Omega-S: A Functional Resilience Index for LLM Fine-Tuning
Paper • 2608.03887 • Published • 7
Dataset Viewer
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
_nota: string
hallazgo_de_codigo: struct<descripcion: string, impacto_medido: string>
child 0, descripcion: string
child 1, impacto_medido: string
descomposicion_de_grados: struct<_metodo: string, q_proj_rama_cuadrada: struct<capa0: double, capa8: double, capa16: double, c (... 176 chars omitted)
child 0, _metodo: string
child 1, q_proj_rama_cuadrada: struct<capa0: double, capa8: double, capa16: double, capa24: double, capa31: double, media_capas_8_3 (... 10 chars omitted)
child 0, capa0: double
child 1, capa8: double
child 2, capa16: double
child 3, capa24: double
child 4, capa31: double
child 5, media_capas_8_31: double
child 2, v_proj_rama_gram: struct<capa0: double, capa8: double, capa16: double, capa24: double, capa31: double, media: double>
child 0, capa0: double
child 1, capa8: double
child 2, capa16: double
child 3, capa24: double
child 4, capa31: double
child 5, media: double
lectura: string
siguiente_experimento: string
caveat: string
elasticidad_de_los_cuatro_factores: struct<_nota: string, _metodo: string, mediana: struct<C: double, D: double, M_lambda2: double, Coex (... 921 chars omitted)
child 0, _nota: string
child 1, _metodo: string
child 2, mediana: struct<C: double, D: double, M_lambda2: double, Coex: double>
child 0, C: double
child 1, D: double
child 2, M_lambda2: double
child 3, Coex: double
child 3, por_modulo: struct<L0.q_proj: struct<rama:
...
double
child 3, ppl_after_B: double
child 4, strength: double
child 4, l2sp: struct<absolute: double, ratio: double, task_A: double, ppl_after_B: double, strength: double>
child 0, absolute: double
child 1, ratio: double
child 2, task_A: double
child 3, ppl_after_B: double
child 4, strength: double
child 5, omega_minv: struct<absolute: double, ratio: double, task_A: double, ppl_after_B: double, strength: double>
child 0, absolute: double
child 1, ratio: double
child 2, task_A: double
child 3, ppl_after_B: double
child 4, strength: double
child 6, omega_minv_tuned: struct<absolute: double, ratio: double, task_A: double, ppl_after_B: double, strength: double>
child 0, absolute: double
child 1, ratio: double
child 2, task_A: double
child 3, ppl_after_B: double
child 4, strength: double
_findings: list<item: string>
child 0, item: string
_protocol: struct<model: string, lora: string, tasks: string, metric: string, penalty_schedule: string, strengt (... 91 chars omitted)
child 0, model: string
child 1, lora: string
child 2, tasks: string
child 3, metric: string
child 4, penalty_schedule: string
child 5, strength: string
child 6, tuning_seed: int64
child 7, evaluation_seeds: list<item: int64>
child 0, item: int64
child 8, note_on_tuning: string
_primary_metric: string
_what: string
_known_gaps: list<item: string>
child 0, item: string
_ties: string
to
{'_what': Value('string'), '_protocol': {'model': Value('string'), 'lora': Value('string'), 'tasks': Value('string'), 'metric': Value('string'), 'penalty_schedule': Value('string'), 'strength': Value('string'), 'tuning_seed': Value('int64'), 'evaluation_seeds': List(Value('int64')), 'note_on_tuning': Value('string')}, '_primary_metric': Value('string'), '_ties': Value('string'), 'arms': {'none': {'42': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '123': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '456': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '789': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '1011': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '2022': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_
...
