assasinatee commited on
Commit
e2e5e67
·
verified ·
1 Parent(s): cb38560

Update DiCoTA: remove S2TT and repeated-tag rows, trim schema

Browse files

367,791 rows; no audio; remove S2TT; exclude 169 repeated-structure rows; remove five numeric difficulty/compression columns and simplify source note.

README.md CHANGED
@@ -8,7 +8,6 @@ configs:
8
  path: data/rl-*.parquet
9
  language:
10
  - en
11
- - zh
12
  pretty_name: DiCoTA
13
  size_categories:
14
  - 100K<n<1M
@@ -18,32 +17,29 @@ size_categories:
18
 
19
  DiCoTA is a text-only derived annotation dataset for difficulty-adaptive compressed chain-of-thought audio reasoning. It is derived from the annotations of [Audio-Reasoner-CoTA](https://huggingface.co/datasets/zhifeixie/Audio-Reasoner-CoTA). **This repository does not redistribute audio.** Each row retains the stable source coordinates needed to retrieve the corresponding CoTA audio row.
20
 
 
 
21
  ## Splits and counts
22
 
23
  | Split | Rows |
24
  |---|---:|
25
- | `sft` | 399,713 |
26
- | `rl` | 171,331 |
27
- | Total | 571,044 |
28
 
29
  | Subset | SFT | RL | Total |
30
  |---|---:|---:|---:|
31
- | S2TT | 142,149 | 60,935 | 203,084 |
32
- | audiocap | 68,283 | 29,266 | 97,549 |
33
  | clotho | 5,967 | 2,558 | 8,525 |
34
  | complex_audio | 13,410 | 5,747 | 19,157 |
35
- | meld | 20,424 | 8,754 | 29,178 |
36
- | multispeaker | 59,704 | 25,588 | 85,292 |
37
- | musicbench | 89,776 | 38,483 | 128,259 |
38
 
39
- The starting full split contains 574,994 rows. The release excludes:
40
 
41
- - 3,863 rows with unresolved AudioBox PQ/PC metadata; see `exclusions/unresolved_audiobox_exclusions.jsonl`.
42
  - 87 rows in the reviewed malformed compressed-answer denylist; see `exclusions/quality_exclusions.jsonl`.
43
-
44
- The two exclusion sets do not overlap.
45
-
46
- No additional full-data structural filtering is applied. In the retained rows, 179 compressed answers contain repeated opening tags, so rating-token insertion produces more than three rating-token occurrences. These rows are retained under the declared full-data release rule.
47
 
48
  ## Loading
49
 
@@ -56,7 +52,7 @@ print(ds[0])
56
 
57
  ## Joining audio back from CoTA
58
 
59
- This dataset stores only derived text annotations and difficulty metadata. Audio is not uploaded here. Use:
60
 
