Datasets:

zesen-kth
/

tiny-llm

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schedule: string
method: string
horizon: int64
lr: double
beta1: double
beta2: double
id: string
mean: double
sd: double
runs: list<item: struct<id: string, seed: int64, loss: double, jobId: string, artifacts: struct<metrics.js (... 155 chars omitted)
  child 0, item: struct<id: string, seed: int64, loss: double, jobId: string, artifacts: struct<metrics.jsonl: string (... 143 chars omitted)
      child 0, id: string
      child 1, seed: int64
      child 2, loss: double
      child 3, jobId: string
      child 4, artifacts: struct<metrics.jsonl: string, result.json: string, resolved.yaml: string, receipt.json: string, requ (... 70 chars omitted)
          child 0, metrics.jsonl: string
          child 1, result.json: string
          child 2, resolved.yaml: string
          child 3, receipt.json: string
          child 4, request.json: string
          child 5, environment.json: string
          child 6, continuation.json: string
rank: int64
logs/wsd-awc8-h40.records.jsonl.gz: string
curves/awc4-cosine-h40.jsonl.gz: string
logs/cosine-sync-h20.records.jsonl.gz: string
provenance/repack.json: string
logs/wsd-sync-h40.records.jsonl.gz: string
worker-comparison.pdf: string
sources/campaigns.jsonl.gz: string
logs/wsd-awc8-h40.metrics.jsonl.gz: string
dataset.json: string
curves/awc8-cosine-h20.jsonl.gz: string
schedule-comparison-sync.pdf: string
schedule-comparison-sync.png: string
curves/sync-wsd-h40.jsonl.gz: string
curves/awc8-cosine-h80.jsonl.gz: string
worker-comparison-mob
...
sonl.gz: string
curves/awc8-wsd-h160.jsonl.gz: string
schedule-comparison-sync.svg: string
curves/awc8-wsd-h120.jsonl.gz: string
curves/sync-wsd-h20.jsonl.gz: string
curves/sync-cosine-h20.jsonl.gz: string
logs/cosine-awc4-h20.records.jsonl.gz: string
logs/wsd-sync-h120.metrics.jsonl.gz: string
manifest.json: string
logs/wsd-sync-h160.metrics.jsonl.gz: string
runs.json: string
schedule-comparison-awc8.png: string
logs/cosine-awc8-h80.metrics.jsonl.gz: string
logs/wsd-sync-h160.records.jsonl.gz: string
logs/wsd-sync-h80.metrics.jsonl.gz: string
logs/wsd-awc8-h80.records.jsonl.gz: string
logs/cosine-awc4-h40.metrics.jsonl.gz: string
logs/cosine-awc8-h80.records.jsonl.gz: string
worker-comparison.svg: string
schedule-comparison.pdf: string
logs/cosine-awc4-h80.records.jsonl.gz: string
winners.json: string
worker-comparison-mobile.svg: string
logs/cosine-awc4-h80.metrics.jsonl.gz: string
curves/sync-wsd-h80.jsonl.gz: string
schedule-comparison.png: string
winners.csv: string
worker-comparison.png: string
curves/awc8-cosine-h40.jsonl.gz: string
logs/wsd-sync-h120.records.jsonl.gz: string
logs/cosine-awc4-h40.records.jsonl.gz: string
logs/wsd-awc8-h20.records.jsonl.gz: string
gh200-throughput.png: string
logs/wsd-sync-h40.metrics.jsonl.gz: string
curves/sync-wsd-h120.jsonl.gz: string
runs.csv: string
performance.csv: string
performance.json: string
shared/environments.jsonl.gz: string
curves/sync-cosine-h80.jsonl.gz: string
performance-runs.json: string
performance-runs.csv: string
to
{'configurations.csv': Value('string'), 'configurations.json': Value('string'), 'curves/awc4-cosine-h20.jsonl.gz': Value('string'), 'curves/awc4-cosine-h40.jsonl.gz': Value('string'), 'curves/awc4-cosine-h80.jsonl.gz': Value('string'), 'curves/awc8-cosine-h20.jsonl.gz': Value('string'), 'curves/awc8-cosine-h40.jsonl.gz': Value('string'), 'curves/awc8-cosine-h80.jsonl.gz': Value('string'), 'curves/awc8-wsd-h120.jsonl.gz': Value('string'), 'curves/awc8-wsd-h160.jsonl.gz': Value('string'), 'curves/awc8-wsd-h20.jsonl.gz': Value('string'), 'curves/awc8-wsd-h40.jsonl.gz': Value('string'), 'curves/awc8-wsd-h80.jsonl.gz': Value('string'), 'curves/sync-cosine-h20.jsonl.gz': Value('string'), 'curves/sync-cosine-h40.jsonl.gz': Value('string'), 'curves/sync-cosine-h80.jsonl.gz': Value('string'), 'curves/sync-wsd-h120.jsonl.gz': Value('string'), 'curves/sync-wsd-h160.jsonl.gz': Value('string'), 'curves/sync-wsd-h20.jsonl.gz': Value('string'), 'curves/sync-wsd-h40.jsonl.gz': Value('string'), 'curves/sync-wsd-h80.jsonl.gz': Value('string'), 'dataset.json': Value('string'), 'gh200-throughput.pdf': Value('string'), 'gh200-throughput.png': Value('string'), 'gh200-throughput.svg': Value('string'), 'logs/cosine-awc4-h20.metrics.jsonl.gz': Value('string'), 'logs/cosine-awc4-h20.records.jsonl.gz': Value('string'), 'logs/cosine-awc4-h40.metrics.jsonl.gz': Value('string'), 'logs/cosine-awc4-h40.records.jsonl.gz': Value('string'), 'logs/cosine-awc4-h80.metrics.jsonl.gz': Value('string'), 'logs/cosine
