Hugging Face's logo

Datasets:
tasksource
/
tasksource-jev-typed-decisions

Tasks:
Zero-Shot Classification
Text Classification
Question Answering
Modalities:
Text
Formats:
parquet
Languages:
English
multilingual
Size:
1M - 10M
Tags:
tasksource
jev
system-one
runtime-defined-decisions
decision-models
multiple-choice
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
3

Add quality-filtered validation and test splits to filtered-full

#3
by bowang0911 - opened
base: refs/heads/main
←
from: refs/pr/3
Discussion Files changed
+1541
-6
Files changed (7) hide show
  1. README.md +13 -3
  2. filtered-full/README.md +4 -3
  3. filtered-full/test-00000-of-00001.parquet +3 -0
  4. filtered-full/validation-00000-of-00001.parquet +3 -0
  5. filtered-full/validation-test-decisions.jsonl.gz +3 -0
  6. filtered-full/validation-test-filtering-report.json +1421 -0
  7. filtered-full/validation-test.md +94 -0
README.md CHANGED
@@ -39,6 +39,10 @@ configs:
39
  data_files:
40
  - split: train
41
  path: filtered-full/train-*.parquet
 
 
 
 
42
  dataset_info:
43
  - config_name: default
44
  features:
@@ -156,8 +160,14 @@ dataset_info:
156
  - name: train
157
  num_bytes: 1633306939
158
  num_examples: 1305478
159
- download_size: 335270018
160
- dataset_size: 1633306939
 
 
 
 
 
 
161
  ---
162
 
163
  # tasksource-jev-typed-decisions
@@ -221,7 +231,7 @@ print(row["state"], row["question"], row["options"], row["target"])
221
  - `default`: a steered mix of about 1M train rows. Sources are first gated on label correctness, then weighted by how interesting they are and how close they sit to the zone of proximal development (judged by decision models). Two-option tasks get fewer rows, and procedural generators get 12%. No row is repeated. Validation and test are the full eval splits, restricted to the mixed sources. Buckets and shares are in [`jev_mixes.py`](https://github.com/sileod/tasksource/blob/main/src/tasksource/metadata/jev_mixes.py) and per-source scores in `jev_source_scores.csv`.
222
  - `full`: every row that passed the build, with per-source caps (about 2.5M train rows).
223
 
224
- - `filtered-full`: 1,305,478 retained `train` decisions from the frozen `full` revision `d2ab1d12be4463fb7ac1f877ed2193b3bb59642a`. Applies commercial-license metadata gating, benchmark-overlap/structural filters, and conservative GLM-5.3-Flash review. Uncertain cases are retained; original row values and soft targets are preserved. No validation or test split is added. This is a filtered historical snapshot and excludes additions made after that revision. See [filtering methods and counts](filtered-full/README.md).
225
 
226
  ## Format
227
 
 
39
  data_files:
40
  - split: train
41
  path: filtered-full/train-*.parquet
42
+ - split: validation
43
+ path: filtered-full/validation-*.parquet
44
+ - split: test
45
+ path: filtered-full/test-*.parquet
46
  dataset_info:
47
  - config_name: default
48
  features:
 
160
  - name: train
161
  num_bytes: 1633306939
162
  num_examples: 1305478
163
+ - name: validation
164
+ num_bytes: 17814120
165
+ num_examples: 14731
166
+ - name: test
167
+ num_bytes: 17734798
168
+ num_examples: 14745
169
+ download_size: 342344028
170
+ dataset_size: 1668855857
171
  ---
172
 
173
  # tasksource-jev-typed-decisions
 
231
  - `default`: a steered mix of about 1M train rows. Sources are first gated on label correctness, then weighted by how interesting they are and how close they sit to the zone of proximal development (judged by decision models). Two-option tasks get fewer rows, and procedural generators get 12%. No row is repeated. Validation and test are the full eval splits, restricted to the mixed sources. Buckets and shares are in [`jev_mixes.py`](https://github.com/sileod/tasksource/blob/main/src/tasksource/metadata/jev_mixes.py) and per-source scores in `jev_source_scores.csv`.
232
  - `full`: every row that passed the build, with per-source caps (about 2.5M train rows).
233
 
234
+ - `filtered-full`: 1,305,478 training decisions, 14,731 validation decisions, and 14,745 test decisions retained from the frozen `full` revision `d2ab1d12be4463fb7ac1f877ed2193b3bb59642a`. Training combines preprocessing exclusions with conservative GLM-5.3-Flash review; validation/test use quality-only review to preserve evaluation coverage. Original row values and soft targets are preserved. Later repository additions are outside this snapshot. See [training filtering methods](filtered-full/README.md) and [validation/test filtering methods](filtered-full/validation-test.md).
235
 
236
  ## Format
237
 
filtered-full/README.md CHANGED
@@ -14,10 +14,11 @@ train = load_dataset(
14
  )
15
  ```
16
 
17
- Only `train` is supplied. Existing `default` and `full` data, configurations, and
18
- validation/test splits are unchanged.
 
19
 
20
- ## Selection counts
21
 
22
  | Stage | Decisions |
23
  |---|---:|
 
14
  )
15
  ```
16
 
17
+ This document describes the training cleanup. Quality-filtered `validation` and `test` splits
18
+ are also available; see [their separate methods and counts](validation-test.md). Existing
19
+ `default` and `full` data, configurations, and validation/test splits are unchanged.
20
 
