Datasets:
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: DatasetGenerationError
Exception: TypeError
Message: Couldn't cast array of type struct<span_id: string, fraction_kept: double> to double
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
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 2312, in cast_table_to_schema
cast_array_to_feature(
~~~~~~~~~~~~~~~~~~~~~^
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
feature,
^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2068, in cast_array_to_feature
_c(array.field(name) if name in array_fields else null_array, subfeature)
~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2118, in cast_array_to_feature
casted_array_values = _c(array.values, feature.feature)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2152, in cast_array_to_feature
return array_cast(
array,
...<2 lines>...
allow_decimal_to_str=allow_decimal_to_str,
)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2016, in array_cast
raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
TypeError: Couldn't cast array of type struct<span_id: string, fraction_kept: double> to double
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1880, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
id string | ratio float64 | label_type string | target_output string | kept_mask list | context string | query string | metadata dict |
|---|---|---|---|---|---|---|---|
2hop__132812_120632 | 0.2 | abstain | I cannot determine the answer to this question from the available context. | [
true,
true,
true,
false,
false,
false,
false,
true,
true,
false,
false,
false,
false,
false,
false,
true,
true,
true,
true,
true,
true,
true,
true,
true,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,... | [Dodge 400]
The Dodge 400 was a mid-size car built by Dodge. It was similar to the Chrysler LeBaron of the same era. The 400 was introduced for the 1982 model year but renamed and merged into the Dodge 600 lineup just two years later. An improved version of Chrysler's K-cars, it was called the Super K platform.
[Presi... | What year was the end of the company that built the White armored car? | {
"compressor": "snapkv",
"model_name": "Qwen/Qwen2.5-7B-Instruct",
"ctx_len": 2190,
"n_kept": 438,
"answer": "1980",
"answer_aliases": [],
"supporting_spans": [
{
"start": 686,
"end": 690,
"span_id": "2hop__132812_120632/p7",
"source": "musique_qd_subanswer"
},
{
... |
2hop__489283_18120 | 0.5 | abstain | I cannot determine the answer to this question from the available context. | [
true,
true,
true,
true,
true,
true,
true,
false,
true,
false,
true,
true,
true,
true,
true,
true,
false,
true,
true,
true,
true,
true,
true,
true,
false,
false,
true,
true,
true,
true,
true,
false,
false,
true,
false,
true,
false,
false,
tr... | [Chicago Cubs]
On September 23, 1908, the Cubs and New York Giants were involved in a tight pennant race. The two clubs were tied in the bottom of the ninth inning at the Polo Grounds, and N.Y. had runners on first and third and two outs when Al Bridwell singled, scoring Moose McCormick from third with the Giants' appa... | When Anson led the Chicago Cubs, how many pennants did they win in the league that employed Peter McLaughlin? | {
"compressor": "kvzip",
"model_name": "Qwen/Qwen2.5-7B-Instruct",
"ctx_len": 4096,
"n_kept": 2048,
"answer": "six",
"answer_aliases": [],
"supporting_spans": [
{
"start": 1528,
"end": 1530,
"span_id": "2hop__489283_18120/p7",
"source": "musique_qd_subanswer"
},
{
... |
2hop__767276_77166 | 0.3 | confident_answer | fifteen | [
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
false,
true,
false,
false,
false,
false,
false,
false,
true,
false,
true,
true,
false,
true,
false,
false,
false,
false,
false,
false,
false,
true,
fals... | [United Nations Security Council]
The Security Council consists of fifteen members. The great powers that were the victors of World War II -- the Soviet Union (now represented by the Russian Federation), the United Kingdom, France, the Republic of China (now represented by the People's Republic of China), and the Unite... | How many members does the legislator of United Nations Security Council Resolution 731 have? | {
"compressor": "expected_attention",
"model_name": "Qwen/Qwen2.5-7B-Instruct",
"ctx_len": 2119,
"n_kept": 636,
"answer": "fifteen",
"answer_aliases": [],
"supporting_spans": [
{
"start": 11,
"end": 12,
"span_id": "2hop__767276_77166/p0",
"source": "musique_qd_subanswer"
},... |
2hop__63241_65123 | 0.5 | abstain | I cannot determine the answer to this question from the available context. | [
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true... | [The Story (song)]
``The Story ''is a song released as a single by American folk rock singer Brandi Carlile, written by Phil Hanseroth, from her 2007 album The Story. It was featured in Grey's Anatomy in 2007 and is on Grey's Anatomy Soundtrack album 3 (released September 11).
