Hugging Face's logo

Datasets:
0arch-io
/
kisoku-1.6b-eval

Languages:
English
Tags:
evaluation
lm-evaluation-harness
kisoku
benchmark
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
gsm8k: list<item: string>
  child 0, item: string
hellaswag: list<item: string>
  child 0, item: string
piqa: list<item: string>
  child 0, item: string
winogrande: list<item: string>
  child 0, item: string
arc_easy: list<item: string>
  child 0, item: string
arc_challenge: list<item: string>
  child 0, item: string
mmlu: list<item: string>
  child 0, item: string
mmlu_probed: list<item: string>
  child 0, item: string
bench: struct<gsm8k: struct<items: int64, probed: int64, any_probe_hit: int64, all_probes_hit: int64, pct_a (... 3522 chars omitted)
  child 0, gsm8k: struct<items: int64, probed: int64, any_probe_hit: int64, all_probes_hit: int64, pct_any: double, pc (... 233 chars omitted)
      child 0, items: int64
      child 1, probed: int64
      child 2, any_probe_hit: int64
      child 3, all_probes_hit: int64
      child 4, pct_any: double
      child 5, pct_all: double
      child 6, items_hit_by_dataset: struct<finemath-4plus: int64, nemotron-cc-v2.1-hqs: int64, megamath-web-pro: int64, nemotron-cc-math (... 66 chars omitted)
          child 0, finemath-4plus: int64
          child 1, nemotron-cc-v2.1-hqs: int64
          child 2, megamath-web-pro: int64
          child 3, nemotron-cc-math-4plus: int64
          child 4, open-web-math: int64
          child 5, ultra-fineweb-en-text: int64
      child 7, hit_ids: list<item: string>
          child 0, item: string
  child 1, mmlu: struct<items: int64, probed: int64, any_probe_hit: int64, all_probes_hit: int64, pct_an
...
, triviaqa: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
      child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
          child 0, id: string
          child 1, ds: string
          child 2, shard: string
          child 3, rec: int64
          child 4, ctx: string
  child 7, arc_challenge: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
      child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
          child 0, id: string
          child 1, ds: string
          child 2, shard: string
          child 3, rec: int64
          child 4, ctx: string
  child 8, humaneval_prompt: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
      child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
          child 0, id: string
          child 1, ds: string
          child 2, shard: string
          child 3, rec: int64
          child 4, ctx: string
  child 9, winogrande: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
      child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
          child 0, id: string
          child 1, ds: string
          child 2, shard: string
          child 3, rec: int64
          child 4, ctx: string
method: struct<minlen: int64, window: int64>
  child 0, minlen: int64
  child 1, window: int64
to
{'corpus': {'finemath-4plus': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'longctx-code-repos': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'longctx-pg19': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'longctx-science-pdfs': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'megamath-web-pro': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'nemotron-cc-math-4plus': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'nemotron-cc-v2.1-hqs': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'open-web-math': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'openthoughts3-text': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'starcoderdata-text': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'ultra-fineweb-en-text': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}}, 'bench': {'gsm8k': {'items': Value('int64'), 'probed': Value('int64'), 'any_probe_hit': Value('int64'), 'all_probes_hit': Value('int64'), 'pct_any': Value('float64'), 'pct_all': Value('float64'), 'items_hit_by_dataset': {'finemath-4plus': Value('int64'), 'nemotron-cc-v2.1-hqs': Value('int64'), 'megamath-web-pro': Value('int64
...
lue('string'))}}, 'examples': {'gsm8k': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'mmlu': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'arc_easy': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'piqa': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'hellaswag': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'humaneval_solution': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'triviaqa': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'arc_challenge': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'humaneval_prompt': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'winogrande': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')})}, 'method': {'minlen': Value('int64'), 'window': 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 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, 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 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, 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
              gsm8k: list<item: string>
                child 0, item: string
              hellaswag: list<item: string>
                child 0, item: string
              piqa: list<item: string>
                child 0, item: string
              winogrande: list<item: string>
                child 0, item: string
              arc_easy: list<item: string>
