Hugging Face's logo

Datasets:
0xSero
/
glm-5.3-reap-fidelity-study

Languages:
English
ArXiv:
Tags:
reap
mixture-of-experts
glm
expert-pruning
evaluation
kl-divergence
License:
Dataset card Files
xet
 Community
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
reference: string
candidate: string
metric: string
direction: string
units: string
value: double
positions: int64
vocab: int64
windows: int64
ci95_low: double
ci95_high: double
bootstrap_b: int64
bootstrap_seed: int64
interval_kind: string
cluster_unit: string
estimator: struct<accumulation_dtype: string, support: string, stored_dtype: string>
  child 0, accumulation_dtype: string
  child 1, support: string
  child 2, stored_dtype: string
median_tokenwise: double
p95_tokenwise: double
per_window_mean: list<item: double>
  child 0, item: double
summary: struct<model: string, n: int64, correct: int64, accuracy: double, truncated: int64, no_answer: int64 (... 233 chars omitted)
  child 0, model: string
  child 1, n: int64
  child 2, correct: int64
  child 3, accuracy: double
  child 4, truncated: int64
  child 5, no_answer: int64
  child 6, mean_tokens: double
  child 7, wall_seconds: double
  child 8, by_domain: struct<Physics: struct<correct: int64, n: int64, acc: double>, Chemistry: struct<correct: int64, n:  (... 76 chars omitted)
      child 0, Physics: struct<correct: int64, n: int64, acc: double>
          child 0, correct: int64
          child 1, n: int64
          child 2, acc: double
      child 1, Chemistry: struct<correct: int64, n: int64, acc: double>
          child 0, correct: int64
          child 1, n: int64
          child 2, acc: double
      child 2, Biology: struct<correct: int64, n: int64, acc: double>
          child 0, correct: int64
          child 1, n: int64
          child 2, acc: double
results: list<item: struct<i: int64, domain: string, gold: string, pred: string, finish: string, tokens: int6 (... 17 chars omitted)
  child 0, item: struct<i: int64, domain: string, gold: string, pred: string, finish: string, tokens: int64, text: st (... 5 chars omitted)
      child 0, i: int64
      child 1, domain: string
      child 2, gold: string
      child 3, pred: string
      child 4, finish: string
      child 5, tokens: int64
      child 6, text: string
to
{'summary': {'model': Value('string'), 'n': Value('int64'), 'correct': Value('int64'), 'accuracy': Value('float64'), 'truncated': Value('int64'), 'no_answer': Value('int64'), 'mean_tokens': Value('float64'), 'wall_seconds': Value('float64'), 'by_domain': {'Physics': {'correct': Value('int64'), 'n': Value('int64'), 'acc': Value('float64')}, 'Chemistry': {'correct': Value('int64'), 'n': Value('int64'), 'acc': Value('float64')}, 'Biology': {'correct': Value('int64'), 'n': Value('int64'), 'acc': Value('float64')}}}, 'results': List({'i': Value('int64'), 'domain': Value('string'), 'gold': Value('string'), 'pred': Value('string'), 'finish': Value('string'), 'tokens': Value('int64'), 'text': 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
              reference: string
              candidate: string
              metric: string
              direction: string
              units: string
              value: double
              positions: int64
              vocab: int64
              windows: int64
              ci95_low: double
              ci95_high: double
              bootstrap_b: int64
              bootstrap_seed: int64
              interval_kind: string
              cluster_unit: string
              estimator: struct<accumulation_dtype: string, support: string, stored_dtype: string>
                child 0, accumulation_dtype: string
                child 1, support: string
                child 2, stored_dtype: string
              median_tokenwise: double
              p95_tokenwise: double
              per_window_mean: list<item: double>
                child 0, item: double
              summary: struct<model: string, n: int64, correct: int64, accuracy: double, truncated: int64, no_answer: int64 (... 233 chars omitted)
                child 0, model: string
                child 1, n: int64
                child 2, correct: int64
                child 3, accuracy: double
                child 4, truncated: int64
                child 5, no_answer: int64
                child 6, mean_tokens: double
                child 7, wall_seconds: double
                child 8, by_domain: struct<Physics: struct<correct: int64, n: int64, acc: double>, Chemistry: struct<correct: int64, n:  (... 76 chars omitted)
                    child 0, Physics: struct<correct: int64, n: int64, acc: double>
                        child 0, correct: int64
                        child 1, n: int64
                        child 2, acc: double
                    child 1, Chemistry: struct<correct: int64, n: int64, acc: double>
                        child 0, correct: int64
                        child 1, n: int64
                        child 2, acc: double
                    child 2, Biology: struct<correct: int64, n: int64, acc: double>
                        child 0, correct: int64
                        child 1, n: int64
                        child 2, acc: double
              results: list<item: struct<i: int64, domain: string, gold: string, pred: string, finish: string, tokens: int6 (... 17 chars omitted)
                child 0, item: struct<i: int64, domain: string, gold: string, pred: string, finish: string, tokens: int64, text: st (... 5 chars omitted)
                    child 0, i: int64
                    child 1, domain: string
                    child 2, gold: string
                    child 3, pred: string
                    child 4, finish: string
                    child 5, tokens: int64
                    child 6, text: string
              to
              {'summary': {'model': Value('string'), 'n': Value('int64'), 'correct': Value('int64'), 'accuracy': Value('float64'), 'truncated': Value('int64'), 'no_answer': Value('int64'), 'mean_tokens': Value('float64'), 'wall_seconds': Value('float64'), 'by_domain': {'Physics': {'correct': Value('int64'), 'n': Value('int64'), 'acc': Value('float64')}, 'Chemistry': {'correct': Value('int64'), 'n': Value('int64'), 'acc': Value('float64')}, 'Biology': {'correct': Value('int64'), 'n': Value('int64'), 'acc': Value('float64')}}}, 'results': List({'i': Value('int64'), 'domain': Value('string'), 'gold': Value('string'), 'pred': Value('string'), 'finish': Value('string'), 'tokens': Value('int64'), 'text': 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.