[Grey's Anatomy (season 14)]
Grey's Anato... | When did the season of Grey's Anatomy where Derek Shepard died begin filming? | {
"compressor": "snapkv",
"model_name": "Qwen/Qwen2.5-7B-Instruct",
"ctx_len": 2803,
"n_kept": 1402,
"answer": "filming for the eleventh season would begin on July 25, 2014",
"answer_aliases": [],
"supporting_spans": [
{
"start": 552,
"end": 556,
"span_id": "2hop__63241_65123/p5",
... |
2hop__33881_158398 | 0.5 | abstain | I cannot determine the answer to this question from the available context. | [
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
true,
false,
false,
true,
true,
true,
tr... | [SQL:2008]
SQL:2008 is the sixth revision of the ISO and ANSI standard for the SQL database query language. It was formally adopted in July 2008. The standard consists of 9 parts which are described in detail in SQL.
[Alps]
This may be consistent with the theory that in Greek Alpes is a name of non-Indo-European origi... | Who is the most prominent database language is named after? | {
"compressor": "snapkv",
"model_name": "Qwen/Qwen2.5-7B-Instruct",
"ctx_len": 1993,
"n_kept": 996,
"answer": "Michael Widenius's daughter",
"answer_aliases": [
"Michael Widenius"
],
"supporting_spans": [
{
"start": 493,
"end": 494,
"span_id": "2hop__33881_158398/p4",
"so... |
2hop__153313_150998 | 0.05 | abstain | I cannot determine the answer to this question from the available context. | [
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
false,
true,
false,
false,
false,
false,
false,
false,
false,
fa... | [The Angry Beavers]
The Angry Beavers is an American animated sitcom created by Mitch Schauer for Nickelodeon. The series revolves around Daggett and Norbert Beaver, two young beaver brothers who have moved out of their parents home to become bachelors in the forest near the fictional Wayouttatown, Oregon. The show pre... | Who produced the series Acme Animation Factory is part of? | {
"compressor": "kvzip",
"model_name": "Qwen/Qwen2.5-7B-Instruct",
"ctx_len": 2526,
"n_kept": 126,
"answer": "Warner Bros.",
"answer_aliases": [],
"supporting_spans": [
{
"start": 547,
"end": 551,
"span_id": "2hop__153313_150998/p4",
"source": "musique_qd_subanswer"
},
... |
2hop__18025_34539 | 0.1 | abstain | I cannot determine the answer to this question from the available context. | [false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,fal(...TRUNCATED) | "[Arsenal F.C.]\nArsenal's tally of 13 League Championships is the third highest in English football(...TRUNCATED) | "Are attendance remittances that important to the league whose trophy was won by an undefeated Arsen(...TRUNCATED) | {"compressor":"kvzip","model_name":"Qwen/Qwen2.5-7B-Instruct","ctx_len":3882,"n_kept":388,"answer":"(...TRUNCATED) |
2hop__2964_2998 | 0.05 | abstain | I cannot determine the answer to this question from the available context. | [true,true,true,false,false,false,false,false,false,false,false,false,false,false,false,false,false,(...TRUNCATED) | "[Anna Chandy]\nJustice Anna Chandy (1905 - 1996), also known as Anna Chandi, was the first female j(...TRUNCATED) | When did the original judge of American Idol who was a choreographer quit? | {"compressor":"snapkv","model_name":"Qwen/Qwen2.5-7B-Instruct","ctx_len":1989,"n_kept":99,"answer":"(...TRUNCATED) |