                child 0, item: string
              arc_challenge: list<item: string>
                child 0, item: string
              mmlu: list<item: string>
                child 0, item: string
              mmlu_probed: list<item: string>
                child 0, item: string
              bench: struct<gsm8k: struct<items: int64, probed: int64, any_probe_hit: int64, all_probes_hit: int64, pct_a (... 3522 chars omitted)
                child 0, gsm8k: struct<items: int64, probed: int64, any_probe_hit: int64, all_probes_hit: int64, pct_any: double, pc (... 233 chars omitted)
                    child 0, items: int64
                    child 1, probed: int64
                    child 2, any_probe_hit: int64
                    child 3, all_probes_hit: int64
                    child 4, pct_any: double
                    child 5, pct_all: double
                    child 6, items_hit_by_dataset: struct<finemath-4plus: int64, nemotron-cc-v2.1-hqs: int64, megamath-web-pro: int64, nemotron-cc-math (... 66 chars omitted)
                        child 0, finemath-4plus: int64
                        child 1, nemotron-cc-v2.1-hqs: int64
                        child 2, megamath-web-pro: int64
                        child 3, nemotron-cc-math-4plus: int64
                        child 4, open-web-math: int64
                        child 5, ultra-fineweb-en-text: int64
                    child 7, hit_ids: list<item: string>
                        child 0, item: string
                child 1, mmlu: struct<items: int64, probed: int64, any_probe_hit: int64, all_probes_hit: int64, pct_an
              ...
              , triviaqa: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
                    child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
                        child 0, id: string
                        child 1, ds: string
                        child 2, shard: string
                        child 3, rec: int64
                        child 4, ctx: string
                child 7, arc_challenge: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
                    child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
                        child 0, id: string
                        child 1, ds: string
                        child 2, shard: string
                        child 3, rec: int64
                        child 4, ctx: string
                child 8, humaneval_prompt: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
                    child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
                        child 0, id: string
                        child 1, ds: string
                        child 2, shard: string
                        child 3, rec: int64
                        child 4, ctx: string
                child 9, winogrande: list<item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>>
                    child 0, item: struct<id: string, ds: string, shard: string, rec: int64, ctx: string>
                        child 0, id: string
                        child 1, ds: string
                        child 2, shard: string
                        child 3, rec: int64
                        child 4, ctx: string
              method: struct<minlen: int64, window: int64>
                child 0, minlen: int64
                child 1, window: int64
              to
              {'corpus': {'finemath-4plus': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'longctx-code-repos': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'longctx-pg19': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'longctx-science-pdfs': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'megamath-web-pro': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'nemotron-cc-math-4plus': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'nemotron-cc-v2.1-hqs': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'open-web-math': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'openthoughts3-text': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'starcoderdata-text': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}, 'ultra-fineweb-en-text': {'shards': Value('int64'), 'records': Value('int64'), 'text_gb': Value('float64')}}, 'bench': {'gsm8k': {'items': Value('int64'), 'probed': Value('int64'), 'any_probe_hit': Value('int64'), 'all_probes_hit': Value('int64'), 'pct_any': Value('float64'), 'pct_all': Value('float64'), 'items_hit_by_dataset': {'finemath-4plus': Value('int64'), 'nemotron-cc-v2.1-hqs': Value('int64'), 'megamath-web-pro': Value('int64
              ...
              lue('string'))}}, 'examples': {'gsm8k': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'mmlu': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'arc_easy': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'piqa': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'hellaswag': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'humaneval_solution': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'triviaqa': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'arc_challenge': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'humaneval_prompt': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')}), 'winogrande': List({'id': Value('string'), 'ds': Value('string'), 'shard': Value('string'), 'rec': Value('int64'), 'ctx': Value('string')})}, 'method': {'minlen': Value('int64'), 'window': 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.

