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
schema_version: int64
timestamp_utc: string
model: string
sglang_commit: string
tp_size: int64
prompt: string
max_tokens: int64
temperature: int64
diff_threshold: double
dumper_filter: string
num_tensor_files: int64
pass_label: string
source: string
num_layers_passed: int64
num_layers_compared: int64
max_rel_diff: double
run_path: string
num_layers_failed: int64
date: timestamp[s]
hardware: string
ci_run_id: string
push_index: int64
to
{'date': Value('timestamp[s]'), 'model': Value('string'), 'run_path': Value('string'), 'sglang_commit': Value('string'), 'push_index': Value('int64'), 'hardware': Value('string'), 'tp_size': Value('int64'), 'pass_label': Value('string'), 'ci_run_id': Value('string'), 'num_layers_compared': Value('int64'), 'num_layers_passed': Value('int64'), 'num_layers_failed': Value('int64'), 'max_rel_diff': Value('float64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
                  return get_rows(
                         ^^^^^^^^^
                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 77, in get_rows
                  rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2690, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2227, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2251, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 494, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 384, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 299, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 128, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2321, in table_cast
                  return cast_table_to_schema(table, schema)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2249, in cast_table_to_schema
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              schema_version: int64
              timestamp_utc: string
              model: string
              sglang_commit: string
              tp_size: int64
              prompt: string
              max_tokens: int64
              temperature: int64
              diff_threshold: double
              dumper_filter: string
              num_tensor_files: int64
              pass_label: string
              source: string
              num_layers_passed: int64
              num_layers_compared: int64
              max_rel_diff: double
              run_path: string
              num_layers_failed: int64
              date: timestamp[s]
              hardware: string
              ci_run_id: string
              push_index: int64
              to
              {'date': Value('timestamp[s]'), 'model': Value('string'), 'run_path': Value('string'), 'sglang_commit': Value('string'), 'push_index': Value('int64'), 'hardware': Value('string'), 'tp_size': Value('int64'), 'pass_label': Value('string'), 'ci_run_id': Value('string'), 'num_layers_compared': Value('int64'), 'num_layers_passed': Value('int64'), 'num_layers_failed': Value('int64'), 'max_rel_diff': 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.

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

Check out the documentation for more information.

sglang nightly precision baselines

Day-over-day per-layer hidden_states dumps captured by sglang's debug_utils/dumper, used as rolling baselines for the nightly precision-regression CI job.

Layout

<model_id_sanitized>/<YYYY>/<MM>/<DD>/run-<sglang_sha7>/
    meta.json                # schema_version=1 run config + result summary
    comparator_report.jsonl  # one ComparisonTensorRecord per line + summary
    tensors/                 # raw .pt files (torch.load-able dict)
        step=*___rank=*___dump_index=*___name=...pt

model_id_sanitized replaces / with __ (HF-friendly).

manifest.jsonl at the repo root: one JSONL row per run, append-only.

Dump format

Each .pt file is torch.save({"value": tensor, "meta": dict}). meta carries step / rank / dump_index / name / sglang_parallel_info / layer_id etc; the filename also encodes the discriminative subset for fast read_meta() parsing.

Diff strategy

Hidden states are captured at each DecoderLayer entry (inputs.1), which is a per-tp-rank partial sum (sglang's LayerCommunicator defers the cross-layer allreduce to the next layer's prepare_*). The comparator's --override-dims bs h[tp:partial] sums the 8 rank tensors on the h axis before diffing the target run against the previous-day baseline; this is deterministic for fixed TP/seed/prompt.

Downloads last month
8