21
+ ## Training selection counts
22
 
23
  | Stage | Decisions |
24
  |---|---:|
filtered-full/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:36abb9d322092a217a77fec0a56c0dc7bb6b2c748e8eafc6c0742fc9e98f32ac
3
+ size 3516042
filtered-full/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:662a32ff08b79f7077d3ef3ccd2e986004466db8bd4c966b44de3836fe023f8b
3
+ size 3557968
filtered-full/validation-test-decisions.jsonl.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0a30b16005e53470aa86a5eed7bb43377ccb2eb408485276d1ce05971fad9779
3
+ size 2252636
filtered-full/validation-test-filtering-report.json ADDED
@@ -0,0 +1,1421 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": "tasksource/tasksource-jev-typed-decisions",
3
+ "config": "filtered-full",
4
+ "source_config": "full",
5
+ "source_revision": "d2ab1d12be4463fb7ac1f877ed2193b3bb59642a",
6
+ "policy": "quality_only",
7
+ "model": "glm_5_3_flash",
8
+ "workers": 768,
9
+ "splits": {
10
+ "validation": {
11
+ "source_rows": 15000,
12
+ "retained_rows": 14731,
13
+ "dispositions": {
14
+ "keep": 13640,
15
+ "uncertain": 1091,
16
+ "drop": 269
17
+ },
18
+ "soft_target_rows_retained": 1418,
19
+ "sources_retained": {
20
+ "AES2-essay-scoring": 16,
21
+ "AdjectiveScaleProbe-nli": 27,
22
+ "AmbigNQ-clarifying-question": 23,
23
+ "BeaverTails": 23,
24
+ "CONDAQA": 25,
25
+ "CREAK": 24,
26
+ "ConTRoL-nli": 127,
27
+ "Dilemmas_Disagreement": 11,
28
+ "Dynasent_Disagreement": 11,
29
+ "FLD.v2/default": 16,
30
+ "FLD.v2/star": 11,
31
+ "FLUTE": 22,
32
+ "FOL-nli": 81,
33
+ "HatemojiBuild": 22,
34
+ "HelpSteer/coherence": 3,
35
+ "HelpSteer/complexity": 2,
36
+ "HelpSteer/correctness": 2,
37
+ "HelpSteer/helpfulness": 5,
38
+ "HelpSteer/verbosity": 2,
39
+ "HelpSteer2/coherence": 4,
40
+ "HelpSteer2/complexity": 10,
41
+ "HelpSteer2/correctness": 7,
42
+ "HelpSteer2/helpfulness": 4,
43
+ "HelpSteer2/verbosity": 4,
44
+ "HelpSteer3/edit_quality": 24,
45
+ "HelpSteer3/feedback": 9,
46
+ "HelpSteer3/preference": 25,
47
+ "HelpSteer3/preference_strength": 8,
48
+ "HelpSteer3/principle": 11,
49
+ "I2D2": 21,
50
+ "IntentGrasp/all": 73,
51
+ "LogicNLI": 27,
52
+ "MOH": 23,
53
+ "MSciNLI": 24,
54
+ "MedQA-USMLE-4-options-hf": 61,
55
+ "PARADISE": 58,
56
+ "PARARULE-Plus": 22,
57
+ "PKU-SafeRLHF/helpfulness": 24,
58
+ "PKU-SafeRLHF/safety": 24,
59
+ "Pol_NLI": 25,
60
+ "Politeness_Disagreement": 12,
61
+ "Prompt-injection-dataset/full": 21,
62
+ "PromptShield": 21,
63
+ "ReSQ": 43,
64
+ "SBIC_Disagreement": 11,
65
+ "SChem_Disagreement": 11,
66
+ "SDOH-NLI": 13,
67
+ "SHP": 25,
68
+ "Sarcasm_News_Headline": 22,
69
+ "ScienceQA_text_only": 47,
70
+ "ShellRisk-Bench": 22,
71
+ "SpaRTUN": 48,
72
+ "SpaceNLI": 29,
73
+ "Touche23-ValueEval": 12,
74
+ "TroFi": 22,
75
+ "TuringBench": 50,
76
+ "UNLI": 475,
77
+ "UltraFeedback-paired": 51,
78
+ "VUAC": 12,
79
+ "WANLI": 82,
80
+ "acceptability-prediction/binary_votes": 19,
81
+ "acceptability-prediction/rating_votes": 55,
82
+ "add_one_rte": 22,
83
+ "ade_corpus_v2/Ade_corpus_v2_classification": 19,
84
+ "ag_news": 29,
85
+ "agent_action_safety": 12,
86
+ "ai2_arc/ARC-Challenge/challenge": 16,
87
+ "ai2_arc/ARC-Easy/challenge": 16,
88
+ "amazon_counterfactual/en": 21,
89
+ "amazon_polarity/amazon_polarity": 22,
90
+ "ambient": 23,
91
+ "anli/a1": 27,
92
+ "anli/a2": 29,
93
+ "anli/a3": 28,
94
+ "app_reviews": 26,
95
+ "apt": 22,
96
+ "arct": 46,
97
+ "args_me": 20,
98
+ "argument-feedback": 17,
99
+ "art": 48,
100
+ "auditor_review": 28,
101
+ "autotnli": 23,
102
+ "avicenna": 22,
103
+ "babi_nli/basic-coreference": 2,
104
+ "babi_nli/basic-deduction": 2,
105
+ "babi_nli/basic-induction": 2,
106
+ "babi_nli/compound-coreference": 5,
107
+ "babi_nli/conjunction": 3,
108
+ "babi_nli/counting": 2,
109
+ "babi_nli/indefinite-knowledge": 2,
110
+ "babi_nli/lists-sets": 1,
111
+ "babi_nli/path-finding": 1,
112
+ "babi_nli/positional-reasoning": 2,
113
+ "babi_nli/simple-negation": 2,
114
+ "babi_nli/single-supporting-fact": 1,
115
+ "babi_nli/size-reasoning": 1,
116
+ "babi_nli/three-arg-relations": 1,
117
+ "babi_nli/three-supporting-facts": 1,
118
+ "babi_nli/time-reasoning": 2,
119
+ "babi_nli/two-arg-relations": 1,
120
+ "babi_nli/two-supporting-facts": 1,
121
+ "babi_nli/yes-no-questions": 1,
122
+ "balanced-copa": 25,
123
+ "banking77": 39,
124
+ "biosift-nli": 20,
125
+ "blog_authorship_corpus/age": 7,
126
+ "blog_authorship_corpus/gender": 13,
127
+ "blog_authorship_corpus/job": 8,
128
+ "boolq-natural-perturbations": 19,
129
+ "brainteasers/SP": 19,
130
+ "brainteasers/WP": 21,
131
+ "breaking_nli": 30,
132
+ "chaos-mnli-ambiguity/votes": 74,
133
+ "chatbot_arena_conversations": 50,
134
+ "chemprot/chemprot_full_source": 31,
135
+ "cicero": 57,
136
+ "circa": 29,
137
+ "citation_intent": 33,
138
+ "civil_comments/identity_attack_share": 9,
139
+ "civil_comments/insult_share": 9,
140
+ "civil_comments/obscene_share": 9,
141
+ "civil_comments/severe_toxicity_share": 8,
142
+ "civil_comments/sexual_explicit_share": 8,
143
+ "civil_comments/threat_share": 8,
144
+ "civil_comments/toxicity_share": 8,
145
+ "cladder": 21,
146
+ "clcd-english": 23,
147
+ "clinc_oos/plus": 41,
148
+ "cloth": 59,
149
+ "clutrr": 29,
150
+ "cnli": 27,
151
+ "codah/codah": 30,
152
+ "code_x_glue_cc_defect_detection": 24,
153
+ "com2sense": 21,
154
+ "commonsense_qa": 62,
155
+ "commonsense_qa_2.0": 22,
156
+ "conceptrules_v2": 34,
157
+ "conj_nli": 28,
158
+ "conll2003/ner_tags": 225,
159
+ "contract-nli/contractnli_a/seg": 14,
160
+ "contract-nli/contractnli_b/full": 14,
161
+ "corr2cause": 44,
162
+ "cos_e/v1.0": 63,
163
+ "cosmos_qa": 57,
164
+ "counterfactually-augmented-imdb": 22,
165
+ "counterfactually-augmented-snli": 28,
166
+ "crowdflower/airline-sentiment": 4,
167
+ "crowdflower/corporate-messaging": 5,
168
+ "crowdflower/political-media-audience": 2,
169
+ "crowdflower/political-media-bias": 3,
170
+ "crowdflower/political-media-message": 3,
171
+ "crowdflower/sentiment_nuclear_power": 4,
172
+ "crowdflower/text_emotion": 3,
173
+ "crowdflower/tweet_global_warming": 3,
174
+ "cycic_classification": 21,
175
+ "cycic_multiplechoice": 62,
176
+ "dadc-limit-nli": 21,
177
+ "dataset_train_nli": 112,
178
+ "dbpedia_14/dbpedia_14": 16,
179
+ "defeasible-nli/atomic": 8,
180
+ "defeasible-nli/snli": 8,
181
+ "defeasible-nli/social": 6,
182
+ "definite_pronoun_resolution": 48,
183
+ "dgen": 33,
184
+ "discosense": 61,
185
+ "discovery/discovery": 39,
186
+ "disrpt/eng.dep.scidtb.rels": 40,
187
+ "dnc": 21,
188
+ "dnd_style_intents": 33,
189
+ "doc-nli": 111,
190
+ "docred": 59,
191
+ "dream": 30,
192
+ "dynahate": 23,
193
+ "dynasent/r1_votes": 111,
194
+ "dynasent/r2_votes": 111,
195
+ "e-CARE": 49,
196
+ "ekar_english": 61,
197
+ "emo/emo2019": 14,
198
+ "emotion": 33,
199
+ "english-grading/cohesion": 2,
200
+ "english-grading/conventions": 5,
201
+ "english-grading/grammar": 2,
202
+ "english-grading/phraseology": 2,
203
+ "english-grading/syntax": 2,
204
+ "english-grading/vocabulary": 1,
205
+ "equate": 28,
206
+ "esci": 15,
207
+ "ethics/commonsense": 6,
208
+ "ethics/deontology": 6,
209
+ "ethics/justice": 5,
210
+ "ethics/virtue": 6,
211
+ "ethos/binary": 11,
212
+ "ethos/multilabel": 12,
213
+ "feasibilityQA": 22,
214
+ "fever-evidence-related": 13,
215
+ "few_rel/default": 22,
216
+ "fig-qa": 50,
217
+ "financial_phrasebank/sentences_allagree": 15,
218
+ "folio": 28,
219
+ "fool-me-twice": 22,
220
+ "fracas": 17,
221
+ "gen_debiased_nli/mnli_par_z": 2,
222
+ "gen_debiased_nli/mnli_seq_z": 1,
223
+ "github-issue-similarity": 23,
224
+ "glue/cola": 4,
225
+ "glue/mnli": 3,
226
+ "glue/mrpc": 4,
227
+ "glue/qnli": 3,
228
+ "glue/qqp": 2,
229
+ "glue/rte": 2,
230
+ "glue/sst2": 2,
231
+ "glue/stsb": 90,
232
+ "glue/wnli": 2,
233
+ "go_emotions/simplified": 20,
234
+ "goal-step-wikihow/goal": 21,
235
+ "goal-step-wikihow/order": 21,
236
+ "goal-step-wikihow/step": 18,
237
+ "google_wellformed_query": 29,
238
+ "hans": 23,
239
+ "hate_speech18": 22,
240
+ "hate_speech_offensive": 28,
241
+ "head_qa/en": 62,
242
+ "headline_cause/en_simple": 28,
243
+ "health_fact": 29,
244