Kisoku 1.6B evaluation results

Raw per-sample outputs of lm-evaluation-harness 0.4.13 for every model in the Kisoku 1.6B technical report, published so the report's tables can be checked from the source files rather than taken on trust. Each run folder holds the harness results_*.json (aggregate metrics, config, task versions) and one samples_<task>_*.jsonl per task (every prompt, model output and score).

Layout 

results/<run-name>/<model>/results_<timestamp>.json
results/<run-name>/<model>/samples_<task>_<timestamp>.jsonl
contamination/    contamination audit: scripts, patterns, hits, summaries, per-dataset outputs (out/)
logs/<batch>/     harness stream logs and the run scripts for each evaluation batch

The run folder name is the original name used when the evaluation was launched (model, suite, and for RULER the context length, for example kisoku-longC-core or qwen3-1.7b-yarn2-ruler-65536). Runs prefixed with _ are diagnostics, probes or alternate-precision checks. Suffix -clean marks runs on the decontaminated task variants. The same run reached us in more than one archive in several cases; the copies were byte-identical, so one copy is kept.

Runs

Run folder Model Tasks Results file date
_bf16-kisoku-v1-core kisoku-v1-3b-base hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-10-05
_bf16-kisoku-v1-gsm8k kisoku-v1-3b-base gsm8k 2026-10-05
_bf16-kisoku-v1-humaneval kisoku-v1-3b-base humaneval 2026-10-05
_diag kisoku-1.6b-base-longA-2500 ruler_cwe, ruler_vt, ruler_fwe, ruler_qa_squad 2026-10-01
_mn-base kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-01
_mn-clean _kisoku-longA-clean ruler (13 subtasks) 2026-10-01
_mn-cleanmin _kisoku-longA-cleanmin ruler (13 subtasks) 2026-10-01
_reppen-kisoku-longA-ruler-4096 kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-01
_reppen-kisoku-longA-ruler-8192 kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-01
_ruler-probe Qwen3-1.7B-Base ruler (13 subtasks) 2026-10-01
gemma-3-1b-bbh gemma-3-1b-pt bbh_fewshot (27 subtasks) 2026-09-30
gemma-3-1b-bbhws gemma-3-1b-pt bbh_ws (27 subtasks) 2026-10-01
gemma-3-1b-core gemma-3-1b-pt hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-10-01
gemma-3-1b-drop gemma-3-1b-pt drop 2026-09-30
gemma-3-1b-gsm8k gemma-3-1b-pt gsm8k 2026-10-01
gemma-3-1b-humaneval gemma-3-1b-pt humaneval 2026-10-01
gemma-3-1b-mmlu gemma-3-1b-pt mmlu (4 subtasks) 2026-10-01
gemma-3-1b-ruler-16384 gemma-3-1b-pt ruler (13 subtasks) 2026-10-02
gemma-3-1b-ruler-32768 gemma-3-1b-pt ruler (13 subtasks) 2026-10-02
gemma-3-1b-ruler-4096 gemma-3-1b-pt ruler (13 subtasks) 2026-10-01
gemma-3-1b-ruler-8192 gemma-3-1b-pt ruler (13 subtasks) 2026-10-01
gemma-3-1b-triviaqa gemma-3-1b-pt triviaqa 2026-09-30
granite-4.0-1b-ruler-131072 granite-4.0-1b-base ruler (13 subtasks) 2026-10-04
granite-4.0-1b-ruler-16384 granite-4.0-1b-base ruler (13 subtasks) 2026-10-02
granite-4.0-1b-ruler-32768 granite-4.0-1b-base ruler (13 subtasks) 2026-10-02
granite-4.0-1b-ruler-4096 granite-4.0-1b-base ruler (13 subtasks) 2026-10-02
granite-4.0-1b-ruler-65536 granite-4.0-1b-base ruler (13 subtasks) 2026-10-03
granite-4.0-1b-ruler-8192 granite-4.0-1b-base ruler (13 subtasks) 2026-10-02
kisoku-chat5-bbhws kisoku-1.6b-chat-sft005 bbh_ws (27 subtasks) 2026-10-05