61
  ```text
62
  source_config
@@ -96,56 +92,22 @@ For efficient batch use, group DiCoTA rows by `(source_config, source_split)` an
96
  | `source_file` | Source annotation JSON path used to build this row. |
97
  | `source_row` | Zero-based row index in the corresponding CoTA split. |
98
  | `user` | Original question, including its `<audio>` marker. |
99
- | `assistant` | Compressed chain of thought with difficulty rating tokens inserted. |
100
- | `difficulty_norm` | Euclidean norm of `(D_P, D_C, D_R)`. |
101
- | `D_P`, `D_C`, `D_R` | Three difficulty axes used for rating-token insertion. |
102
  | `Reasoning_Demand` | Integer teacher score in `[1, 5]`. |
103
  | `Integration_Score` | Integer teacher score in `[1, 5]`. |
104
  | `Question_Quality` | Integer teacher score in `[1, 5]`. |
105
  | `PQ`, `PC` | AudioBox perceptual-quality and perceptual-complexity scores. |
106
- | `target_total`, `orig_len`, `compression_ratio` | Compression metadata retained from generation. |
107
-
108
- Other output rows are not modified or recomputed.
109
-
110
- ## Rating-token construction
111
-
112
- The assistant is the original compressed answer with one rating token inserted immediately after each opening `<PLANNING>`, `<CAPTION>`, and `<REASONING>` tag:
113
-
114
- ```text
115
- <PLANNING><Rn>...
116
- <CAPTION><Rn>...
117
- <REASONING><Rn>...
118
- ```
119
-
120
- The difficulty axes are:
121
-
122
- ```text
123
- D_P = 1 - Question_Quality / 5
124
- D_C = 0.5 * (Integration_Score / 5)
125
- + 0.5 * (1 - ((PQ / 10) + (PC / 10)) / 2)
126
- D_R = 0.5 * (Reasoning_Demand / 5)
127
- + 0.5 * (Integration_Score / 5)
128
- ```
129
-
130
- For each axis, `score * 5` maps to:
131
-
132
- | Range | Token |
133
- |---|---|
134
- | `[0, 1.5)` | `R1` |
135
- | `[1.5, 2.5)` | `R2` |
136
- | `[2.5, 3.5)` | `R3` |
137
- | `[3.5, 4.5)` | `R4` |
138
- | `[4.5, 5]` | `R5` |
139
 
140
  ## Reproducibility
141
 
142
  `manifest/manifest.json` records:
143
 
144
  - final split and subset counts,
145
- - exclusion counts,
146
  - output Parquet sizes, row counts, and SHA-256 hashes,
147
  - hashes for the 75 source annotation JSON files.
148
 
149
  ## Source
150
 
151
- This derived annotation dataset was built from [Audio-Reasoner-CoTA](https://huggingface.co/datasets/zhifeixie/Audio-Reasoner-CoTA). Please also cite the Audio-Reasoner work when using this data.
 
8
  path: data/rl-*.parquet
9
  language:
10
  - en
 
11
  pretty_name: DiCoTA
12
  size_categories:
13
  - 100K<n<1M
 
17
 
18
  DiCoTA is a text-only derived annotation dataset for difficulty-adaptive compressed chain-of-thought audio reasoning. It is derived from the annotations of [Audio-Reasoner-CoTA](https://huggingface.co/datasets/zhifeixie/Audio-Reasoner-CoTA). **This repository does not redistribute audio.** Each row retains the stable source coordinates needed to retrieve the corresponding CoTA audio row.
19
 
20
+ The S2TT subset is excluded from this release.
21
+
22
  ## Splits and counts
23
 
24
  | Split | Rows |
25
  |---|---:|
26
+ | `sft` | 257,440 |
27
+ | `rl` | 110,351 |
28
+ | Total | 367,791 |
29
 
30
  | Subset | SFT | RL | Total |
31
  |---|---:|---:|---:|
32
+ | audiocap | 68,220 | 29,241 | 97,461 |
 
33
  | clotho | 5,967 | 2,558 | 8,525 |
34
  | complex_audio | 13,410 | 5,747 | 19,157 |
35
+ | meld | 20,423 | 8,752 | 29,175 |
36
+ | multispeaker | 59,694 | 25,586 | 85,280 |
37
+ | musicbench | 89,726 | 38,467 | 128,193 |
38
 
39
+ From the 368,047 non-S2TT rows in the starting full split, this release excludes:
40
 
 
41
  - 87 rows in the reviewed malformed compressed-answer denylist; see `exclusions/quality_exclusions.jsonl`.
42
+ - 169 non-S2TT rows whose compressed answers contain repeated opening tags; see `exclusions/repeated_structure_exclusions.jsonl`.
 
 
 
43
 
44
  ## Loading
45
 
 
52
 
53
  ## Joining audio back from CoTA
54
 
55
+ This dataset stores only derived text annotations. Audio is not uploaded here. Use:
56
 
57
  ```text
58
  source_config
 
92
  | `source_file` | Source annotation JSON path used to build this row. |
93
  | `source_row` | Zero-based row index in the corresponding CoTA split. |
94
  | `user` | Original question, including its `<audio>` marker. |
95
+ | `assistant` | Compressed chain of thought with rating tokens. |
 
 
96
  | `Reasoning_Demand` | Integer teacher score in `[1, 5]`. |
97
  | `Integration_Score` | Integer teacher score in `[1, 5]`. |
98
  | `Question_Quality` | Integer teacher score in `[1, 5]`. |
99
  | `PQ`, `PC` | AudioBox perceptual-quality and perceptual-complexity scores. |
100
+ | `target_total`, `orig_len` | Compression metadata retained from generation. |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
101
 
102
  ## Reproducibility
103
 
104
  `manifest/manifest.json` records:
105
 
106
  - final split and subset counts,
107
+ - exclusion policy and counts,
108
  - output Parquet sizes, row counts, and SHA-256 hashes,
109
  - hashes for the 75 source annotation JSON files.
110
 