...
onl.gz': Value('string'), 'logs/wsd-sync-h40.records.jsonl.gz': Value('string'), 'logs/wsd-sync-h80.metrics.jsonl.gz': Value('string'), 'logs/wsd-sync-h80.records.jsonl.gz': Value('string'), 'manifest.json': Value('string'), 'matched.csv': Value('string'), 'matched.json': Value('string'), 'performance-runs.csv': Value('string'), 'performance-runs.json': Value('string'), 'performance.csv': Value('string'), 'performance.json': Value('string'), 'provenance/pre-repack-checksums.json.gz': Value('string'), 'provenance/repack.json': Value('string'), 'runs.csv': Value('string'), 'runs.json': Value('string'), 'schedule-comparison-awc8.pdf': Value('string'), 'schedule-comparison-awc8.png': Value('string'), 'schedule-comparison-awc8.svg': Value('string'), 'schedule-comparison-sync.pdf': Value('string'), 'schedule-comparison-sync.png': Value('string'), 'schedule-comparison-sync.svg': Value('string'), 'schedule-comparison.pdf': Value('string'), 'schedule-comparison.png': Value('string'), 'schedule-comparison.svg': Value('string'), 'shared/environments.jsonl.gz': Value('string'), 'sources/campaigns.jsonl.gz': Value('string'), 'sources.json': Value('string'), 'winners.csv': Value('string'), 'winners.json': Value('string'), 'worker-comparison-mobile.pdf': Value('string'), 'worker-comparison-mobile.png': Value('string'), 'worker-comparison-mobile.svg': Value('string'), 'worker-comparison.pdf': Value('string'), 'worker-comparison.png': Value('string'), 'worker-comparison.svg': 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
              schedule: string
              method: string
              horizon: int64
              lr: double
              beta1: double
              beta2: double
              id: string
              mean: double
              sd: double
              runs: list<item: struct<id: string, seed: int64, loss: double, jobId: string, artifacts: struct<metrics.js (... 155 chars omitted)
                child 0, item: struct<id: string, seed: int64, loss: double, jobId: string, artifacts: struct<metrics.jsonl: string (... 143 chars omitted)
                    child 0, id: string
                    child 1, seed: int64
                    child 2, loss: double
                    child 3, jobId: string
                    child 4, artifacts: struct<metrics.jsonl: string, result.json: string, resolved.yaml: string, receipt.json: string, requ (... 70 chars omitted)
                        child 0, metrics.jsonl: string
                        child 1, result.json: string
                        child 2, resolved.yaml: string
                        child 3, receipt.json: string
                        child 4, request.json: string
                        child 5, environment.json: string
                        child 6, continuation.json: string
              rank: int64
              logs/wsd-awc8-h40.records.jsonl.gz: string
              curves/awc4-cosine-h40.jsonl.gz: string
              logs/cosine-sync-h20.records.jsonl.gz: string
              provenance/repack.json: string
              logs/wsd-sync-h40.records.jsonl.gz: string
              worker-comparison.pdf: string
              sources/campaigns.jsonl.gz: string
              logs/wsd-awc8-h40.metrics.jsonl.gz: string
              dataset.json: string
              curves/awc8-cosine-h20.jsonl.gz: string
              schedule-comparison-sync.pdf: string
              schedule-comparison-sync.png: string
              curves/sync-wsd-h40.jsonl.gz: string
              curves/awc8-cosine-h80.jsonl.gz: string
              worker-comparison-mob
              ...
              sonl.gz: string
              curves/awc8-wsd-h160.jsonl.gz: string
              schedule-comparison-sync.svg: string
              curves/awc8-wsd-h120.jsonl.gz: string
              curves/sync-wsd-h20.jsonl.gz: string
              curves/sync-cosine-h20.jsonl.gz: string
              logs/cosine-awc4-h20.records.jsonl.gz: string
              logs/wsd-sync-h120.metrics.jsonl.gz: string
              manifest.json: string
              logs/wsd-sync-h160.metrics.jsonl.gz: string
              runs.json: string
              schedule-comparison-awc8.png: string
              logs/cosine-awc8-h80.metrics.jsonl.gz: string
              logs/wsd-sync-h160.records.jsonl.gz: string