kisoku-chat5-core kisoku-1.6b-chat-sft005 hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-10-05
kisoku-chat5-gsm8k kisoku-1.6b-chat-sft005 gsm8k 2026-10-05
kisoku-chat5-mmlu kisoku-1.6b-chat-sft005 mmlu (4 subtasks) 2026-10-05
kisoku-chat5-triviaqa kisoku-1.6b-chat-sft005 triviaqa 2026-10-05
kisoku-chat5-yarn2-ruler-32768 kisoku-1.6b-chat-sft005-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-chat5-yarn2-ruler-4096 kisoku-1.6b-chat-sft005-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-chat5-yarn2-ruler-65536 kisoku-1.6b-chat-sft005-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-longA-ruler-16384 kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-01
kisoku-longA-ruler-32768 kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-02
kisoku-longA-ruler-4096 kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-01
kisoku-longA-ruler-8192 kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-01
kisoku-longA-ruler32k kisoku-1.6b-base-longA-2500 ruler (13 subtasks) 2026-10-01
kisoku-longC-bbhws kisoku-1.6b-base-longC-1299 bbh_ws (27 subtasks) 2026-10-07
kisoku-longC-core kisoku-1.6b-base-longC-1299 hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-10-07
kisoku-longC-box2plain-ruler-131072 kisoku-1.6b-base-longC-1299 ruler (13 subtasks) 2026-10-07
kisoku-longC-box2plain-ruler-65536 kisoku-1.6b-base-longC-1299 ruler (13 subtasks) 2026-10-05
kisoku-longC-gsm8k kisoku-1.6b-base-longC-1299 gsm8k 2026-10-07
kisoku-longC-humaneval kisoku-1.6b-base-longC-1299 humaneval 2026-10-07
kisoku-longC-mmlu kisoku-1.6b-base-longC-1299 mmlu (4 subtasks) 2026-10-07
kisoku-longC-ruler-16384 kisoku-1.6b-base-longC-1299 ruler (13 subtasks) 2026-10-05
kisoku-longC-ruler-32768 kisoku-1.6b-base-longC-1299 ruler (13 subtasks) 2026-10-05
kisoku-longC-ruler-4096 kisoku-1.6b-base-longC-1299 ruler (13 subtasks) 2026-10-05
kisoku-longC-ruler-65536 kisoku-1.6b-base-longC-1299 ruler (13 subtasks) 2026-10-04
kisoku-longC-ruler-8192 kisoku-1.6b-base-longC-1299 ruler (13 subtasks) 2026-10-05
kisoku-longC-triviaqa kisoku-1.6b-base-longC-1299 triviaqa 2026-10-07
kisoku-longC-yarn-temp-ruler-65536 kisoku-longC-yarn-temp ruler (13 subtasks) 2026-10-06
kisoku-longC-yarn1.5-ruler-65536 kisoku-longC-yarn1.5 ruler (13 subtasks) 2026-10-05
kisoku-longC-yarn2-interp-ruler-65536 kisoku-longC-yarn2-interp ruler (13 subtasks) 2026-10-05
kisoku-longC-yarn2-ruler-131072 kisoku-1.6b-base-longC-1299-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-longC-yarn2-ruler-16384 kisoku-1.6b-base-longC-1299-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-longC-yarn2-ruler-32768 kisoku-1.6b-base-longC-1299-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-longC-yarn2-ruler-4096 kisoku-1.6b-base-longC-1299-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-longC-yarn2-ruler-65536 kisoku-1.6b-base-longC-1299-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-longC-yarn2-ruler-8192 kisoku-1.6b-base-longC-1299-yarn2 ruler (13 subtasks) 2026-10-05
kisoku-s1-bbh kisoku-1.6b-base-s1-198999 bbh_fewshot (27 subtasks) 2026-09-30
kisoku-s1-bbhws-clean kisoku-1.6b-base-s1-198999 bbh_ws (27 subtasks) 2026-10-01
kisoku-s1-bbhws kisoku-1.6b-base-s1-198999 bbh_ws (27 subtasks) 2026-10-01