+ "hellaswag": 61,
245
+ "help-nli": 27,
246
+ "hh-rlhf/harmless-base": 12,
247
+ "hh-rlhf/helpful-base": 12,
248
+ "hh-rlhf/helpful-online": 11,
249
+ "hh-rlhf/helpful-rejection-sampled": 14,
250
+ "hlgd": 14,
251
+ "hope_edi/english": 28,
252
+ "hover": 22,
253
+ "humicroedit/subtask-1": 5,
254
+ "humicroedit/subtask-2": 10,
255
+ "hyperpartisan_news": 22,
256
+ "idioms-nli": 23,
257
+ "imdb": 23,
258
+ "implicatures": 50,
259
+ "implicit-hate-stg1": 28,
260
+ "insincere-questions": 25,
261
+ "it-support-tickets": 31,
262
+ "jigsaw_toxicity": 22,
263
+ "joci": 30,
264
+ "language-identification": 17,
265
+ "lewidi/hs_brexit": 20,
266
+ "lewidi/md_agreement": 20,
267
+ "lewidi/mp": 19,
268
+ "lex_glue/case_hold": 63,
269
+ "lex_glue/ledgar": 12,
270
+ "lex_glue/scotus": 11,
271
+ "lex_glue/unfair_tos": 11,
272
+ "lexical_relation_classification/BLESS": 6,
273
+ "lexical_relation_classification/CogALexV": 5,
274
+ "lexical_relation_classification/EVALution": 6,
275
+ "lexical_relation_classification/K&H+N": 8,
276
+ "lexical_relation_classification/ROOT09": 4,
277
+ "liar": 17,
278
+ "lifecycle-entailment": 22,
279
+ "lingnli": 29,
280
+ "linguisticprobing/bigram_shift": 1,
281
+ "linguisticprobing/coordination_inversion": 1,
282
+ "linguisticprobing/obj_number": 1,
283
+ "logical-entailment": 21,
284
+ "logical-fallacy": 29,
285
+ "logiqa": 58,
286
+ "logiqa-2.0-nli": 31,
287
+ "lonli": 30,
288
+ "lsat-ar": 58,
289
+ "lsat-rc": 32,
290
+ "lsat_qa/all": 12,
291
+ "math_qa": 52,
292
+ "mc_taco": 23,
293
+ "mctest-nli": 111,
294
+ "medical_questions_pairs": 22,
295
+ "medmcqa": 57,
296
+ "mindgames": 23,
297
+ "missing-item-prediction/contrastive": 11,
298
+ "monli": 22,
299
+ "monotonicity-entailment": 22,
300
+ "moral_stories/full": 48,
301
+ "mpe": 24,
302
+ "multilingual/AfriSenti-twitter-sentiment/amh": 3,
303
+ "multilingual/AfriSenti-twitter-sentiment/arq": 3,
304
+ "multilingual/AfriSenti-twitter-sentiment/ary": 3,
305
+ "multilingual/AfriSenti-twitter-sentiment/hau": 3,
306
+ "multilingual/AfriSenti-twitter-sentiment/ibo": 2,
307
+ "multilingual/AfriSenti-twitter-sentiment/kin": 4,
308
+ "multilingual/AfriSenti-twitter-sentiment/pcm": 2,
309
+ "multilingual/AfriSenti-twitter-sentiment/por": 2,
310
+ "multilingual/AfriSenti-twitter-sentiment/swa": 2,
311
+ "multilingual/AfriSenti-twitter-sentiment/tso": 2,
312
+ "multilingual/AfriSenti-twitter-sentiment/twi": 2,
313
+ "multilingual/AfriSenti-twitter-sentiment/yor": 3,
314
+ "multilingual/MLMA_hate_speech": 17,
315
+ "multilingual/NusaX-senti/ace": 3,
316
+ "multilingual/NusaX-senti/ban": 3,
317
+ "multilingual/NusaX-senti/bbc": 2,
318
+ "multilingual/NusaX-senti/bjn": 3,
319
+ "multilingual/NusaX-senti/bug": 2,
320
+ "multilingual/NusaX-senti/eng": 2,
321
+ "multilingual/NusaX-senti/ind": 2,
322
+ "multilingual/NusaX-senti/jav": 2,
323
+ "multilingual/NusaX-senti/mad": 2,
324
+ "multilingual/NusaX-senti/min": 2,
325
+ "multilingual/NusaX-senti/nij": 3,
326
+ "multilingual/NusaX-senti/sun": 2,
327
+ "multilingual/amazon_reviews_multi/all_languages": 27,
328
+ "multilingual/americas_nli/all_languages": 30,
329
+ "multilingual/clue/afqmc": 11,
330
+ "multilingual/clue/ocnli": 9,
331
+ "multilingual/clue/tnews": 7,
332
+ "multilingual/disrpt/deu.rst.pcc.rels": 4,
333
+ "multilingual/disrpt/eus.rst.ert.rels": 3,
334
+ "multilingual/disrpt/fas.rst.prstc.rels": 4,
335
+ "multilingual/disrpt/fra.sdrt.annodis.rels": 4,
336
+ "multilingual/disrpt/nld.rst.nldt.rels": 4,
337
+ "multilingual/disrpt/por.rst.cstn.rels": 3,
338
+ "multilingual/disrpt/rus.rst.rrt.rels": 3,
339
+ "multilingual/disrpt/spa.rst.rststb.rels": 4,
340
+ "multilingual/disrpt/tha.pdtb.tdtb.rels": 6,
341
+ "multilingual/disrpt/zho.rst.gcdt.rels": 4,
342
+ "multilingual/exams/multilingual": 59,
343
+ "multilingual/indic_glue/actsa-sc.te/sentiment": 5,
344
+ "multilingual/indic_glue/bbca.hi/news": 5,
345
+ "multilingual/indic_glue/iitp-mr.hi/sentiment": 5,
346
+ "multilingual/indic_glue/iitp-pr.hi/sentiment": 4,
347
+ "multilingual/indic_glue/inltkh": 4,
348
+ "multilingual/indic_glue/inltkh.te/sentiment": 4,
349
+ "multilingual/indic_glue/md.hi/discourse_mode": 4,
350
+ "multilingual/indic_glue/sna.bn/news": 5,
351
+ "multilingual/indic_glue/wstp": 62,
352
+ "multilingual/klue/nli": 12,
353
+ "multilingual/klue/sts": 11,
354
+ "multilingual/klue/ynat": 9,
355
+ "multilingual/language-identification": 17,
356
+ "multilingual/masakhanews/amh": 4,
357
+ "multilingual/masakhanews/eng": 3,
358
+ "multilingual/masakhanews/fra": 1,
359
+ "multilingual/masakhanews/hau": 2,
360
+ "multilingual/masakhanews/ibo": 1,
361
+ "multilingual/masakhanews/lin": 1,
362
+ "multilingual/masakhanews/lug": 1,
363
+ "multilingual/masakhanews/orm": 4,
364
+ "multilingual/masakhanews/pcm": 1,
365
+ "multilingual/masakhanews/run": 1,
366
+ "multilingual/masakhanews/sna": 1,
367
+ "multilingual/masakhanews/som": 1,
368
+ "multilingual/masakhanews/swa": 1,
369
+ "multilingual/masakhanews/tir": 1,
370
+ "multilingual/masakhanews/xho": 4,
371
+ "multilingual/masakhanews/yor": 1,
372
+ "multilingual/massive": 38,
373
+ "multilingual/miam": 35,
374
+ "multilingual/mms": 29,
375
+ "multilingual/mtop": 39,
376
+ "multilingual/multilingual-NLI-26lang-2mil7": 30,
377
+ "multilingual/multilingual-sentiments/all": 28,
378
+ "multilingual/oasst1_pairwise_rlhf_reward": 50,
379
+ "multilingual/offenseval_2020/ar": 6,
380
+ "multilingual/offenseval_2020/da": 5,
381
+ "multilingual/offenseval_2020/gr": 5,
382
+ "multilingual/offenseval_2020/tr": 6,
383
+ "multilingual/offenseval_dravidian/kannada": 13,
384
+ "multilingual/offenseval_dravidian/malayalam": 11,
385
+ "multilingual/offenseval_dravidian/tamil": 9,
386
+ "multilingual/paws-x/de": 3,
387
+ "multilingual/paws-x/en": 3,
388
+ "multilingual/paws-x/es": 6,
389
+ "multilingual/paws-x/fr": 4,
390
+ "multilingual/paws-x/ja": 2,
391
+ "multilingual/paws-x/ko": 2,
392
+ "multilingual/paws-x/zh": 2,
393
+ "multilingual/tweet_sentiment_multilingual": 28,
394
+ "multilingual/tydi-as2-balanced": 22,
395
+ "multilingual/universal-joy": 29,
396
+ "multilingual/wili_2018": 38,
397
+ "multilingual/x-fact": 16,
398
+ "multilingual/x-stance": 22,
399
+ "multilingual/xcopa/et": 4,
400
+ "multilingual/xcopa/ht": 3,
401
+ "multilingual/xcopa/id": 2,
402
+ "multilingual/xcopa/it": 3,
403
+ "multilingual/xcopa/qu": 2,
404
+ "multilingual/xcopa/sw": 2,
405
+ "multilingual/xcopa/ta": 2,
406
+ "multilingual/xcopa/th": 2,
407
+ "multilingual/xcopa/tr": 2,
408
+ "multilingual/xcopa/translation-et": 2,
409
+ "multilingual/xcopa/translation-ht": 2,
410
+ "multilingual/xcopa/translation-id": 3,
411
+ "multilingual/xcopa/translation-it": 2,
412
+ "multilingual/xcopa/translation-sw": 2,
413
+ "multilingual/xcopa/translation-ta": 2,
414
+ "multilingual/xcopa/translation-th": 2,
415
+ "multilingual/xcopa/translation-tr": 2,
416
+ "multilingual/xcopa/translation-vi": 2,
417
+ "multilingual/xcopa/translation-zh": 2,
418
+ "multilingual/xcopa/vi": 3,
419
+ "multilingual/xcopa/zh": 2,
420
+ "multilingual/xcsr/X-CODAH-ar": 2,
421
+ "multilingual/xcsr/X-CODAH-de": 2,
422
+ "multilingual/xcsr/X-CODAH-en": 2,
423
+ "multilingual/xcsr/X-CODAH-es": 2,
424
+ "multilingual/xcsr/X-CODAH-fr": 2,
425
+ "multilingual/xcsr/X-CODAH-hi": 1,
426
+ "multilingual/xcsr/X-CODAH-it": 3,
427
+ "multilingual/xcsr/X-CODAH-jap": 3,
428
+ "multilingual/xcsr/X-CODAH-nl": 2,
429
+ "multilingual/xcsr/X-CODAH-pl": 2,
430
+ "multilingual/xcsr/X-CODAH-pt": 2,
431
+ "multilingual/xcsr/X-CODAH-ru": 2,
432
+ "multilingual/xcsr/X-CODAH-sw": 2,
433
+ "multilingual/xcsr/X-CODAH-ur": 2,
434
+ "multilingual/xcsr/X-CODAH-vi": 2,
435
+ "multilingual/xcsr/X-CODAH-zh": 4,
436
+ "multilingual/xcsr/X-CSQA-ar": 2,
437
+ "multilingual/xcsr/X-CSQA-de": 2,
438
+ "multilingual/xcsr/X-CSQA-en": 2,
439
+ "multilingual/xcsr/X-CSQA-es": 3,
440
+ "multilingual/xcsr/X-CSQA-fr": 2,
441
+ "multilingual/xcsr/X-CSQA-hi": 3,
442
+ "multilingual/xcsr/X-CSQA-it": 1,
443
+ "multilingual/xcsr/X-CSQA-jap": 1,
444
+ "multilingual/xcsr/X-CSQA-nl": 1,
445
+ "multilingual/xcsr/X-CSQA-pl": 1,
446
+ "multilingual/xcsr/X-CSQA-pt": 2,