': Value('float64')}}, 'plasticity_tests': {'omega_lib_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_raw_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_spectral_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'l2sp_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_minv_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_minv_tuned_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}}, '_findings': List(Value('string')), '_known_gaps': List(Value('string'))}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 489, 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 2818, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 2369, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
_nota: string
hallazgo_de_codigo: struct<descripcion: string, impacto_medido: string>
child 0, descripcion: string
child 1, impacto_medido: string
descomposicion_de_grados: struct<_metodo: string, q_proj_rama_cuadrada: struct<capa0: double, capa8: double, capa16: double, c (... 176 chars omitted)
child 0, _metodo: string
child 1, q_proj_rama_cuadrada: struct<capa0: double, capa8: double, capa16: double, capa24: double, capa31: double, media_capas_8_3 (... 10 chars omitted)
child 0, capa0: double
child 1, capa8: double
child 2, capa16: double
child 3, capa24: double
child 4, capa31: double
child 5, media_capas_8_31: double
child 2, v_proj_rama_gram: struct<capa0: double, capa8: double, capa16: double, capa24: double, capa31: double, media: double>
child 0, capa0: double
child 1, capa8: double
child 2, capa16: double
child 3, capa24: double
child 4, capa31: double
child 5, media: double
lectura: string
siguiente_experimento: string
caveat: string
elasticidad_de_los_cuatro_factores: struct<_nota: string, _metodo: string, mediana: struct<C: double, D: double, M_lambda2: double, Coex (... 921 chars omitted)
child 0, _nota: string
child 1, _metodo: string
child 2, mediana: struct<C: double, D: double, M_lambda2: double, Coex: double>
child 0, C: double
child 1, D: double
child 2, M_lambda2: double
child 3, Coex: double
child 3, por_modulo: struct<L0.q_proj: struct<rama:
...
double
child 3, ppl_after_B: double
child 4, strength: double
child 4, l2sp: struct<absolute: double, ratio: double, task_A: double, ppl_after_B: double, strength: double>
child 0, absolute: double
child 1, ratio: double
child 2, task_A: double
child 3, ppl_after_B: double
child 4, strength: double
child 5, omega_minv: struct<absolute: double, ratio: double, task_A: double, ppl_after_B: double, strength: double>
child 0, absolute: double
child 1, ratio: double
child 2, task_A: double
child 3, ppl_after_B: double
child 4, strength: double
child 6, omega_minv_tuned: struct<absolute: double, ratio: double, task_A: double, ppl_after_B: double, strength: double>
child 0, absolute: double
child 1, ratio: double
child 2, task_A: double
child 3, ppl_after_B: double
child 4, strength: double
_findings: list<item: string>
child 0, item: string
_protocol: struct<model: string, lora: string, tasks: string, metric: string, penalty_schedule: string, strengt (... 91 chars omitted)
child 0, model: string
child 1, lora: string
child 2, tasks: string
child 3, metric: string
child 4, penalty_schedule: string
child 5, strength: string
child 6, tuning_seed: int64
child 7, evaluation_seeds: list<item: int64>
child 0, item: int64
child 8, note_on_tuning: string
_primary_metric: string
_what: string
_known_gaps: list<item: string>
child 0, item: string
_ties: string
to
{'_what': Value('string'), '_protocol': {'model': Value('string'), 'lora': Value('string'), 'tasks': Value('string'), 'metric': Value('string'), 'penalty_schedule': Value('string'), 'strength': Value('string'), 'tuning_seed': Value('int64'), 'evaluation_seeds': List(Value('int64')), 'note_on_tuning': Value('string')}, '_primary_metric': Value('string'), '_ties': Value('string'), 'arms': {'none': {'42': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '123': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '456': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '789': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '1011': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_after_B': Value('float64')}, '2022': {'humaneval_after_A': Value('float64'), 'humaneval_after_B': Value('float64'), 'retention_pct': Value('float64'), 'ppl_after_A': Value('float64'), 'ppl_
...
': Value('float64')}}, 'plasticity_tests': {'omega_lib_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_raw_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_spectral_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'l2sp_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_minv_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}, 'omega_minv_tuned_vs_none': {'wins': Value('int64'), 'losses': Value('int64'), 'ties': Value('int64'), 'sign_test_p': Value('float64'), 'wilcoxon_W': Value('int64'), 'wilcoxon_p': Value('float64'), 'mean_difference': Value('float64')}}, '_findings': List(Value('string')), '_known_gaps': List(Value('string'))}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Omega-S: per-seed retention results
Per-seed measurements behind the retention tables of Omega-S, a data-free regularisation penalty for low-rank fine-tuning. Llama-3-8B with LoRA, code to prose, HumanEval over ten seeds, plus the tuned weight-decay and EWC arms and the negative controls.
Paper: https://arxiv.org/abs/2608.03887 Code: https://github.com/BiomeMakers/OmegaS-LLM
- Downloads last month
- 26