kisoku-s1-core kisoku-1.6b-base-s1-198999 hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-09-30
kisoku-s1-drop kisoku-1.6b-base-s1-198999 drop 2026-09-30
kisoku-s1-gsm8k-clean kisoku-1.6b-base-s1-198999 gsm8k 2026-10-01
kisoku-s1-gsm8k kisoku-1.6b-base-s1-198999 gsm8k 2026-09-30
kisoku-s1-humaneval-clean kisoku-1.6b-base-s1-198999 humaneval 2026-10-01
kisoku-s1-humaneval kisoku-1.6b-base-s1-198999 humaneval 2026-09-30
kisoku-s1-mmlu kisoku-1.6b-base-s1-198999 mmlu (4 subtasks) 2026-10-01
kisoku-s1-triviaqa-clean kisoku-1.6b-base-s1-198999 triviaqa 2026-10-01
kisoku-s1-triviaqa kisoku-1.6b-base-s1-198999 triviaqa 2026-09-30
kisoku-s3-bbh kisoku-1.6b-base-s3-242999 bbh_fewshot (27 subtasks) 2026-09-30
kisoku-s3-bbhws-clean kisoku-1.6b-base-s3-242999 bbh_ws (27 subtasks) 2026-10-01
kisoku-s3-bbhws kisoku-1.6b-base-s3-242999 bbh_ws (27 subtasks) 2026-10-01
kisoku-s3-core kisoku-1.6b-base-s3-242999 hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-09-30
kisoku-s3-drop kisoku-1.6b-base-s3-242999 drop 2026-09-30
kisoku-s3-gsm8k-clean kisoku-1.6b-base-s3-242999 gsm8k 2026-10-01
kisoku-s3-gsm8k kisoku-1.6b-base-s3-242999 gsm8k 2026-09-30
kisoku-s3-humaneval-clean kisoku-1.6b-base-s3-242999 humaneval 2026-10-01
kisoku-s3-humaneval kisoku-1.6b-base-s3-242999 humaneval 2026-09-30
kisoku-s3-mmlu kisoku-1.6b-base-s3-242999 mmlu (4 subtasks) 2026-09-30
kisoku-s3-triviaqa-clean kisoku-1.6b-base-s3-242999 triviaqa 2026-10-01
kisoku-s3-triviaqa kisoku-1.6b-base-s3-242999 triviaqa 2026-09-30
kisoku-v1-bbhws kisoku-v1-3b-base bbh_ws (27 subtasks) 2026-10-05
kisoku-v1-core kisoku-v1-3b-base hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-10-05
kisoku-v1-gsm8k kisoku-v1-3b-base gsm8k 2026-10-05
kisoku-v1-humaneval kisoku-v1-3b-base humaneval 2026-10-05
kisoku-v1-mmlu kisoku-v1-3b-base mmlu (4 subtasks) 2026-10-05
kisoku-v1-triviaqa kisoku-v1-3b-base triviaqa 2026-10-05
lfm2.5-1.2b-ruler-16384 LFM2.5-1.2B-Base ruler (13 subtasks) 2026-10-02
lfm2.5-1.2b-ruler-32768 LFM2.5-1.2B-Base ruler (13 subtasks) 2026-10-02
lfm2.5-1.2b-ruler-4096 LFM2.5-1.2B-Base ruler (13 subtasks) 2026-10-01
lfm2.5-1.2b-ruler-65536 LFM2.5-1.2B-Base ruler (13 subtasks) 2026-10-04
lfm2.5-1.2b-ruler-8192 LFM2.5-1.2B-Base ruler (13 subtasks) 2026-10-01
llama-3.2-1b-bbh Llama-3.2-1B bbh_fewshot (27 subtasks) 2026-09-30
llama-3.2-1b-bbhws Llama-3.2-1B bbh_ws (27 subtasks) 2026-10-01
llama-3.2-1b-core Llama-3.2-1B hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-09-30
llama-3.2-1b-box2sanity-ruler-4096 Llama-3.2-1B ruler (13 subtasks) 2026-10-05
llama-3.2-1b-drop Llama-3.2-1B drop 2026-09-30
llama-3.2-1b-gsm8k Llama-3.2-1B gsm8k 2026-09-30
llama-3.2-1b-humaneval Llama-3.2-1B humaneval 2026-09-30
llama-3.2-1b-mmlu Llama-3.2-1B mmlu (4 subtasks) 2026-09-30
llama-3.2-1b-ruler-131072 Llama-3.2-1B ruler (13 subtasks) 2026-10-04
llama-3.2-1b-ruler-16384 Llama-3.2-1B ruler (13 subtasks) 2026-10-02
llama-3.2-1b-ruler-32768 Llama-3.2-1B ruler (13 subtasks) 2026-10-02
llama-3.2-1b-ruler-4096 Llama-3.2-1B ruler (13 subtasks) 2026-10-01
llama-3.2-1b-ruler-65536 Llama-3.2-1B ruler (13 subtasks) 2026-10-03
llama-3.2-1b-ruler-8192 Llama-3.2-1B ruler (13 subtasks) 2026-10-01
llama-3.2-1b-triviaqa Llama-3.2-1B triviaqa 2026-09-30