447
+ "multilingual/xcsr/X-CSQA-ru": 1,
448
+ "multilingual/xcsr/X-CSQA-sw": 1,
449
+ "multilingual/xcsr/X-CSQA-ur": 1,
450
+ "multilingual/xcsr/X-CSQA-vi": 1,
451
+ "multilingual/xcsr/X-CSQA-zh": 1,
452
+ "multilingual/xglue/nc": 6,
453
+ "multilingual/xglue/qadsm": 8,
454
+ "multilingual/xglue/qam": 6,
455
+ "multilingual/xglue/wpr": 3,
456
+ "multilingual/xlwic/xlwic_de_de": 4,
457
+ "multilingual/xlwic/xlwic_en_ko": 8,
458
+ "multilingual/xlwic/xlwic_fr_fr": 6,
459
+ "multilingual/xlwic/xlwic_it_it": 4,
460
+ "multilingual/xnli": 84,
461
+ "multilingual/xstory_cloze/ar": 3,
462
+ "multilingual/xstory_cloze/en": 3,
463
+ "multilingual/xstory_cloze/es": 2,
464
+ "multilingual/xstory_cloze/eu": 3,
465
+ "multilingual/xstory_cloze/hi": 2,
466
+ "multilingual/xstory_cloze/id": 2,
467
+ "multilingual/xstory_cloze/my": 3,
468
+ "multilingual/xstory_cloze/ru": 2,
469
+ "multilingual/xstory_cloze/sw": 2,
470
+ "multilingual/xstory_cloze/te": 2,
471
+ "multilingual/xstory_cloze/zh": 2,
472
+ "mutual": 57,
473
+ "nan-nli": 14,
474
+ "natural-language-satisfiability": 23,
475
+ "naturallogic": 27,
476
+ "nlgraph": 21,
477
+ "nli-veridicality-transitivity": 22,
478
+ "nli4ct_semeval2024": 20,
479
+ "nli_fever": 24,
480
+ "nope": 27,
481
+ "numer_sense": 33,
482
+ "oasst2/creativity": 14,
483
+ "oasst2/helpfulness": 14,
484
+ "oasst2/humor": 12,
485
+ "oasst2/quality": 12,
486
+ "oasst2/toxicity": 12,
487
+ "oasst2/violence": 12,
488
+ "oasst2_pairwise_rlhf_reward": 49,
489
+ "onestop_qa": 31,
490
+ "open_question_type": 34,
491
+ "openbookqa": 31,
492
+ "parade": 22,
493
+ "patent-phrase-similarity": 33,
494
+ "path-naturalness-prediction": 50,
495
+ "paws/labeled_final": 11,
496
+ "paws/labeled_swap": 9,
497
+ "phrase_similarity": 21,
498
+ "piqa": 51,
499
+ "poem_sentiment": 15,
500
+ "pragmeval/emergent": 2,
501
+ "pragmeval/emobank-arousal": 2,
502
+ "pragmeval/emobank-dominance": 2,
503
+ "pragmeval/emobank-valence": 2,
504
+ "pragmeval/gum": 2,
505
+ "pragmeval/mrda": 2,
506
+ "pragmeval/pdtb": 2,
507
+ "pragmeval/persuasiveness-claimtype": 2,
508
+ "pragmeval/persuasiveness-eloquence": 2,
509
+ "pragmeval/persuasiveness-premisetype": 2,
510
+ "pragmeval/persuasiveness-relevance": 2,
511
+ "pragmeval/persuasiveness-specificity": 1,
512
+ "pragmeval/persuasiveness-strength": 2,
513
+ "pragmeval/sarcasm": 1,
514
+ "pragmeval/squinky-formality": 1,
515
+ "pragmeval/squinky-implicature": 1,
516
+ "pragmeval/squinky-informativeness": 1,
517
+ "pragmeval/stac": 2,
518
+ "pragmeval/switchboard": 1,
519
+ "pragmeval/verifiability": 1,
520
+ "privacy-200k-Mistral-Large-3": 29,
521
+ "prm800k_dpo/solution": 124,
522
+ "prm800k_dpo/step": 113,
523
+ "probability_words_nli/reasoning_1hop": 14,
524
+ "probability_words_nli/reasoning_2hop": 11,
525
+ "procedural-typed-decisions/arithmetic": 120,
526
+ "procedural-typed-decisions/entity_belief_tracking": 120,
527
+ "procedural-typed-decisions/event_state_reconstruction": 120,
528
+ "procedural-typed-decisions/evidence_sufficiency": 120,
529
+ "procedural-typed-decisions/multi_view_adjudication": 120,
530
+ "procedural-typed-decisions/needle_retrieval": 120,
531
+ "procedural-typed-decisions/partial_observation_calibration": 120,
532
+ "procedural-typed-decisions/policy_applicability": 117,
533
+ "procedural-typed-decisions/policy_under_uncertainty": 120,
534
+ "procedural-typed-decisions/record_aggregation": 120,
535
+ "procedural-typed-decisions/state_perturbation": 120,
536
+ "procedural-typed-decisions/table_lookup": 60,
537
+ "procedural-typed-decisions/taxonomy_routing": 120,
538
+ "prompt-injection-dataset": 23,
539
+ "prompt-injections": 22,
540
+ "proofwriter": 30,
541
+ "propsegment/nli": 26,
542
+ "prost": 30,
543
+ "proto_qa/proto_qa": 74,
544
+ "qasc": 38,
545
+ "quail": 57,
546
+ "quarel": 10,
547
+ "quartz": 51,
548
+ "race-c": 26,
549
+ "race/high": 31,
550
+ "race/middle": 29,
551
+ "recast/recast_factuality": 3,
552
+ "recast/recast_megaveridicality": 3,
553
+ "recast/recast_ner": 3,
554
+ "recast/recast_puns": 1,
555
+ "recast/recast_sentiment": 4,
556
+ "recast/recast_verbcorner": 5,
557
+ "recast/recast_verbnet": 2,
558
+ "recast_white/dpr": 6,
559
+ "recast_white/fnplus": 10,
560
+ "recast_white/sprl": 6,
561
+ "reclor": 31,
562
+ "regset": 32,
563
+ "resnli": 22,
564
+ "riddle_sense": 60,
565
+ "robustLR": 28,
566
+ "robust_nli/IS_CS": 1,
567
+ "robust_nli/LI_LI": 1,
568
+ "robust_nli/PI_CD": 1,
569
+ "robust_nli_is_sd": 2,
570
+ "robust_nli_li_ts": 3,
571
+ "rotten_tomatoes": 22,
572
+ "ruletaker": 21,
573
+ "rumoureval_2019/RumourEval2019": 29,
574
+ "safe-guard-prompt-injection": 22,
575
+ "scicite": 25,
576
+ "scientific-exaggeration-detection": 10,
577
+ "scifact_entailment": 31,
578
+ "sciie": 34,
579
+ "scinli": 28,
580
+ "sciq": 31,
581
+ "scitail/snli_format": 25,
582
+ "scone": 22,
583
+ "scruples/verdict_votes": 75,
584
+ "seahorse_summarization_evaluation": 110,
585
+ "sem_eval_2010_task_8": 33,
586
+ "semantic_fragments_nli": 30,
587
+ "sen-making/1": 24,
588
+ "sen-making/2": 25,
589
+ "sharc": 27,
590
+ "shell-safety-v2": 30,
591
+ "sherliic": 20,
592
+ "sick/label": 29,
593
+ "sick/relatedness": 74,
594
+ "silicone/dyda_da": 8,
595
+ "silicone/maptask": 6,
596
+ "silicone/oasis": 8,
597
+ "silicone/sem": 5,
598
+ "simple_pair": 17,
599
+ "sms_spam": 22,
600
+ "snips_built_in_intents": 16,
601
+ "snli": 29,
602
+ "social_i_qa": 60,
603
+ "spartqa-mchoice": 89,
604
+ "spartqa-yn": 45,
605
+ "stackoverflow-questions": 28,
606
+ "starcon": 22,
607
+ "stepgame": 32,
608
+ "strategy-qa": 23,
609
+ "sts-companion": 90,
610
+ "subjectivity": 22,
611
+ "summarize_from_feedback/comparisons": 244,
612
+ "super_glue/axg": 5,
613
+ "super_glue/boolq": 2,
614
+ "super_glue/boolq_passage": 4,
615
+ "super_glue/cb": 5,
616
+ "super_glue/copa": 6,
617
+ "super_glue/multirc": 2,
618
+ "super_glue/wic": 3,
619
+ "swag/regular": 61,
620
+ "syntactic-augmentation-nli": 27,
621
+ "synthetic-instruct-gptj-pairwise": 26,
622
+ "synthetic-retrieval-NLI/binary": 8,
623
+ "synthetic-retrieval-NLI/count": 10,
624
+ "synthetic-retrieval-NLI/position": 10,
625
+ "tasksource_dpo_pairs": 231,
626
+ "temporal-nli": 28,
627
+ "tomi-nli": 34,
628
+ "toxic-chat/toxicchat0124/jailbreaking": 11,
629
+ "toxic-chat/toxicchat0124/toxicity": 12,
630
+ "toxic_conversations": 22,
631
+ "toxigen-data/annotated": 22,
632
+ "tracie": 27,
633
+ "trec": 38,
634
+ "tweet_eval/emoji": 4,
635
+ "tweet_eval/emotion": 2,
636
+ "tweet_eval/hate": 2,
637
+ "tweet_eval/irony": 5,
638
+ "tweet_eval/offensive": 3,
639
+ "tweet_eval/sentiment": 2,
640
+ "tweet_eval/stance_abortion": 2,
641
+ "tweet_eval/stance_atheism": 2,
642
+ "tweet_eval/stance_climate": 2,
643
+ "tweet_eval/stance_feminist": 2,
644
+ "tweet_eval/stance_hillary": 2,
645
+ "tweets_hate_speech_detection": 24,
646
+ "twentyquestions": 36,
647
+ "twitter-financial-news-sentiment": 29,
648
+ "utilitarianism": 11,
649
+ "vitaminc": 31,
650
+ "webgpt_comparisons": 27,
651
+ "wice": 28,
652
+ "wiki_hop/original": 70,
653
+ "wiki_qa": 23,
654
+ "wikimedqa/medwiki": 73,
655
+ "wildguardmix-cleaned/prompt_harm": 6,
656
+ "wildguardmix-cleaned/response_harm": 7,
657
+ "wildguardmix-cleaned/response_refusal": 9,
658
+ "winodict": 49,
659
+ "winogrande/winogrande_xl": 47,
660
+ "winowhy": 22,
661
+ "wiqa": 58,
662
+ "wnut_17/wnut_17": 223,
663
+ "wouldyourather": 29,
664
+ "yahoo_answers_topics": 33,
665
+ "yelp_review_full/yelp_review_full": 28
666
+ },
667
+ "kinds_retained": {
668
+ "choice": 11679,
669
+ "noul": 2042,
670
+ "score": 1010
671
+ },
672
+ "license_use_retained": {
673
+ "commercial": 8452,
674
+ "non-commercial": 1581,
675
+ "unspecified": 4698
676
+ },
677
+ "partially_retained_groups": 39,
678
+ "row_content_sha256": "de5be6ee6a34f82a2f2e388ac437a009abc4749bbb5e9d4cbfc0e6321b90339d",
679
+ "sha256": "662a32ff08b79f7077d3ef3ccd2e986004466db8bd4c966b44de3836fe023f8b",
680
+ "parquet_bytes": 3557968,
681
+ "arrow_bytes": 17814120
682
+ },
683
+ "test": {
684
+ "source_rows": 15000,
685
+ "retained_rows": 14745,
686
+ "dispositions": {
687
+ "keep": 13561,
