SenseMath: Do LLMs Have Number Sense? Evaluating Shortcut Use, Judgment, and Generation
Paper • 2604.01988 • Published
Dataset Viewer
The dataset viewer is not available for this split.
Cannot extract the features (columns) for the split 'train' of the config 'apply_d16' of the dataset.
Error code: FeaturesError
Exception: ValueError
Message: Value is too big!
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 244, in compute_first_rows_from_streaming_response
iterable_dataset = iterable_dataset._resolve_features()
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 4645, in _resolve_features
features = _infer_features_from_batch(self.with_format(None)._head())
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2866, in _head
return next(iter(self.iter(batch_size=n)))
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3073, in iter
for key, pa_table in ex_iterable.iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2581, in _iter_arrow
yield from self.ex_iterable._iter_arrow()
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 251, in _generate_tables
batch = "\n".join(ujson_dumps(x) for x in ujson_loads(full_data)).encode()
~~~~~~~~~~~^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
return pd.io.json.ujson_loads(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
ValueError: Value is too big!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.
SenseMath
Paper · GitHub · Hugging Face
Version: revised-20261109. MIT license.
| Data | File | Items |
|---|---|---|
| Apply | sensemath_v2_d{2,4,8,16}.json |
4,800 |
| J1 | judge_j1.json |
300 |
| J1 paper subset (default evaluation) | subsets/judge_j1_paper_filtered_261.json |
261 |
| J2 | judge_j2.json |
80 |
Files are in benchmark/ on GitHub and data/ on Hugging Face.
Apply has 8 categories and strong/weak/control variants; each file contains 400 families.
Use pure_math.question, options, and answer. Judge uses prompt and ground_truth.
label_source is human, legacy, rule, or pending; labels are not all human-validated.
Eight J1 labels outside the paper subset are null and excluded from scoring.
The 261-item subset is part of J1, not additional data. solution_length_limited flags truncated J2 solutions.
manifest.json provides file hashes. Read JSON with Python to preserve large integers.
Run
pip install openai
# Set OPENAI_API_KEY in your environment.
python code/run_inference_api.py --step inference --model gpt-4o-mini --scale 4
python code/run_inference_api.py --step analyze --scale 4
python code/run_judge_tasks.py --step validate --task all
python code/run_judge_tasks.py --step inference --task all --api --api-model gpt-4o-mini
python code/run_judge_tasks.py --step analyze
Apply runs CoT, NC (number-sense prompting), and Strict conditions; inference calls the model API.
Judge defaults to the 261-item J1 subset; use --j1-profile full for the 300-item set in both inference and analysis.
Results are saved in results/revised-20261109/. These commands compute answer accuracy; shortcut-use annotation is separate.
- Downloads last month
- 108