GLM-5.3 REAP Fidelity Study

The evidence behind the GLM-5.3 REAP model series: how each pruned cut was measured, and why it prunes the way it does.

What we measured, and how

Every number here is KL divergence vs the full BF16 model — a token-by-token “how differently do these two models predict the next token” score over a sealed 25-prompt panel across the full 154,880-token vocabulary. 0 = identical, lower = closer. It is a far more sensitive test than benchmark accuracy: it catches quality loss long before a leaderboard would.

Finding 1 — which experts to keep matters more than anything

Pruning to the same size (keep 192 of 256 experts), only changing how experts are ranked:

Criterion KL vs BF16
max-over-domain (massmax) 0.361 ← used by the series
domain-balanced 0.399
reap × max-over-domain 0.406
mean routed mass 0.569
frequency only 0.635 the “obvious” choice
random 0.685
stock REAP (frequency-blind mean) 0.889

Ranking experts by their largest share of any single domain’s work (so every domain keeps its specialists) beats the frequency-based choice almost 2×. That is the whole idea behind the series.

Finding 2 — the size ladder

Same criterion, different number of experts kept:

Cut Experts KL vs BF16
base (unpruned, 3-bit) 256 0.089
661B 224 0.195
615B 208 0.283
569B 192 0.361
533B 180 0.428
500B 168 0.511

Finding 3 — the prune dominates the quant format

At the same expert count, EXL3 (3-bit) and W4A16 (INT4) land in the same place — 0.361 vs 0.357 at keep-192, 0.511 vs 0.506 at keep-168. The pruning drives the fidelity; the quant format barely moves it. So pick the format your hardware wants (EXL3 for Blackwell/consumer, W4A16 for Hopper) and choose the size by how much fidelity you can spend.

Contents

  • kld/ — every KL run: the criterion sweep, the size ladder, EXL3 vs W4A16, NVFP4.
  • plans/ — the exact kept-expert lists per layer.
  • gpqa/ — GPQA-Diamond runs for spot-checking downstream accuracy.
  • panel.json, frontier_table.txt, *_summary.json — the sealed eval panel and roll-ups.

Inputs come from glm-5.3-reap-observations-v1.

Credits

Downloads last month
409
Total file size:
10.4 MB

Collection including 0xSero/glm-5.3-reap-fidelity-study

Paper for 0xSero/glm-5.3-reap-fidelity-study