688
+ "uncertain": 1184,
689
+ "drop": 255
690
+ },
691
+ "soft_target_rows_retained": 1390,
692
+ "sources_retained": {
693
+ "AES2-essay-scoring": 17,
694
+ "AdjectiveScaleProbe-nli": 29,
695
+ "AmbigNQ-clarifying-question": 21,
696
+ "BeaverTails": 23,
697
+ "CONDAQA": 26,
698
+ "CREAK": 23,
699
+ "ConTRoL-nli": 134,
700
+ "Dilemmas_Disagreement": 11,
701
+ "Dynasent_Disagreement": 11,
702
+ "FLD.v2/default": 13,
703
+ "FLD.v2/star": 15,
704
+ "FLUTE": 23,
705
+ "FOL-nli": 81,
706
+ "HatemojiBuild": 23,
707
+ "HelpSteer/coherence": 3,
708
+ "HelpSteer/complexity": 3,
709
+ "HelpSteer/correctness": 4,
710
+ "HelpSteer/helpfulness": 3,
711
+ "HelpSteer/verbosity": 2,
712
+ "HelpSteer2/coherence": 6,
713
+ "HelpSteer2/complexity": 8,
714
+ "HelpSteer2/correctness": 4,
715
+ "HelpSteer2/helpfulness": 6,
716
+ "HelpSteer2/verbosity": 4,
717
+ "HelpSteer3/edit_quality": 26,
718
+ "HelpSteer3/feedback": 12,
719
+ "HelpSteer3/preference": 23,
720
+ "HelpSteer3/preference_strength": 10,
721
+ "HelpSteer3/principle": 10,
722
+ "I2D2": 23,
723
+ "IntentGrasp/all": 73,
724
+ "LogicNLI": 29,
725
+ "MOH": 24,
726
+ "MSciNLI": 26,
727
+ "MedQA-USMLE-4-options-hf": 61,
728
+ "PARADISE": 61,
729
+ "PARARULE-Plus": 22,
730
+ "PKU-SafeRLHF/helpfulness": 24,
731
+ "PKU-SafeRLHF/safety": 24,
732
+ "Pol_NLI": 23,
733
+ "Politeness_Disagreement": 12,
734
+ "Prompt-injection-dataset/full": 21,
735
+ "PromptShield": 24,
736
+ "ReSQ": 43,
737
+ "SBIC_Disagreement": 12,
738
+ "SChem_Disagreement": 12,
739
+ "SDOH-NLI": 11,
740
+ "SHP": 24,
741
+ "Sarcasm_News_Headline": 23,
742
+ "ScienceQA_text_only": 46,
743
+ "ShellRisk-Bench": 22,
744
+ "SpaRTUN": 48,
745
+ "SpaceNLI": 28,
746
+ "Touche23-ValueEval": 13,
747
+ "TroFi": 22,
748
+ "TuringBench": 50,
749
+ "UNLI": 505,
750
+ "UltraFeedback-paired": 53,
751
+ "VUAC": 11,
752
+ "WANLI": 85,
753
+ "acceptability-prediction/binary_votes": 22,
754
+ "acceptability-prediction/rating_votes": 59,
755
+ "add_one_rte": 22,
756
+ "ade_corpus_v2/Ade_corpus_v2_classification": 18,
757
+ "ag_news": 28,
758
+ "agent_action_safety": 13,
759
+ "ai2_arc/ARC-Challenge/challenge": 17,
760
+ "ai2_arc/ARC-Easy/challenge": 17,
761
+ "amazon_counterfactual/en": 23,
762
+ "amazon_polarity/amazon_polarity": 22,
763
+ "ambient": 24,
764
+ "anli/a1": 29,
765
+ "anli/a2": 22,
766
+ "anli/a3": 25,
767
+ "app_reviews": 25,
768
+ "apt": 22,
769
+ "arct": 47,
770
+ "args_me": 22,
771
+ "argument-feedback": 14,
772
+ "art": 48,
773
+ "auditor_review": 29,
774
+ "autotnli": 26,
775
+ "avicenna": 22,
776
+ "babi_nli/basic-coreference": 2,
777
+ "babi_nli/basic-deduction": 2,
778
+ "babi_nli/basic-induction": 2,
779
+ "babi_nli/compound-coreference": 2,
780
+ "babi_nli/conjunction": 1,
781
+ "babi_nli/counting": 4,
782
+ "babi_nli/indefinite-knowledge": 1,
783
+ "babi_nli/lists-sets": 1,
784
+ "babi_nli/path-finding": 1,
785
+ "babi_nli/positional-reasoning": 2,
786
+ "babi_nli/simple-negation": 1,
787
+ "babi_nli/single-supporting-fact": 1,
788
+ "babi_nli/size-reasoning": 3,
789
+ "babi_nli/three-arg-relations": 1,
790
+ "babi_nli/three-supporting-facts": 1,
791
+ "babi_nli/time-reasoning": 1,
792
+ "babi_nli/two-arg-relations": 4,
793
+ "babi_nli/two-supporting-facts": 1,
794
+ "babi_nli/yes-no-questions": 4,
795
+ "balanced-copa": 27,
796
+ "banking77": 38,
797
+ "biosift-nli": 22,
798
+ "blog_authorship_corpus/age": 9,
799
+ "blog_authorship_corpus/gender": 12,
800
+ "blog_authorship_corpus/job": 8,
801
+ "boolq-natural-perturbations": 21,
802
+ "brainteasers/SP": 22,
803
+ "brainteasers/WP": 23,
804
+ "breaking_nli": 29,
805
+ "chaos-mnli-ambiguity/votes": 78,
806
+ "chatbot_arena_conversations": 50,
807
+ "chemprot/chemprot_full_source": 31,
808
+ "cicero": 57,
809
+ "circa": 29,
810
+ "citation_intent": 34,
811
+ "civil_comments/identity_attack_share": 10,
812
+ "civil_comments/insult_share": 9,
813
+ "civil_comments/obscene_share": 9,
814
+ "civil_comments/severe_toxicity_share": 9,
815
+ "civil_comments/sexual_explicit_share": 9,
816
+ "civil_comments/threat_share": 9,
817
+ "civil_comments/toxicity_share": 9,
818
+ "cladder": 22,
819
+ "clcd-english": 24,
820
+ "clinc_oos/plus": 38,
821
+ "cloth": 66,
822
+ "clutrr": 30,
823
+ "cnli": 28,
824
+ "codah/codah": 31,
825
+ "code_x_glue_cc_defect_detection": 23,
826
+ "com2sense": 26,
827
+ "commonsense_qa": 64,
828
+ "commonsense_qa_2.0": 23,
829
+ "conceptrules_v2": 33,
830
+ "conj_nli": 27,
831
+ "conll2003/ner_tags": 222,
832
+ "contract-nli/contractnli_a/seg": 13,
833
+ "contract-nli/contractnli_b/full": 14,
834
+ "corr2cause": 42,
835
+ "cos_e/v1.0": 63,
836
+ "cosmos_qa": 60,
837
+ "counterfactually-augmented-imdb": 22,
838
+ "counterfactually-augmented-snli": 26,
839
+ "crowdflower/airline-sentiment": 3,
840
+ "crowdflower/corporate-messaging": 3,
841
+ "crowdflower/political-media-audience": 6,
842
+ "crowdflower/political-media-bias": 2,
843
+ "crowdflower/political-media-message": 3,
844
+ "crowdflower/sentiment_nuclear_power": 3,
845
+ "crowdflower/text_emotion": 4,
846
+ "crowdflower/tweet_global_warming": 3,
847
+ "cycic_classification": 23,
848
+ "cycic_multiplechoice": 62,
849
+ "dadc-limit-nli": 21,
850
+ "dataset_train_nli": 111,
851
+ "dbpedia_14/dbpedia_14": 18,
852
+ "defeasible-nli/atomic": 8,
853
+ "defeasible-nli/snli": 7,
854
+ "defeasible-nli/social": 7,
855
+ "definite_pronoun_resolution": 52,
856
+ "dgen": 32,
857
+ "discosense": 59,
858
+ "discovery/discovery": 41,
859
+ "disrpt/eng.dep.scidtb.rels": 41,
860
+ "dnc": 23,
861
+ "dnd_style_intents": 34,
862
+ "doc-nli": 109,
863
+ "docred": 60,
864
+ "dream": 31,
865
+ "dynahate": 22,
866
+ "dynasent/r1_votes": 117,
867
+ "dynasent/r2_votes": 120,
868
+ "e-CARE": 49,
869
+ "ekar_english": 65,
870
+ "emo/emo2019": 14,
871
+ "emotion": 33,
872
+ "english-grading/cohesion": 3,
873
+ "english-grading/conventions": 3,
874
+ "english-grading/grammar": 2,
875
+ "english-grading/phraseology": 2,
876
+ "english-grading/syntax": 2,
877
+ "english-grading/vocabulary": 2,
878
+ "equate": 26,
879
+ "esci": 13,
880
+ "ethics/commonsense": 6,
881
+ "ethics/deontology": 6,
882
+ "ethics/justice": 5,
883
+ "ethics/virtue": 5,
884
+ "ethos/binary": 13,
885
+ "ethos/multilabel": 11,
886
+ "feasibilityQA": 22,
887
+ "fever-evidence-related": 14,
888
+ "few_rel/default": 22,
889
+ "fig-qa": 49,
890
+ "financial_phrasebank/sentences_allagree": 14,
891
+ "folio": 28,
892
+ "fool-me-twice": 22,
893
+ "fracas": 19,
894
+ "gen_debiased_nli/mnli_seq_z": 1,
895
+ "gen_debiased_nli/mnli_z_aug": 1,
896
+ "github-issue-similarity": 22,
897
+ "glue/cola": 4,
898
+ "glue/mnli": 3,
899
+ "glue/mrpc": 1,
900
+ "glue/qnli": 3,
901
+ "glue/qqp": 2,
902
+ "glue/rte": 2,
903
+ "glue/sst2": 3,
904
+ "glue/wnli": 2,
905
+ "go_emotions/simplified": 20,
906
+ "goal-step-wikihow/goal": 18,
907
+ "goal-step-wikihow/order": 20,
908
+ "goal-step-wikihow/step": 22,
909
+ "google_wellformed_query": 31,
910
+ "hans": 22,
911
+ "hate_speech18": 24,
912
+ "hate_speech_offensive": 28,
913
+ "head_qa/en": 58,
914
+ "headline_cause/en_simple": 27,
915
+ "health_fact": 27,
916
+ "hellaswag": 61,
917
+ "help-nli": 35,
918
+ "hh-rlhf/harmless-base": 14,
919
+ "hh-rlhf/helpful-base": 13,
920
+ "hh-rlhf/helpful-online": 12,
921
+ "hh-rlhf/helpful-rejection-sampled": 10,
922
+ "hlgd": 20,
923
+ "hope_edi/english": 26,
924
+ "hover": 21,
925
+ "humicroedit/subtask-1": 10,
926
+ "humicroedit/subtask-2": 4,
927
+ "hyperpartisan_news": 15,
928
+ "idioms-nli": 25,
929
+ "imdb": 22,
930
+ "implicatures": 47,
931
+ "implicit-hate-stg1": 32,
932
+ "insincere-questions": 22,
933
+ "it-support-tickets": 35,
934
+ "jigsaw_toxicity": 22,
935
+ "joci": 28,
936
+ "language-identification": 16,
937
+ "lewidi/hs_brexit": 21,
938
+ "lewidi/md_agreement": 22,
939
+ "lewidi/mp": 21,
940
+ "lex_glue/case_hold": 62,
941
+ "lex_glue/ledgar": 11,
942
+ "lex_glue/scotus": 10,
943
+ "lex_glue/unfair_tos": 11,
944
+ "lexical_relation_classification/BLESS": 6,
945
+ "lexical_relation_classification/CogALexV": 7,
946
+ "lexical_relation_classification/EVALution": 6,
947
+ "lexical_relation_classification/K&H+N": 5,
