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
total_num_rows: int64
shard_rows: struct<s3://marin-us-east-02a/marin/tokenized/code-contests-noblock-chat-d1fb27/2026.07.17/train/par (... 24 chars omitted)
  child 0, s3://marin-us-east-02a/marin/tokenized/code-contests-noblock-chat-d1fb27/2026.07.17/train/part-00000 (... 9 chars omitted): int64
is_finished: bool
finished_shards: list<item: string>
  child 0, item: string
field_counts: struct<assistant_masks: int64, input_ids: int64>
  child 0, assistant_masks: int64
  child 1, input_ids: int64
field_counts_by_shard: struct<>
layout: string
metadata: struct<preprocessor_metadata: struct<>>
  child 0, preprocessor_metadata: struct<>
codecs: null
node_type: string
chunk_key_encoding: struct<name: string>
  child 0, name: string
chunk_grid: struct<configuration: struct<chunk_shape: list<item: int64>>, name: string>
  child 0, configuration: struct<chunk_shape: list<item: int64>>
      child 0, chunk_shape: list<item: int64>
          child 0, item: int64
  child 1, name: string
fill_value: int64
zarr_format: int64
shape: list<item: int64>
  child 0, item: int64
data_type: string
to
{'chunk_grid': {'configuration': {'chunk_shape': List(Value('int64'))}, 'name': Value('string')}, 'chunk_key_encoding': {'name': Value('string')}, 'codecs': List({'configuration': {'chunk_shape': List(Value('int64')), 'codecs': List({'configuration': Json(decode=True), 'name': Value('string')}), 'index_codecs': List(Json(decode=True))}, 'name': Value('string')}), 'data_type': Value('string'), 'fill_value': Value('int64'), 'node_type': Value('string'), 'shape': List(Value('int64')), 'zarr_format': Value('int64')}
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 489, 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 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              total_num_rows: int64
              shard_rows: struct<s3://marin-us-east-02a/marin/tokenized/code-contests-noblock-chat-d1fb27/2026.07.17/train/par (... 24 chars omitted)
                child 0, s3://marin-us-east-02a/marin/tokenized/code-contests-noblock-chat-d1fb27/2026.07.17/train/part-00000 (... 9 chars omitted): int64
              is_finished: bool
              finished_shards: list<item: string>
                child 0, item: string
              field_counts: struct<assistant_masks: int64, input_ids: int64>
                child 0, assistant_masks: int64
                child 1, input_ids: int64
              field_counts_by_shard: struct<>
              layout: string
              metadata: struct<preprocessor_metadata: struct<>>
                child 0, preprocessor_metadata: struct<>
              codecs: null
              node_type: string
              chunk_key_encoding: struct<name: string>
                child 0, name: string
              chunk_grid: struct<configuration: struct<chunk_shape: list<item: int64>>, name: string>
                child 0, configuration: struct<chunk_shape: list<item: int64>>
                    child 0, chunk_shape: list<item: int64>
                        child 0, item: int64
                child 1, name: string
              fill_value: int64
              zarr_format: int64
              shape: list<item: int64>
                child 0, item: int64
              data_type: string
              to
              {'chunk_grid': {'configuration': {'chunk_shape': List(Value('int64'))}, 'name': Value('string')}, 'chunk_key_encoding': {'name': Value('string')}, 'codecs': List({'configuration': {'chunk_shape': List(Value('int64')), 'codecs': List({'configuration': Json(decode=True), 'name': Value('string')}), 'index_codecs': List(Json(decode=True))}, 'name': Value('string')}), 'data_type': Value('string'), 'fill_value': Value('int64'), 'node_type': Value('string'), 'shape': List(Value('int64')), 'zarr_format': Value('int64')}
              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.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Grug 67B agentic SFT training dataset

This directory is a local, revision-pinned reconstruction of the exact 29-component mixture consumed by grug_67b_a2b_sft_s3_agentic.

The reconstruction has two representations:

  • converted_hf/ contains the readable converted datasets. Each component is checked out at the full Hugging Face commit recorded by the corresponding Marin document artifact. These 29 snapshots contain 77,012 conversations and occupy about 1.67 GB before filesystem overhead.
  • token_caches/ contains the token IDs, assistant masks, offsets, ledgers, and statistics that Levanter actually read. These are byte-for-byte copies from the 29 immutable s3://marin-us-east-02a/marin/tokenized/*-chat-d1fb27/2026.07.17 cache prefixes and contain 797,783,562 tokens.

manifest.json maps every source trace repository to its converted repository, full Hugging Face revision, document artifact, token-cache URI, row and token counts, object-store ETags, and local SHA-256 checksums. wandb_training_data_config.json is the data/trainer projection from the authoritative W&B run. chat_template.jinja is the exact template embedded in every cache; its SHA-256 is 7481acabcc6916e27df9b536fdeff24c4d326c04e5e387ea3d5c9863da499577.

The aggregate reproduces the recorded training length:

ceil(5 epochs * 797,783,562 tokens / (32,768 tokens * batch 64)) = 1,903 steps

All 29 components have weight 1.0. Equal component weights describe Marin's blockwise mixture sampling; concatenating the Parquet files is useful for inspection but is not equivalent to the shuffled training stream. Exact replay should use wandb_training_data_config.json and the local token caches.

The conversion batch produced 32 usable repositories, but the training run selected only these 29. The three successful conversions omitted from training were exp_rpt_methods2test-large-v3, exp_rpt_pymethods2test-large, and exp_rpt_pymethods2test-v3. Two other attempted conversions failed and never entered the mixture.

Raw cache .artifact.json files are deliberately absent. Their command-line provenance contains plaintext credentials. The manifest retains their SHA-256 digests and a sanitized provenance projection without copying those credentials.

Downloads last month
63