), '705': Value('string'), '713': Value('string'), '1020': Value('string'), '1077': Value('string'), '1167': Value('string'), '1237': Value('string'), '1288': Value('string'), '1592': Value('string'), '1611': Value('string'), '1622': Value('string'), '1738': Value('string'), '1781': Value('string'), '1795': Value('string'), '1855': Value('string'), '1901': Value('string'), '1980': Value('string'), '2173': Value('string'), '2248': Value('string'), '2379': Value('string'), '2392': Value('string'), '2401': Value('string'), '2414': Value('string'), '2427': Value('string'), '2472': Value('string'), '2528': Value('string'), '2694': Value('string'), '2726': Value('string'), '2747': Value('string'), '2884': Value('string'), '2905': Value('string'), '3172': Value('string'), '3277': Value('string'), '3366': Value('string'), '3506': Value('string'), '3639': Value('string'), '3693': Value('string'), '3724': Value('string'), '3874': Value('string'), '3910': Value('string'), '3991': Value('string'), '4018': Value('string'), '4053': Value('string'), '4066': Value('string'), '4115': Value('string'), '4132': Value('string'), '4324': Value('string'), '4383': Value('string'), '4422': Valu
...
ing'), '22749': Value('string'), '22771': Value('string'), '22792': Value('string'), '22845': Value('string'), '23023': Value('string'), '23096': Value('string'), '23283': Value('string'), '23335': Value('string'), '23387': Value('string'), '23409': Value('string'), '23494': Value('string'), '23525': Value('string'), '23724': Value('string'), '23780': Value('string'), '24008': Value('string'), '24045': Value('string'), '24114': Value('string'), '24191': Value('string'), '24276': Value('string'), '24336': Value('string'), '24380': Value('string'), '24454': Value('string'), '24470': Value('string'), '24471': Value('string')}, 'runs': {'original': {'tokens': List({'tokenId': Value('int64'), 'text': Value('string'), 'kind': Value('string')}), 'topK': Value('int64'), 'topIds': List(Value('int64')), 'topProbs': List(Value('float64')), 'topTexts': List(Value('string')), 'nLayer': Value('int64'), 'nHead': Value('int64'), 'attention': Value('string'), 'concepts': {'offsets': List(Value('int64')), 'features': List(Value('int64')), 'acts': List(Value('float64'))}}, 'steered': {'tokens': List({'tokenId': Value('int64'), 'text': Value('string'), 'kind': Value('string')}), 'topK': Value('int64'), 'topIds': List(Value('int64')), 'topProbs': List(Value('float64')), 'topTexts': List(Value('string')), 'nLayer': Value('int64'), 'nHead': Value('int64'), 'attention': Value('string'), 'concepts': {'offsets': List(Value('int64')), 'features': List(Value('int64')), 'acts': List(Value('float64'))}}}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 544, 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 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 361, 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 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              name: string
              source: string
              split_layer: int64
              hook: string
              n_layer: int64
              n_head: int64
              d_model: int64
              vocab_size: int64
              n_positions: int64
              prepend_bos: bool
              bos_token_id: int64
              top_k: int64
              quantization: string
              files: struct<embeddings: string, part_a: string, part_b: string>
                child 0, embeddings: string
                child 1, part_a: string
                child 2, part_b: string
              external_data: struct<part_b: list<item: string>>
                child 0, part_b: list<item: string>
                    child 0, item: string
              bytes: struct<part_a.onnx: int64, part_b.onnx: int64, embeddings.f16.bin: int64>
                child 0, part_a.onnx: int64
                child 1, part_b.onnx: int64
                child 2, embeddings.f16.bin: int64
              sha256: struct<part_a.onnx: string, part_b.onnx: string, embeddings.f16.bin: string>
                child 0, part_a.onnx: string
                child 1, part_b.onnx: string
                child 2, embeddings.f16.bin: string
              verification: struct<fp32: struct<prompts: int64, positions: int64, torch_vs_hf_max_abs: double, onnx_vs_hf_max_ab (... 655 chars omitted)
                child 0, fp32: struct<prompts: int64, positions: int64, torch_vs_hf_max_abs: double, onnx_vs_hf_max_abs: double, kl (... 266 chars omitted)
                    child 0, prompts: int64
                    child 1, positions: int64
                    child 2, torch_vs_hf_max_abs: double
                    child 3, onnx_vs_hf_max_abs: double
                    child 4, kl_mean: double
                    child 5, kl_max: double
                    child 6, top1_agreement: double
                    child 7, argmax_match: bool
                    child 8, greedy: struct<prompts: int64, tokens: int64, identical: int64, mean_matching_prefix
              ...
              , 21015: string
                child 241, 21141: string
                child 242, 21282: string
                child 243, 21309: string
                child 244, 21327: string
                child 245, 21387: string
                child 246, 21439: string
                child 247, 21452: string
                child 248, 21525: string
                child 249, 21539: string
                child 250, 21571: string
                child 251, 21687: string
                child 252, 21717: string
                child 253, 21718: string
                child 254, 21782: string
                child 255, 21842: string
                child 256, 21849: string
                child 257, 21952: string
                child 258, 22021: string
                child 259, 22029: string
                child 260, 22097: string
                child 261, 22134: string
                child 262, 22141: string
                child 263, 22328: string
                child 264, 22371: string
                child 265, 22422: string
                child 266, 22434: string
                child 267, 22486: string
                child 268, 22543: string
                child 269, 22614: string
                child 270, 22710: string
                child 271, 22730: string
                child 272, 22749: string
                child 273, 22771: string
                child 274, 22792: string
                child 275, 22845: string
                child 276, 23023: string
                child 277, 23096: string
                child 278, 23283: string
                child 279, 23335: string
                child 280, 23387: string
                child 281, 23409: string
                child 282, 23494: string
                child 283, 23525: string
                child 284, 23724: string
                child 285, 23780: string
                child 286, 24008: string
                child 287, 24045: string
                child 288, 24114: string
                child 289, 24191: string
                child 290, 24276: string
                child 291, 24336: string
                child 292, 24380: string
                child 293, 24454: string
                child 294, 24470: string
                child 295, 24471: string
              to
              {'format': Value('int64'), 'model': Value('string'), 'prompt': Value('string'), 'maxNewTokens': Value('int64'), 'steer': {'vectorId': Value('string'), 'strength': Value('float64'), 'label': Value('string')}, 'labels': {'144': Value('string'), '196': Value('string'), '344': Value('string'), '612': Value('string'), '705': Value('string'), '713': Value('string'), '1020': Value('string'), '1077': Value('string'), '1167': Value('string'), '1237': Value('string'), '1288': Value('string'), '1592': Value('string'), '1611': Value('string'), '1622': Value('string'), '1738': Value('string'), '1781': Value('string'), '1795': Value('string'), '1855': Value('string'), '1901': Value('string'), '1980': Value('string'), '2173': Value('string'), '2248': Value('string'), '2379': Value('string'), '2392': Value('string'), '2401': Value('string'), '2414': Value('string'), '2427': Value('string'), '2472': Value('string'), '2528': Value('string'), '2694': Value('string'), '2726': Value('string'), '2747': Value('string'), '2884': Value('string'), '2905': Value('string'), '3172': Value('string'), '3277': Value('string'), '3366': Value('string'), '3506': Value('string'), '3639': Value('string'), '3693': Value('string'), '3724': Value('string'), '3874': Value('string'), '3910': Value('string'), '3991': Value('string'), '4018': Value('string'), '4053': Value('string'), '4066': Value('string'), '4115': Value('string'), '4132': Value('string'), '4324': Value('string'), '4383': Value('string'), '4422': Valu
              ...
              ing'), '22749': Value('string'), '22771': Value('string'), '22792': Value('string'), '22845': Value('string'), '23023': Value('string'), '23096': Value('string'), '23283': Value('string'), '23335': Value('string'), '23387': Value('string'), '23409': Value('string'), '23494': Value('string'), '23525': Value('string'), '23724': Value('string'), '23780': Value('string'), '24008': Value('string'), '24045': Value('string'), '24114': Value('string'), '24191': Value('string'), '24276': Value('string'), '24336': Value('string'), '24380': Value('string'), '24454': Value('string'), '24470': Value('string'), '24471': Value('string')}, 'runs': {'original': {'tokens': List({'tokenId': Value('int64'), 'text': Value('string'), 'kind': Value('string')}), 'topK': Value('int64'), 'topIds': List(Value('int64')), 'topProbs': List(Value('float64')), 'topTexts': List(Value('string')), 'nLayer': Value('int64'), 'nHead': Value('int64'), 'attention': Value('string'), 'concepts': {'offsets': List(Value('int64')), 'features': List(Value('int64')), 'acts': List(Value('float64'))}}, 'steered': {'tokens': List({'tokenId': Value('int64'), 'text': Value('string'), 'kind': Value('string')}), 'topK': Value('int64'), 'topIds': List(Value('int64')), 'topProbs': List(Value('float64')), 'topTexts': List(Value('string')), 'nLayer': Value('int64'), 'nHead': Value('int64'), 'attention': Value('string'), 'concepts': {'offsets': List(Value('int64')), 'features': List(Value('int64')), 'acts': List(Value('float64'))}}}}
              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.