948
+ "lexical_relation_classification/ROOT09": 5,
949
+ "liar": 16,
950
+ "lifecycle-entailment": 21,
951
+ "lingnli": 28,
952
+ "linguisticprobing/bigram_shift": 1,
953
+ "linguisticprobing/coordination_inversion": 1,
954
+ "logical-entailment": 22,
955
+ "logical-fallacy": 34,
956
+ "logiqa": 55,
957
+ "logiqa-2.0-nli": 29,
958
+ "lonli": 28,
959
+ "lsat-ar": 27,
960
+ "lsat-rc": 30,
961
+ "lsat_qa/all": 16,
962
+ "math_qa": 52,
963
+ "mc_taco": 22,
964
+ "mctest-nli": 111,
965
+ "medical_questions_pairs": 21,
966
+ "medmcqa": 59,
967
+ "mindgames": 22,
968
+ "missing-item-prediction/contrastive": 11,
969
+ "monli": 22,
970
+ "monotonicity-entailment": 23,
971
+ "moral_stories/full": 49,
972
+ "mpe": 30,
973
+ "multilingual/AfriSenti-twitter-sentiment/amh": 3,
974
+ "multilingual/AfriSenti-twitter-sentiment/arq": 3,
975
+ "multilingual/AfriSenti-twitter-sentiment/ary": 4,
976
+ "multilingual/AfriSenti-twitter-sentiment/hau": 2,
977
+ "multilingual/AfriSenti-twitter-sentiment/ibo": 2,
978
+ "multilingual/AfriSenti-twitter-sentiment/kin": 3,
979
+ "multilingual/AfriSenti-twitter-sentiment/pcm": 3,
980
+ "multilingual/AfriSenti-twitter-sentiment/por": 2,
981
+ "multilingual/AfriSenti-twitter-sentiment/swa": 2,
982
+ "multilingual/AfriSenti-twitter-sentiment/tso": 2,
983
+ "multilingual/AfriSenti-twitter-sentiment/twi": 2,
984
+ "multilingual/AfriSenti-twitter-sentiment/yor": 2,
985
+ "multilingual/MLMA_hate_speech": 16,
986
+ "multilingual/NusaX-senti/ace": 3,
987
+ "multilingual/NusaX-senti/ban": 3,
988
+ "multilingual/NusaX-senti/bbc": 3,
989
+ "multilingual/NusaX-senti/bjn": 2,
990
+ "multilingual/NusaX-senti/bug": 2,
991
+ "multilingual/NusaX-senti/eng": 2,
992
+ "multilingual/NusaX-senti/ind": 3,
993
+ "multilingual/NusaX-senti/jav": 2,
994
+ "multilingual/NusaX-senti/mad": 2,
995
+ "multilingual/NusaX-senti/min": 2,
996
+ "multilingual/NusaX-senti/nij": 3,
997
+ "multilingual/NusaX-senti/sun": 2,
998
+ "multilingual/amazon_reviews_multi/all_languages": 29,
999
+ "multilingual/americas_nli/all_languages": 28,
1000
+ "multilingual/clue/afqmc": 12,
1001
+ "multilingual/clue/ocnli": 8,
1002
+ "multilingual/clue/tnews": 8,
1003
+ "multilingual/disrpt/deu.rst.pcc.rels": 5,
1004
+ "multilingual/disrpt/eus.rst.ert.rels": 6,
1005
+ "multilingual/disrpt/fas.rst.prstc.rels": 3,
1006
+ "multilingual/disrpt/fra.sdrt.annodis.rels": 3,
1007
+ "multilingual/disrpt/nld.rst.nldt.rels": 3,
1008
+ "multilingual/disrpt/por.rst.cstn.rels": 3,
1009
+ "multilingual/disrpt/rus.rst.rrt.rels": 5,
1010
+ "multilingual/disrpt/spa.rst.rststb.rels": 6,
1011
+ "multilingual/disrpt/tha.pdtb.tdtb.rels": 3,
1012
+ "multilingual/disrpt/zho.rst.gcdt.rels": 3,
1013
+ "multilingual/exams/multilingual": 59,
1014
+ "multilingual/indic_glue/actsa-sc.te/sentiment": 4,
1015
+ "multilingual/indic_glue/bbca.hi/news": 5,
1016
+ "multilingual/indic_glue/iitp-mr.hi/sentiment": 7,
1017
+ "multilingual/indic_glue/iitp-pr.hi/sentiment": 3,
1018
+ "multilingual/indic_glue/inltkh": 3,
1019
+ "multilingual/indic_glue/inltkh.te/sentiment": 4,
1020
+ "multilingual/indic_glue/md.hi/discourse_mode": 4,
1021
+ "multilingual/indic_glue/sna.bn/news": 3,
1022
+ "multilingual/indic_glue/wstp": 59,
1023
+ "multilingual/klue/nli": 14,
1024
+ "multilingual/klue/sts": 10,
1025
+ "multilingual/klue/ynat": 9,
1026
+ "multilingual/language-identification": 16,
1027
+ "multilingual/masakhanews/amh": 2,
1028
+ "multilingual/masakhanews/eng": 5,
1029
+ "multilingual/masakhanews/fra": 3,
1030
+ "multilingual/masakhanews/hau": 2,
1031
+ "multilingual/masakhanews/ibo": 2,
1032
+ "multilingual/masakhanews/lin": 2,
1033
+ "multilingual/masakhanews/lug": 2,
1034
+ "multilingual/masakhanews/orm": 2,
1035
+ "multilingual/masakhanews/pcm": 1,
1036
+ "multilingual/masakhanews/run": 1,
1037
+ "multilingual/masakhanews/sna": 1,
1038
+ "multilingual/masakhanews/som": 2,
1039
+ "multilingual/masakhanews/swa": 1,
1040
+ "multilingual/masakhanews/tir": 1,
1041
+ "multilingual/masakhanews/xho": 1,
1042
+ "multilingual/masakhanews/yor": 1,
1043
+ "multilingual/massive": 39,
1044
+ "multilingual/miam": 33,
1045
+ "multilingual/mms": 30,
1046
+ "multilingual/mtop": 39,
1047
+ "multilingual/multilingual-NLI-26lang-2mil7": 27,
1048
+ "multilingual/multilingual-sentiments/all": 28,
1049
+ "multilingual/oasst1_pairwise_rlhf_reward": 49,
1050
+ "multilingual/offenseval_2020/ar": 5,
1051
+ "multilingual/offenseval_2020/da": 9,
1052
+ "multilingual/offenseval_2020/gr": 5,
1053
+ "multilingual/offenseval_2020/tr": 4,
1054
+ "multilingual/offenseval_dravidian/kannada": 17,
1055
+ "multilingual/offenseval_dravidian/malayalam": 10,
1056
+ "multilingual/offenseval_dravidian/tamil": 9,
1057
+ "multilingual/paws-x/de": 4,
1058
+ "multilingual/paws-x/en": 3,
1059
+ "multilingual/paws-x/es": 3,
1060
+ "multilingual/paws-x/fr": 3,
1061
+ "multilingual/paws-x/ja": 2,
1062
+ "multilingual/paws-x/ko": 3,
1063
+ "multilingual/paws-x/zh": 2,
1064
+ "multilingual/tweet_sentiment_multilingual": 28,
1065
+ "multilingual/tydi-as2-balanced": 24,
1066
+ "multilingual/universal-joy": 28,
1067
+ "multilingual/wili_2018": 38,
1068
+ "multilingual/x-fact": 18,
1069
+ "multilingual/x-stance": 22,
1070
+ "multilingual/xcopa/et": 3,
1071
+ "multilingual/xcopa/ht": 4,
1072
+ "multilingual/xcopa/id": 2,
1073
+ "multilingual/xcopa/it": 4,
1074
+ "multilingual/xcopa/qu": 3,
1075
+ "multilingual/xcopa/sw": 3,
1076
+ "multilingual/xcopa/ta": 2,
1077
+ "multilingual/xcopa/th": 1,
1078
+ "multilingual/xcopa/tr": 2,
1079
+ "multilingual/xcopa/translation-et": 4,
1080
+ "multilingual/xcopa/translation-ht": 2,
1081
+ "multilingual/xcopa/translation-id": 2,
1082
+ "multilingual/xcopa/translation-it": 2,
1083
+ "multilingual/xcopa/translation-sw": 2,
1084
+ "multilingual/xcopa/translation-ta": 2,
1085
+ "multilingual/xcopa/translation-th": 2,
1086
+ "multilingual/xcopa/translation-tr": 2,
1087
+ "multilingual/xcopa/translation-vi": 2,
1088
+ "multilingual/xcopa/translation-zh": 2,
1089
+ "multilingual/xcopa/vi": 2,
1090
+ "multilingual/xcopa/zh": 2,
1091
+ "multilingual/xcsr/X-CODAH-ar": 3,
1092
+ "multilingual/xcsr/X-CODAH-de": 2,
1093
+ "multilingual/xcsr/X-CODAH-en": 2,
1094
+ "multilingual/xcsr/X-CODAH-es": 3,
1095
+ "multilingual/xcsr/X-CODAH-fr": 2,
1096
+ "multilingual/xcsr/X-CODAH-hi": 2,
1097
+ "multilingual/xcsr/X-CODAH-it": 2,
1098
+ "multilingual/xcsr/X-CODAH-jap": 2,
1099
+ "multilingual/xcsr/X-CODAH-nl": 3,
1100
+ "multilingual/xcsr/X-CODAH-pl": 2,
1101
+ "multilingual/xcsr/X-CODAH-pt": 2,
1102
+ "multilingual/xcsr/X-CODAH-ru": 2,
1103
+ "multilingual/xcsr/X-CODAH-sw": 2,
1104
+ "multilingual/xcsr/X-CODAH-ur": 3,
1105
+ "multilingual/xcsr/X-CODAH-vi": 2,
1106
+ "multilingual/xcsr/X-CODAH-zh": 3,
1107
+ "multilingual/xcsr/X-CSQA-ar": 2,
1108
+ "multilingual/xcsr/X-CSQA-de": 2,
1109
+ "multilingual/xcsr/X-CSQA-en": 2,
1110
+ "multilingual/xcsr/X-CSQA-es": 2,
1111
+ "multilingual/xcsr/X-CSQA-fr": 2,
1112
+ "multilingual/xcsr/X-CSQA-hi": 2,
1113
+ "multilingual/xcsr/X-CSQA-it": 2,
1114
+ "multilingual/xcsr/X-CSQA-jap": 2,
1115
+ "multilingual/xcsr/X-CSQA-nl": 3,
1116
+ "multilingual/xcsr/X-CSQA-pl": 1,
1117
+ "multilingual/xcsr/X-CSQA-pt": 1,
1118
+ "multilingual/xcsr/X-CSQA-ru": 1,
1119
+ "multilingual/xcsr/X-CSQA-sw": 1,
1120
+ "multilingual/xcsr/X-CSQA-ur": 1,
1121
+ "multilingual/xcsr/X-CSQA-vi": 1,
1122
+ "multilingual/xcsr/X-CSQA-zh": 1,
1123
+ "multilingual/xglue/nc": 8,
1124
+ "multilingual/xglue/qadsm": 6,
1125
+ "multilingual/xglue/qam": 4,
1126
+ "multilingual/xglue/wpr": 4,
1127
+ "multilingual/xlwic/xlwic_de_de": 4,
1128
+ "multilingual/xlwic/xlwic_en_ko": 4,
1129
+ "multilingual/xlwic/xlwic_fr_fr": 7,
1130
+ "multilingual/xlwic/xlwic_it_it": 6,
1131
+ "multilingual/xnli": 84,
1132
+ "multilingual/xstory_cloze/ar": 3,
1133
+ "multilingual/xstory_cloze/en": 2,
1134
+ "multilingual/xstory_cloze/es": 3,
1135
+ "multilingual/xstory_cloze/eu": 2,
1136
+ "multilingual/xstory_cloze/hi": 2,
1137
+ "multilingual/xstory_cloze/id": 2,
1138
+ "multilingual/xstory_cloze/my": 3,
1139
+ "multilingual/xstory_cloze/ru": 3,
1140
+ "multilingual/xstory_cloze/sw": 2,
