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
tag: string
boot: string
pid: int64
final: bool
elapsed_s: double
select_experts_calls: int64
moe_layers: int64
n_experts: int64
total_expert_activations: int64
tokens_per_layer: list<item: int64>
  child 0, item: int64
per_layer: list<item: struct<moe_layer: int64, model_layer: int64, total: int64, used_experts: int64, max_over_ (... 52 chars omitted)
  child 0, item: struct<moe_layer: int64, model_layer: int64, total: int64, used_experts: int64, max_over_mean: doubl (... 40 chars omitted)
      child 0, moe_layer: int64
      child 1, model_layer: int64
      child 2, total: int64
      child 3, used_experts: int64
      child 4, max_over_mean: double
      child 5, top16: list<item: list<item: int64>>
          child 0, item: list<item: int64>
              child 0, item: int64
model_layer_offset: int64
experts_per_layer: int64
generations: int64
to
{'moe_layers': Value('int64'), 'experts_per_layer': Value('int64'), 'model_layer_offset': Value('int64'), 'generations': Value('int64'), 'total_expert_activations': Value('int64'), 'per_layer': List({'model_layer': Value('int64'), 'total': Value('int64'), 'used_experts': Value('int64'), 'max_over_mean': Value('float64'), 'topk_coverage': {'8': Value('float64'), '16': Value('float64'), '32': Value('float64'), '64': Value('float64'), '128': Value('float64')}, 'counts': List(Value('int64'))})}
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
              tag: string
              boot: string
              pid: int64
              final: bool
              elapsed_s: double
              select_experts_calls: int64
              moe_layers: int64
              n_experts: int64
              total_expert_activations: int64
              tokens_per_layer: list<item: int64>
                child 0, item: int64
              per_layer: list<item: struct<moe_layer: int64, model_layer: int64, total: int64, used_experts: int64, max_over_ (... 52 chars omitted)
                child 0, item: struct<moe_layer: int64, model_layer: int64, total: int64, used_experts: int64, max_over_mean: doubl (... 40 chars omitted)
                    child 0, moe_layer: int64
                    child 1, model_layer: int64
                    child 2, total: int64
                    child 3, used_experts: int64
                    child 4, max_over_mean: double
                    child 5, top16: list<item: list<item: int64>>
                        child 0, item: list<item: int64>
                            child 0, item: int64
              model_layer_offset: int64
              experts_per_layer: int64
              generations: int64
              to
              {'moe_layers': Value('int64'), 'experts_per_layer': Value('int64'), 'model_layer_offset': Value('int64'), 'generations': Value('int64'), 'total_expert_activations': Value('int64'), 'per_layer': List({'model_layer': Value('int64'), 'total': Value('int64'), 'used_experts': Value('int64'), 'max_over_mean': Value('float64'), 'topk_coverage': {'8': Value('float64'), '16': Value('float64'), '32': Value('float64'), '64': Value('float64'), '128': Value('float64')}, 'counts': List(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.

GLM-5.3-Flash — per-layer MoE expert usage on ARC-AGI-3

Per-layer, per-expert routing counts collected while GLM-5.3-Flash (320B total / 18B active) played the 25 public official ARC-AGI-3 games through the TUFA duck-harness (TAAF).

Collected by instrumenting vLLM's FusedMoERouter.select_experts and accumulating a [MoE layer x expert] count matrix on-GPU.

Model / run configuration

model zai-org/GLM-5.3-Flash (FP8)
serving vLLM (ROCm), tensor-parallel 4
hardware 4x AMD Instinct MI355X (gfx950)
MoE layers 43 (model layers 3–45)
experts / layer 288
experts / token 8 (+1 shared)
total expert activations 28,423,475,384
server generations merged 14

Files

  • expert_counts_merged.npyint64[43, 288] activation counts. Row i is model layer i + 3.
  • expert_counts_merged.json — same data plus per-layer summary stats and top-K coverage fractions (useful for sizing an expert cache).
  • raw/ — the per-rank, per-generation dumps this was aggregated from.

Aggregation

Dumps sharing a boot id are tensor-parallel rank replicas (identical routing) so only one is taken; dumps from different boot ids are separate server generations and are summed.

Top-K routing coverage (fraction of routing mass in the K hottest experts)

layer K=8 K=16 K=32 K=64 K=128
3 9.0% 15.6% 26.0% 41.2% 63.5%
9 17.8% 24.6% 35.3% 49.6% 70.6%
15 17.7% 26.9% 39.3% 55.3% 75.9%
21 20.6% 31.4% 44.6% 60.2% 79.3%
27 13.3% 22.4% 35.6% 53.5% 75.7%
33 21.2% 31.2% 42.1% 56.6% 76.5%
39 17.2% 26.8% 39.6% 54.6% 75.0%
45 19.6% 28.4% 40.6% 58.1% 79.4%
Downloads last month
20