              logs/wsd-sync-h80.metrics.jsonl.gz: string
              logs/wsd-awc8-h80.records.jsonl.gz: string
              logs/cosine-awc4-h40.metrics.jsonl.gz: string
              logs/cosine-awc8-h80.records.jsonl.gz: string
              worker-comparison.svg: string
              schedule-comparison.pdf: string
              logs/cosine-awc4-h80.records.jsonl.gz: string
              winners.json: string
              worker-comparison-mobile.svg: string
              logs/cosine-awc4-h80.metrics.jsonl.gz: string
              curves/sync-wsd-h80.jsonl.gz: string
              schedule-comparison.png: string
              winners.csv: string
              worker-comparison.png: string
              curves/awc8-cosine-h40.jsonl.gz: string
              logs/wsd-sync-h120.records.jsonl.gz: string
              logs/cosine-awc4-h40.records.jsonl.gz: string
              logs/wsd-awc8-h20.records.jsonl.gz: string
              gh200-throughput.png: string
              logs/wsd-sync-h40.metrics.jsonl.gz: string
              curves/sync-wsd-h120.jsonl.gz: string
              runs.csv: string
              performance.csv: string
              performance.json: string
              shared/environments.jsonl.gz: string
              curves/sync-cosine-h80.jsonl.gz: string
              performance-runs.json: string
              performance-runs.csv: string
              to
              {'configurations.csv': Value('string'), 'configurations.json': Value('string'), 'curves/awc4-cosine-h20.jsonl.gz': Value('string'), 'curves/awc4-cosine-h40.jsonl.gz': Value('string'), 'curves/awc4-cosine-h80.jsonl.gz': Value('string'), 'curves/awc8-cosine-h20.jsonl.gz': Value('string'), 'curves/awc8-cosine-h40.jsonl.gz': Value('string'), 'curves/awc8-cosine-h80.jsonl.gz': Value('string'), 'curves/awc8-wsd-h120.jsonl.gz': Value('string'), 'curves/awc8-wsd-h160.jsonl.gz': Value('string'), 'curves/awc8-wsd-h20.jsonl.gz': Value('string'), 'curves/awc8-wsd-h40.jsonl.gz': Value('string'), 'curves/awc8-wsd-h80.jsonl.gz': Value('string'), 'curves/sync-cosine-h20.jsonl.gz': Value('string'), 'curves/sync-cosine-h40.jsonl.gz': Value('string'), 'curves/sync-cosine-h80.jsonl.gz': Value('string'), 'curves/sync-wsd-h120.jsonl.gz': Value('string'), 'curves/sync-wsd-h160.jsonl.gz': Value('string'), 'curves/sync-wsd-h20.jsonl.gz': Value('string'), 'curves/sync-wsd-h40.jsonl.gz': Value('string'), 'curves/sync-wsd-h80.jsonl.gz': Value('string'), 'dataset.json': Value('string'), 'gh200-throughput.pdf': Value('string'), 'gh200-throughput.png': Value('string'), 'gh200-throughput.svg': Value('string'), 'logs/cosine-awc4-h20.metrics.jsonl.gz': Value('string'), 'logs/cosine-awc4-h20.records.jsonl.gz': Value('string'), 'logs/cosine-awc4-h40.metrics.jsonl.gz': Value('string'), 'logs/cosine-awc4-h40.records.jsonl.gz': Value('string'), 'logs/cosine-awc4-h80.metrics.jsonl.gz': Value('string'), 'logs/cosine
              ...
              onl.gz': Value('string'), 'logs/wsd-sync-h40.records.jsonl.gz': Value('string'), 'logs/wsd-sync-h80.metrics.jsonl.gz': Value('string'), 'logs/wsd-sync-h80.records.jsonl.gz': Value('string'), 'manifest.json': Value('string'), 'matched.csv': Value('string'), 'matched.json': Value('string'), 'performance-runs.csv': Value('string'), 'performance-runs.json': Value('string'), 'performance.csv': Value('string'), 'performance.json': Value('string'), 'provenance/pre-repack-checksums.json.gz': Value('string'), 'provenance/repack.json': Value('string'), 'runs.csv': Value('string'), 'runs.json': Value('string'), 'schedule-comparison-awc8.pdf': Value('string'), 'schedule-comparison-awc8.png': Value('string'), 'schedule-comparison-awc8.svg': Value('string'), 'schedule-comparison-sync.pdf': Value('string'), 'schedule-comparison-sync.png': Value('string'), 'schedule-comparison-sync.svg': Value('string'), 'schedule-comparison.pdf': Value('string'), 'schedule-comparison.png': Value('string'), 'schedule-comparison.svg': Value('string'), 'shared/environments.jsonl.gz': Value('string'), 'sources/campaigns.jsonl.gz': Value('string'), 'sources.json': Value('string'), 'winners.csv': Value('string'), 'winners.json': Value('string'), 'worker-comparison-mobile.pdf': Value('string'), 'worker-comparison-mobile.png': Value('string'), 'worker-comparison-mobile.svg': Value('string'), 'worker-comparison.pdf': Value('string'), 'worker-comparison.png': Value('string'), 'worker-comparison.svg': 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.