Prizma model artifacts

Files downloaded by Prizma, a site that runs GPT-2 small entirely in the browser and shows what happens inside it. Everything here is generated by the scripts in the repository's research/ folder; nothing is edited by hand.

models/gpt2-l6/
  manifest.json        shapes, split layer, file hashes, verification against the original model
  embeddings.f16.bin   GPT-2's token-embedding table, raw little-endian fp16 [50257, 768]
  part_a.onnx          blocks 0-5: token embeddings -> residual stream entering block 6, attention
  part_b.onnx          blocks 6-11 + unembedding: residual stream -> logits, top-10, attention
  steering.json        unit-norm steering directions and the strength range that works for each
  demo.json            a precomputed example (original and steered text) shown while the model downloads
  sae/                 the sparse autoencoder the site reads concepts with and steers along
    manifest.json      how to encode, shapes, file names, self-check, explained variance
    encoder.f16.bin    encoder weights, raw little-endian fp16 [24576, 768]
    biases.f32.bin     encoder and decoder biases, fp32
    decoder/NNNN.f16.bin   decoder rows in shards of 256, fp16
    labels.json        one short explanation per feature

GPT-2 small is split at blocks.6.hook_resid_pre so the browser can read and edit the residual stream between the two halves. The transformer blocks use 8-bit weight quantization (MatMulNBits); the unembedding matrix is read from embeddings.f16.bin. manifest.json records how closely the result matches the original model.

Sources

  • openai-community/gpt2 (MIT license).
  • sae/ holds the weights of Joseph Bloom's open residual-stream SAE for blocks.6.hook_resid_pre, jbloom/GPT2-Small-SAEs-Reformatted (MIT license), converted to fp16 without other changes. Some directions in steering.json are decoder rows of the same SAE.
  • sae/labels.json holds the feature explanations published by Neuronpedia for that SAE (their public data export). They are Neuronpedia's work, not ours, and are not covered by the license tag of this card.
Downloads last month
124