2hop__783931_37045 | 0.8 | confident_answer | 4th | [true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true,true(...TRUNCATED) | "[Santa Monica, California]\nEvery fall the Santa Monica Chamber of Commerce hosts The Taste of Sant(...TRUNCATED) | "In what century did the person the Budești Josani church is named after become the influence for r(...TRUNCATED) | {"compressor":"snapkv","model_name":"Qwen/Qwen2.5-7B-Instruct","ctx_len":2498,"n_kept":1998,"answer"(...TRUNCATED) |
2hop__90625_576533 | 0.05 | abstain | I cannot determine the answer to this question from the available context. | [true,true,true,true,false,false,true,true,false,true,true,false,false,false,false,false,false,false(...TRUNCATED) | "[Isabel of Coimbra]\nInfanta Isabel of Coimbra (Isabella of Portugal) (1 March 1432 – 2 December (...TRUNCATED) | Who wed the actress portraying the wife in the king's speech? | {"compressor":"expected_attention","model_name":"Qwen/Qwen2.5-7B-Instruct","ctx_len":1902,"n_kept":9(...TRUNCATED) |
Compression-Aware Abstention — labeled datasets
Compression-labeled training and evaluation data for "Compression-Aware Abstention" (GroundLM @ EMNLP 2026). Derived from MuSiQue 2-hop QA (CC BY 4.0, Trivedi et al., TACL 2022).
Each record pairs a MuSiQue example with a KV-cache compressor's per-token survival mask at one of six retention ratios, plus the label that mask implies.
Fields
| Field | Meaning |
|---|---|
id |
MuSiQue example id (2hop__*) |
ratio |
Target retention ratio: 0.05, 0.10, 0.20, 0.30, 0.50, 0.80 |
kept_mask |
Per-token boolean survival mask over the tokenized context |
label_type |
confident_answer or abstain |
target_output |
Gold answer, or the fixed refusal string |
supporting_spans |
Tight answer-bearing token ranges used for labeling |
metadata |
Compressor name, base model, context length |
Labels follow the rule in the paper: confident_answer when every supporting
span retains at least 0.8 of its tokens, abstain when any span falls below
0.3, dropped otherwise. Masks are generated by the same base model that
consumes them, since eviction patterns are model-specific.
Files
| File | Rows | Notes |
|---|---|---|
train_v1.jsonl / val_v1.jsonl |
2669 / 463 | KVzip masks, Qwen2.5-7B-Instruct. The headline split. |
train_mixture_v1.jsonl |
8772 | KVzip + Expected Attention + SnapKV-context |
train_mixture_v1_holdout.jsonl |
984 | Checkpoint selection only, id-disjoint from train |
val_mixture_v1.jsonl |
1511 | Across all three compressors |
val_v1_expattn.jsonl, val_v1_snapkv.jsonl, val_v1_snapkv_native.jsonl |
414 / 431 / 419 | Same examples, held-out compressors |
train_v1_llama.jsonl / val_v1_llama.jsonl |
3640 / 465 | Llama-3.1-8B-Instruct masks |
train_v1_random.jsonl / val_v1_random.jsonl |
1236 / 468 | Uniform-random masks at matched retention |
test_held_out.jsonl / val_v1_for_selection.jsonl |
100 / 363 | Id-disjoint split of val_v1 |
Training splits apply a 30/70 minority-class floor; validation splits keep the natural class distribution. Train and validation are disjoint by example id.
Caveats
- The class distribution is U-shaped in the retention ratio: almost all
abstainat low retention, almost allconfident_answerat high. Only the r=0.50 slice is close to balanced. Aggregate metrics over the full set reward a policy that thresholds on the ratio alone without reading the context. - Tight-span survival is a surrogate for contextual sufficiency. On a stratified 400-example audit it reaches Cohen's kappa 0.61 against an LLM-judge verdict; disagreements concentrate on multi-hop questions where the answer string survives but the bridging evidence does not.
Code and reproduction: https://github.com/mali-kh/compression-aware-abstention
- Downloads last month
- 117