qwen2.5-1.5b-bbhws-clean Qwen2.5-1.5B bbh_ws (27 subtasks) 2026-10-01
qwen2.5-1.5b-bbhws Qwen2.5-1.5B bbh_ws (27 subtasks) 2026-10-01
qwen2.5-1.5b-core Qwen2.5-1.5B hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-09-30
qwen2.5-1.5b-gsm8k-clean Qwen2.5-1.5B gsm8k 2026-10-01
qwen2.5-1.5b-gsm8k Qwen2.5-1.5B gsm8k 2026-09-30
qwen2.5-1.5b-humaneval-clean Qwen2.5-1.5B humaneval 2026-10-01
qwen2.5-1.5b-humaneval Qwen2.5-1.5B humaneval 2026-09-30
qwen2.5-1.5b-mmlu Qwen2.5-1.5B mmlu (4 subtasks) 2026-09-30
qwen2.5-1.5b-triviaqa-clean Qwen2.5-1.5B triviaqa 2026-10-01
qwen2.5-1.5b-triviaqa Qwen2.5-1.5B triviaqa 2026-10-01
qwen3-0.6b-box2plain-ruler-32768 qwen3-0.6b ruler (13 subtasks) 2026-10-05
qwen3-0.6b-ruler-16384 Qwen3-0.6B-Base ruler (13 subtasks) 2026-10-02
qwen3-0.6b-ruler-32768 Qwen3-0.6B-Base ruler (13 subtasks) 2026-10-02
qwen3-0.6b-ruler-4096 Qwen3-0.6B-Base ruler (13 subtasks) 2026-10-01
qwen3-0.6b-ruler-65536 Qwen3-0.6B-Base ruler (13 subtasks) 2026-10-04
qwen3-0.6b-ruler-8192 Qwen3-0.6B-Base ruler (13 subtasks) 2026-10-01
qwen3-0.6b-yarn2-ruler-16384 qwen3-0.6b-yarn2 ruler (13 subtasks) 2026-10-05
qwen3-0.6b-yarn2-ruler-32768 qwen3-0.6b-yarn2 ruler (13 subtasks) 2026-10-05
qwen3-0.6b-yarn2-ruler-65536 qwen3-0.6b-yarn2 ruler (13 subtasks) 2026-10-05
qwen3-1.7b-box2plain-ruler-32768 qwen3-1.7b ruler (13 subtasks) 2026-10-05
qwen3-1.7b-ruler-16384 Qwen3-1.7B-Base ruler (13 subtasks) 2026-10-02
qwen3-1.7b-ruler-32768 Qwen3-1.7B-Base ruler (13 subtasks) 2026-10-02
qwen3-1.7b-ruler-4096 Qwen3-1.7B-Base ruler (13 subtasks) 2026-10-01
qwen3-1.7b-ruler-65536 Qwen3-1.7B-Base ruler (13 subtasks) 2026-10-04
qwen3-1.7b-ruler-8192 Qwen3-1.7B-Base ruler (13 subtasks) 2026-10-01
qwen3-1.7b-yarn2-ruler-16384 qwen3-1.7b-yarn2 ruler (13 subtasks) 2026-10-05
qwen3-1.7b-yarn2-ruler-32768 qwen3-1.7b-yarn2 ruler (13 subtasks) 2026-10-05
qwen3-1.7b-yarn2-ruler-65536 qwen3-1.7b-yarn2 ruler (13 subtasks) 2026-10-07
qwen3.5-0.8b-ruler-131072 Qwen3.5-0.8B-Base ruler (13 subtasks) 2026-10-04
qwen3.5-0.8b-ruler-16384 Qwen3.5-0.8B-Base ruler (13 subtasks) 2026-10-02
qwen3.5-0.8b-ruler-32768 Qwen3.5-0.8B-Base ruler (13 subtasks) 2026-10-02
qwen3.5-0.8b-ruler-4096 Qwen3.5-0.8B-Base ruler (13 subtasks) 2026-10-01
qwen3.5-0.8b-ruler-65536 Qwen3.5-0.8B-Base ruler (13 subtasks) 2026-10-03
qwen3.5-0.8b-ruler-8192 Qwen3.5-0.8B-Base ruler (13 subtasks) 2026-10-01
qwen3.5-2b-ruler-131072 Qwen3.5-2B-Base ruler (13 subtasks) 2026-10-04
qwen3.5-2b-ruler-16384 Qwen3.5-2B-Base ruler (13 subtasks) 2026-10-02
qwen3.5-2b-ruler-32768 Qwen3.5-2B-Base ruler (13 subtasks) 2026-10-02
qwen3.5-2b-ruler-4096 Qwen3.5-2B-Base ruler (13 subtasks) 2026-10-01
qwen3.5-2b-ruler-65536 Qwen3.5-2B-Base ruler (13 subtasks) 2026-10-03
qwen3.5-2b-ruler-8192 Qwen3.5-2B-Base ruler (13 subtasks) 2026-10-01
smollm2-1.7b-bbh SmolLM2-1.7B bbh_fewshot (27 subtasks) 2026-09-30
smollm2-1.7b-bbhws SmolLM2-1.7B bbh_ws (27 subtasks) 2026-10-01
smollm2-1.7b-core SmolLM2-1.7B hellaswag, arc_easy, arc_challenge, piqa, winogrande 2026-09-30
smollm2-1.7b-gsm8k SmolLM2-1.7B gsm8k 2026-09-30
smollm2-1.7b-humaneval SmolLM2-1.7B humaneval 2026-09-30
smollm2-1.7b-mmlu SmolLM2-1.7B mmlu (4 subtasks) 2026-09-30
smollm2-1.7b-triviaqa SmolLM2-1.7B triviaqa 2026-09-30