tiny-llm retained training evidence

1782 seed/horizon records across 594 configurations: cosine-to-zero and WSD schedules, synchronous and four- and eight-worker decentralized training, at 20 to 160 global tokens per parameter, for 20.4M-parameter models on C4. Snapshot 2026-09-17.

This mirrors doc/data/current-training/ in WangZesen/tiny-llm. The website built from it is at https://wangzesen.github.io/tiny-llm/.

Layout

Retained artifacts are grouped into gzipped JSONL containers, one line per artifact:

{"path":"logs/cosine/sync/h20/lr-0.01-b1-0.9-b2-0.99/s42/metrics.jsonl",
  "bytes":71410,"sha256":"...","text":"<the artifact verbatim>"}

sources.json is the index: it maps every artifact to its container and records the digest of the original file. Metrics containers concatenate one gzip member per line, so offset and compressedBytes address a single run's log directly; every other container is a single stream. Placement is a pure rule, so a misfiled container is detectable.

What is retained, and what is not

9836 artifacts are byte-identical to the files written during training. The 1296 metrics.jsonl logs had two members deleted from every train event: local_grad_norms, local_losses. These are per-worker diagnostics for the packed models; no published figure, table or curve reads them, and they were 37% of the logged bytes.

The transform is a deletion, not a re-serialization: no other byte of any line, and no line, was changed, added or reordered. sources.json.sha256 remains the digest of the original file and can no longer be recomputed from this copy for those logs — but for the cosine runs it stays independently attested by the retained receipt.json, which records the trainer's own hash of metrics.jsonl and is checked on every verification run. The complete pre-repack bundle, with the arrays intact, is this repo at revision per-file-original.

Verify and regenerate

uv run python scripts/export_publication.py --check --bundle hf://zesen-kth/tiny-llm

No GPU, campaign directory or checkpoint is needed. Verification checks every retained byte against sources.json and every derived file against checksums.json.

Downloads last month: 35

Total file size:

98.3 MB