1141
+ "multilingual/xstory_cloze/te": 2,
1142
+ "multilingual/xstory_cloze/zh": 2,
1143
+ "mutual": 60,
1144
+ "nan-nli": 16,
1145
+ "natural-language-satisfiability": 22,
1146
+ "naturallogic": 25,
1147
+ "nlgraph": 26,
1148
+ "nli-veridicality-transitivity": 21,
1149
+ "nli4ct_semeval2024": 22,
1150
+ "nli_fever": 27,
1151
+ "nope": 25,
1152
+ "numer_sense": 33,
1153
+ "oasst2/creativity": 13,
1154
+ "oasst2/helpfulness": 13,
1155
+ "oasst2/humor": 13,
1156
+ "oasst2/quality": 13,
1157
+ "oasst2/toxicity": 13,
1158
+ "oasst2/violence": 13,
1159
+ "oasst2_pairwise_rlhf_reward": 49,
1160
+ "onestop_qa": 31,
1161
+ "open_question_type": 33,
1162
+ "openbookqa": 32,
1163
+ "parade": 21,
1164
+ "patent-phrase-similarity": 33,
1165
+ "path-naturalness-prediction": 50,
1166
+ "paws/labeled_final": 10,
1167
+ "paws/labeled_swap": 11,
1168
+ "phrase_similarity": 20,
1169
+ "piqa": 50,
1170
+ "poem_sentiment": 18,
1171
+ "pragmeval/emergent": 2,
1172
+ "pragmeval/emobank-arousal": 3,
1173
+ "pragmeval/emobank-dominance": 2,
1174
+ "pragmeval/emobank-valence": 2,
1175
+ "pragmeval/gum": 2,
1176
+ "pragmeval/mrda": 2,
1177
+ "pragmeval/pdtb": 1,
1178
+ "pragmeval/persuasiveness-claimtype": 1,
1179
+ "pragmeval/persuasiveness-eloquence": 1,
1180
+ "pragmeval/persuasiveness-premisetype": 1,
1181
+ "pragmeval/persuasiveness-relevance": 1,
1182
+ "pragmeval/persuasiveness-specificity": 1,
1183
+ "pragmeval/persuasiveness-strength": 1,
1184
+ "pragmeval/sarcasm": 1,
1185
+ "pragmeval/squinky-formality": 1,
1186
+ "pragmeval/squinky-implicature": 1,
1187
+ "pragmeval/squinky-informativeness": 1,
1188
+ "pragmeval/stac": 1,
1189
+ "pragmeval/switchboard": 2,
1190
+ "pragmeval/verifiability": 1,
1191
+ "privacy-200k-Mistral-Large-3": 28,
1192
+ "prm800k_dpo/solution": 122,
1193
+ "prm800k_dpo/step": 116,
1194
+ "probability_words_nli/reasoning_1hop": 11,
1195
+ "probability_words_nli/reasoning_2hop": 11,
1196
+ "procedural-typed-decisions/arithmetic": 119,
1197
+ "procedural-typed-decisions/entity_belief_tracking": 120,
1198
+ "procedural-typed-decisions/event_state_reconstruction": 120,
1199
+ "procedural-typed-decisions/evidence_sufficiency": 120,
1200
+ "procedural-typed-decisions/multi_view_adjudication": 120,
1201
+ "procedural-typed-decisions/needle_retrieval": 119,
1202
+ "procedural-typed-decisions/partial_observation_calibration": 120,
1203
+ "procedural-typed-decisions/policy_applicability": 119,
1204
+ "procedural-typed-decisions/policy_under_uncertainty": 120,
1205
+ "procedural-typed-decisions/record_aggregation": 118,
1206
+ "procedural-typed-decisions/state_perturbation": 123,
1207
+ "procedural-typed-decisions/table_lookup": 60,
1208
+ "procedural-typed-decisions/taxonomy_routing": 120,
1209
+ "prompt-injection-dataset": 24,
1210
+ "prompt-injections": 22,
1211
+ "proofwriter": 27,
1212
+ "propsegment/nli": 22,
1213
+ "prost": 31,
1214
+ "proto_qa/proto_qa": 78,
1215
+ "qasc": 37,
1216
+ "quail": 57,
1217
+ "quarel": 11,
1218
+ "quartz": 50,
1219
+ "race-c": 31,
1220
+ "race/high": 26,
1221
+ "race/middle": 32,
1222
+ "recast/recast_factuality": 3,
1223
+ "recast/recast_megaveridicality": 5,
1224
+ "recast/recast_ner": 3,
1225
+ "recast/recast_puns": 3,
1226
+ "recast/recast_sentiment": 3,
1227
+ "recast/recast_verbcorner": 3,
1228
+ "recast/recast_verbnet": 2,
1229
+ "recast_white/dpr": 6,
1230
+ "recast_white/fnplus": 10,
1231
+ "recast_white/sprl": 6,
1232
+ "reclor": 32,
1233
+ "regset": 33,
1234
+ "resnli": 22,
1235
+ "riddle_sense": 65,
1236
+ "robustLR": 32,
1237
+ "robust_nli/IS_CS": 1,
1238
+ "robust_nli/LI_LI": 1,
1239
+ "robust_nli/PI_CD": 1,
1240
+ "robust_nli/ST_NE": 1,
1241
+ "robust_nli_is_sd": 3,
1242
+ "robust_nli_li_ts": 2,
1243
+ "rotten_tomatoes": 22,
1244
+ "ruletaker": 22,
1245
+ "rumoureval_2019/RumourEval2019": 29,
1246
+ "safe-guard-prompt-injection": 21,
1247
+ "scicite": 28,
1248
+ "scientific-exaggeration-detection": 27,
1249
+ "scifact_entailment": 26,
1250
+ "sciie": 35,
1251
+ "scinli": 26,
1252
+ "sciq": 30,
1253
+ "scitail/snli_format": 22,
1254
+ "scone": 23,
1255
+ "scruples/verdict_votes": 78,
1256
+ "seahorse_summarization_evaluation": 111,
1257
+ "sem_eval_2010_task_8": 33,
1258
+ "semantic_fragments_nli": 28,
1259
+ "sen-making/1": 23,
1260
+ "sen-making/2": 28,
1261
+ "sharc": 28,
1262
+ "shell-safety-v2": 28,
1263
+ "sherliic": 22,
1264
+ "sick/label": 29,
1265
+ "sick/relatedness": 79,
1266
+ "silicone/dyda_da": 7,
1267
+ "silicone/maptask": 7,
1268
+ "silicone/oasis": 8,
1269
+ "silicone/sem": 6,
1270
+ "simple_pair": 15,
1271
+ "sms_spam": 21,
1272
+ "snips_built_in_intents": 16,
1273
+ "snli": 28,
1274
+ "social_i_qa": 60,
1275
+ "spartqa-mchoice": 91,
1276
+ "spartqa-yn": 42,
1277
+ "stackoverflow-questions": 29,
1278
+ "starcon": 24,
1279
+ "stepgame": 28,
1280
+ "strategy-qa": 21,
1281
+ "sts-companion": 95,
1282
+ "subjectivity": 19,
1283
+ "summarize_from_feedback/comparisons": 240,
1284
+ "super_glue/axg": 5,
1285
+ "super_glue/boolq": 3,
1286
+ "super_glue/boolq_passage": 3,
1287
+ "super_glue/cb": 6,
1288
+ "super_glue/copa": 1,
1289
+ "super_glue/multirc": 3,
1290
+ "super_glue/wic": 3,
1291
+ "swag/regular": 61,
1292
+ "syntactic-augmentation-nli": 28,
1293
+ "synthetic-instruct-gptj-pairwise": 25,
1294
+ "synthetic-retrieval-NLI/binary": 14,
1295
+ "synthetic-retrieval-NLI/count": 7,
1296
+ "synthetic-retrieval-NLI/position": 8,
1297
+ "tasksource_dpo_pairs": 233,
1298
+ "temporal-nli": 25,
1299
+ "tomi-nli": 32,
1300
+ "toxic-chat/toxicchat0124/jailbreaking": 11,
1301
+ "toxic-chat/toxicchat0124/toxicity": 11,
1302
+ "toxic_conversations": 22,
1303
+ "toxigen-data/annotated": 21,
1304
+ "tracie": 23,
1305
+ "trec": 38,
1306
+ "tweet_eval/emoji": 2,
1307
+ "tweet_eval/emotion": 3,
1308
+ "tweet_eval/hate": 3,
1309
+ "tweet_eval/irony": 2,
1310
+ "tweet_eval/offensive": 2,
1311
+ "tweet_eval/sentiment": 2,
1312
+ "tweet_eval/stance_abortion": 2,
1313
+ "tweet_eval/stance_atheism": 2,
1314
+ "tweet_eval/stance_climate": 2,
1315
+ "tweet_eval/stance_feminist": 6,
1316
+ "tweet_eval/stance_hillary": 2,
1317
+ "tweets_hate_speech_detection": 23,
1318
+ "twentyquestions": 33,
1319
+ "twitter-financial-news-sentiment": 30,
1320
+ "utilitarianism": 12,
1321
+ "vitaminc": 28,
1322
+ "webgpt_comparisons": 27,
1323
+ "wice": 29,
1324
+ "wiki_hop/original": 75,
1325
+ "wiki_qa": 21,
1326
+ "wikimedqa/medwiki": 75,
1327
+ "wildguardmix-cleaned/prompt_harm": 10,
1328
+ "wildguardmix-cleaned/response_harm": 4,
1329
+ "wildguardmix-cleaned/response_refusal": 7,
1330
+ "winodict": 46,
1331
+ "winogrande/winogrande_xl": 49,
1332
+ "winowhy": 21,
1333
+ "wiqa": 58,
1334
+ "wnut_17/wnut_17": 219,
1335
+ "wouldyourather": 32,
1336
+ "yahoo_answers_topics": 29,
1337
+ "yelp_review_full/yelp_review_full": 28
1338
+ },
1339
+ "kinds_retained": {
1340
+ "choice": 11670,
1341
+ "noul": 2104,
1342
+ "score": 971
1343
+ },
1344
+ "license_use_retained": {
1345
+ "commercial": 8570,
1346
+ "non-commercial": 1584,
1347
+ "unspecified": 4591
1348
+ },
1349
+ "partially_retained_groups": 44,
1350
+ "row_content_sha256": "0b5ca7f8290c78e3602d58bdb4348b9814dbfae723a6a507c57178a1351f458d",
1351
+ "sha256": "36abb9d322092a217a77fec0a56c0dc7bb6b2c748e8eafc6c0742fc9e98f32ac",
1352
+ "parquet_bytes": 3516042,
1353
+ "arrow_bytes": 17734798
1354
+ }
1355
+ },
1356
+ "all_reviews_complete": true,
1357
+ "targets_modified": false,
1358
+ "downstream_predictions_consulted": false,
1359
+ "training_exclusions_applied": false,
1360
+ "retention_overrides": {
1361
+ "neutral_or_non_entailment_disagreement_preserved": 32
1362
+ },
1363
+ "rejection_issues": {
1364
+ "wrong_target": 474,
1365
+ "question_mismatch": 2,
1366
+ "missing_input": 33,
1367
+ "invalid_options": 15
1368
+ },
1369
+ "final_challenge_candidates": 691,
1370
+ "final_challenge_restorations": {
1371
+ "manual_ambiguous_procedural_advice_preserved": 1,
1372
+ "final_challenge_preserved_defensible_target": 166
1373
+ },
1374
+ "manual_retention": {
1375