Reading a results file

Open results_*.json and read the results key: one entry per task, with metric names of the form metric,filter.

  • PIQA, ARC (easy and challenge) and HellaSwag: use acc_norm,none.
  • GSM8K: use exact_match,strict-match.
  • HumanEval: use pass@1.
  • MMLU, TriviaQA, BBH and DROP: use the primary metric of each task as reported in the harness output.
  • RULER: the group average stored in results.json is wrong at context lengths other than 4096. Compute the score as the mean of the 13 individual RULER task values at that length (each stored under the key <length>,none, for example 65536,none), ignoring the group entry. Entries with a negative value or fewer than 13 tasks mean the length was not fully evaluated.

The n-samples key gives the number of examples scored, and config.model_args identifies the model and dtype.

Scrubbing note

Absolute machine paths in the files were replaced with <local> (for example pretrained=<local>/kisoku-1.6b-base-longC-1299), the storage bucket name with <bucket>, and host addresses and login names were removed. The model folder name is kept so each model stays identifiable. Metric values and sample contents were not modified. Benchmark text and model outputs that happen to contain example paths or private-range IP addresses (for example router setup answers in HellaSwag) are left as they are.

Links

heldout-chat/

The held-out conversation test from report section 7B. scripts.jsonl holds the 297 simulated-user conversation scripts. rollout-<model>.jsonl holds each model's replies to every script (Kisoku chat passes 5, 6, 8 and 9, and Llama 3.2 1B Instruct and Llama 3.1 8B Instruct as references, each given a system prompt saying it is Kisoku). judge-<model>.jsonl holds the automatic judge's verdict per turn (ok true or false, with a one-sentence problem when false). The acceptable-turn percentages in the report are computed from these. The on-policy correction data built from these rollouts is training data and is not released.
Downloads last month
71
Total file size:
760 MB

Collection including 0arch-io/kisoku-1.6b-eval