+ "validation:tasksource:PARADISE-b363cb34f2:validation:1664": "The question asks for the most appropriate supplied advice. Help affording veterinary care can be relevant to treating a pet tumor; the two reports do not establish an obvious defect. Retain as uncertain."
1376
+ },
1377
+ "challenge_protocol": {
1378
+ "model": "glm_5_3_flash",
1379
+ "workers": 64,
1380
+ "prompt_sha256": "4c7090983d7870e702d1c1ab3b82b55f9faf6350e89073ec2177ee165af862f8",
1381
+ "code_sha256": "b95d6ed1f48fe379fdf826f18b8a61a3d4224180c71cafae2a6167639c13781f",
1382
+ "input_manifest_sha256": "81816eb5a684044f4c17ac86e051158a47ba7edfde45863ef0d992fb09f6bf1d",
1383
+ "judge_sha256": "a9049c1e8ac23f304936b67b3845fa9c71f5e1e5cccb1a64b00500f975cae471",
1384
+ "manual_retention": {
1385
+ "validation:tasksource:PARADISE-b363cb34f2:validation:1664": "The question asks for the most appropriate supplied advice. Help affording veterinary care can be relevant to treating a pet tumor; the two reports do not establish an obvious defect. Retain as uncertain."
1386
+ },
1387
+ "purpose": "Separate follow-up exclusion audit, added after a quality spot check; no downstream model predictions are consulted."
1388
+ },
1389
+ "source_files": {
1390
+ "validation": [
1391
+ {
1392
+ "path": "full/validation-00000-of-00001.parquet",
1393
+ "sha256": "0e901246f91cf3a869b66bd8fb7a404a438472933868b76838c0f3bdcb64f2f3",
1394
+ "rows": 15000
1395
+ }
1396
+ ],
1397
+ "test": [
1398
+ {
1399
+ "path": "full/test-00000-of-00001.parquet",
1400
+ "sha256": "a5f367eaaef12fa78ce943e0bd00d9bbde537c841d64758ca99a19b01a5a2917",
1401
+ "rows": 15000
1402
+ }
1403
+ ]
1404
+ },
1405
+ "review_prompt_sha256": "c3c7ff391a9091b594ab9f3b2c684718b119cb285419aba9fd9e4d1cc964e79a",
1406
+ "review_code_sha256": {
1407
+ "judge.py": "a9049c1e8ac23f304936b67b3845fa9c71f5e1e5cccb1a64b00500f975cae471",
1408
+ "run_filter.py": "9e9f605a1cbcd264610797246b384b4947219fa3749a095ef7bd6cfbff539b46",
1409
+ "retention.py": "6e9dbf4e36c39ec76e60471d92eeb4af4fd60503e02475125896bf0ebc92caae",
1410
+ "prepare_eval.py": "3b2204882bc8941fb807001f706052ff2f9c8033825082d629ca05cbe812afe3",
1411
+ "filter_eval.py": "5fad567f90bcfeea93d1958244eb766354efd770aafcb906cf0a126eb81dfc97"
1412
+ },
1413
+ "input_manifest_sha256": "81816eb5a684044f4c17ac86e051158a47ba7edfde45863ef0d992fb09f6bf1d",
1414
+ "public_decision_ledger_sha256": "0a30b16005e53470aa86a5eed7bb43377ccb2eb408485276d1ce05971fad9779",
1415
+ "validation": {
1416
+ "every_exported_value_verified": true,
1417
+ "heldout_group_ids_disjoint": true,
1418
+ "source_labels_preserved": true,
1419
+ "training_data_unchanged": true
1420
+ }
1421
+ }
filtered-full/validation-test.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Quality-filtered validation and test
2
+
3
+ These splits extend `filtered-full` using the same frozen `full` snapshot as its
4
+ training split: revision `d2ab1d12be4463fb7ac1f877ed2193b3bb59642a`. Each source split has
5
+ 15,000 decisions. The cleanup ran on October 6, 2026. Later repository additions,
6
+ including the subsequent WebInstruct test shard, are outside this snapshot.
7
+
8
+ | Split | Original | Keep | Uncertain, retained | Rejected | Published |
9
+ |---|---:|---:|---:|---:|---:|
10
+ | validation | 15,000 | 13,640 | 1,091 | 269 | 14,731 |
11
+ | test | 15,000 | 13,561 | 1,184 | 255 | 14,745 |
12
+
13
+ All 30,000 decisions received a successful review; there are no unresolved API
14
+ errors. Original `default` and `full` splits and the existing `filtered-full/train`
15
+ files remain unchanged.
16
+
17
+ ```python
18
+ from datasets import load_dataset
19
+
20
+ validation = load_dataset(
21
+ "tasksource/tasksource-jev-typed-decisions", "filtered-full", split="validation"
22
+ )
23
+ test = load_dataset(
24
+ "tasksource/tasksource-jev-typed-decisions", "filtered-full", split="test"
25
+ )
26
+ ```
27
+
28
+ ## Quality-only selection
29
+
30
+ GLM-5.3-Flash (`glm_5_3_flash`) reviewed the source input, question, ordered options,
31
+ and original target distribution through llm-api with 768 workers. The reviewer
32
+ and prompts are the same as in the earlier full training cleanup. The 27 fixed
33
+ known-defect and preservation controls must pass before held-out review begins.
34
+
35
+ A decision is rejected only when both the initial reviewer and an independent
36
+ verification reviewer identify an obvious defect and recommend rejection.
37
+ Uncertain cases, reasonable subjective disagreements, and soft target
38
+ distributions are retained. Existing conservative guards preserve ambiguous
39
+ neutral/non-entailment and emotion-label disagreements. Review confidence is
40
+ not interpreted as a calibrated probability of a wrong label.
41
+
42
+ A spot check identified a debatable procedural-advice rejection, so a separate
43
+ final challenge audit was added. This additional stage uses the same GLM model
44
+ with high reasoning effort and up to 64 concurrent workers. It sees the original
45
+ example and both proposed rejection arguments and tries to falsify those
46
+ arguments. Its protocol and code are recorded separately from the original
47
+ frozen two-reviewer pass. The final challenger must also recommend rejection;
48
+ otherwise the example is retained as uncertain. The challenge has its own 27
49
+ control checks. One explicitly documented procedural-advice case is retained
50
+ after a conservative spot check even if automated reviewers disagree.
51
+
52
+ The challenge considered 691 proposed exclusions
53
+ and restored 167 of them. The
54
+ decision ledger records the third review and any explicit retention rationale.
55
+
56
+ Unlike the training selection, these held-out splits do **not** apply commercial
57
+ license gating, benchmark-source holdouts, training token-budget exclusions, or
58
+ filter an evaluation set against itself as a protected panel. This preserves
59
+ evaluation coverage. License fields retain their original values, including
60
+ `non-commercial` and `unspecified`; source license conditions still apply.
61
+
62
+ No decider/JEV predictions, errors, confidence scores, or downstream benchmark
63
+ scores were consulted in selection. Reviewers see gold targets because this is
64
+ a supervised data-quality audit. The original examples are never relabeled.
65
+
66
+ ## Preservation and verification
67
+
68
+ The export selects the original Parquet rows by ID, preserving all 13 columns
69
+ from that historical release: text, question, option order, target distributions,
70
+ decision kinds, source/group/question IDs, and license metadata. The raw `split`
71
+ field remains `dev` for the Hub's `validation` split and `test` for `test`.
72
+ No `example_id` is synthesized for this older schema.
73
+
74
+ All input hashes are frozen before inference. Each selected row is written in its
75
+ original order, and every exported value is verified after reading the Parquet
76
+ file back. The split metadata and Hugging Face Datasets loading are also checked.
77
+ Filtering acts on decisions, so some groups retain fewer members. Group by both
78
+ `group_id` and `state` when reconstructing related requests.
79
+
80
+ These are LLM-curated evaluation sets, not human-certified corrected benchmarks.
81
+ The reviewer can make mistakes and can change which tasks or difficulty levels
82
+ remain. Results should identify the filtered split and revision and should not
83
+ be compared directly with scores on the unfiltered splits without accounting
84
+ for the selection. Removing obvious defects does not establish universal
85
+ decontamination or independence from model pretraining.
86
+
87
+ ## Audit artifacts
88
+
89
+ - [Machine-readable report](validation-test-filtering-report.json): counts by split,
90
+ source, decision kind, and license; code/prompt fingerprints; input/output hashes.
91
+ - [Decision ledger](validation-test-decisions.jsonl.gz): every original row ID,
92
+ effective disposition, reviewer reasons, and conservative retention overrides.
93
+ Reviewer reasons are model judgments rather than ground truth. Internal service
94
+ credentials, request traces, and provider routing are not included.