111
  ## Source
112
 
113
+ This derived annotation dataset was built from [Audio-Reasoner-CoTA](https://huggingface.co/datasets/zhifeixie/Audio-Reasoner-CoTA).
data/rl-00000.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2f15d7383a34e0e7de65754474e2f2419d0fc8160bb37ec922c898f8f8dae610
3
- size 48195149
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:91714edf361ab58b72831e28be2f542074fb3d3da90e00f7be847bc902747c1a
3
+ size 36602022
data/sft-00000.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e005176960306ffd5d17ed4b92d1442bf249dadf64f0cd0eae38befaeae6790e
3
- size 111114855
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4671e1045c6441c2ebf7afecec0b0ccd33d14d908a4b0e89f18349eae6028344
3
+ size 84410579
exclusions/repeated_structure_exclusions.jsonl ADDED
@@ -0,0 +1,169 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"id":"audiocap/audiocap_open_part10.json#5873","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
2
+ {"id":"audiocap/audiocap_open_part10.json#6357","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
3
+ {"id":"audiocap/audiocap_open_part2.json#1513","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
4
+ {"id":"audiocap/audiocap_open_part2.json#3172","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":21}
5
+ {"id":"audiocap/audiocap_open_part2.json#5316","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
6
+ {"id":"audiocap/audiocap_open_part2.json#5666","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
7
+ {"id":"audiocap/audiocap_open_part3.json#4218","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
8
+ {"id":"audiocap/audiocap_open_part3.json#6156","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
9
+ {"id":"audiocap/audiocap_open_part4.json#2532","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
10
+ {"id":"audiocap/audiocap_open_part4.json#3304","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
11
+ {"id":"audiocap/audiocap_open_part4.json#5635","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
12
+ {"id":"audiocap/audiocap_open_part5.json#1136","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
13
+ {"id":"audiocap/audiocap_open_part5.json#2038","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
14
+ {"id":"audiocap/audiocap_open_part5.json#4335","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
15
+ {"id":"audiocap/audiocap_open_part5.json#7116","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
16
+ {"id":"audiocap/audiocap_open_part6.json#4234","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
17
+ {"id":"audiocap/audiocap_open_part8.json#2480","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
18
+ {"id":"audiocap/audiocap_open_part8.json#3353","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
19
+ {"id":"audiocap/audiocap_open_part8.json#4068","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
20
+ {"id":"audiocap/audiocap_open_part8.json#6213","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
21
+ {"id":"audiocap/audiocap_open_part9.json#3432","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
22
+ {"id":"audiocap/audiocap_open_part9.json#3559","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
23
+ {"id":"audiocap/audiocap_open_part9.json#5774","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
24
+ {"id":"audiocap/audiocap_open_part9.json#628","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
25
+ {"id":"audiocap/audiocap_open_part9.json#894","split":"rl","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
26
+ {"id":"meld/open.json#12309","split":"rl","subset":"meld","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
27
+ {"id":"meld/open.json#3832","split":"rl","subset":"meld","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
28
+ {"id":"multispeaker/multispeaker_choice_part1.json#3116","split":"rl","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
29
+ {"id":"multispeaker/multispeaker_choice_part5.json#7824","split":"rl","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
30
+ {"id":"musicbench/musicbench_choice_part8.json#2889","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
31
+ {"id":"musicbench/musicbench_choice_part9.json#2711","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
32
+ {"id":"musicbench/musicbench_open_part1.json#273","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
33
+ {"id":"musicbench/musicbench_open_part3.json#6426","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
34
+ {"id":"musicbench/musicbench_open_part4.json#3332","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
35
+ {"id":"musicbench/musicbench_open_part5.json#4694","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
36
+ {"id":"musicbench/musicbench_open_part5.json#662","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
37
+ {"id":"musicbench/musicbench_open_part6.json#4064","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
38
+ {"id":"musicbench/musicbench_open_part6.json#6786","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":4}
39
+ {"id":"musicbench/musicbench_open_part7.json#2327","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
40
+ {"id":"musicbench/musicbench_open_part7.json#6403","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
41
+ {"id":"musicbench/musicbench_open_part7.json#7134","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
42
+ {"id":"musicbench/musicbench_open_part9.json#1076","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":11}
43
+ {"id":"musicbench/musicbench_open_part9.json#1337","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
44
+ {"id":"musicbench/musicbench_open_part9.json#377","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
45
+ {"id":"musicbench/musicbench_open_part9.json#779","split":"rl","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
46
+ {"id":"audiocap/audiocap_open_part10.json#4908","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
47
+ {"id":"audiocap/audiocap_open_part10.json#6458","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
48
+ {"id":"audiocap/audiocap_open_part10.json#7074","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
49
+ {"id":"audiocap/audiocap_open_part10.json#7465","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
50
+ {"id":"audiocap/audiocap_open_part2.json#2268","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
51
+ {"id":"audiocap/audiocap_open_part2.json#3196","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
52
+ {"id":"audiocap/audiocap_open_part2.json#4148","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
53
+ {"id":"audiocap/audiocap_open_part2.json#5108","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
54
+ {"id":"audiocap/audiocap_open_part2.json#5325","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
55
+ {"id":"audiocap/audiocap_open_part2.json#5402","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
56
+ {"id":"audiocap/audiocap_open_part2.json#5762","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
57
+ {"id":"audiocap/audiocap_open_part2.json#5787","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":8}
58
+ {"id":"audiocap/audiocap_open_part2.json#7255","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
59
+ {"id":"audiocap/audiocap_open_part3.json#1093","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":15}
60
+ {"id":"audiocap/audiocap_open_part3.json#1561","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
61
+ {"id":"audiocap/audiocap_open_part3.json#2566","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
62
+ {"id":"audiocap/audiocap_open_part3.json#3173","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
63
+ {"id":"audiocap/audiocap_open_part3.json#5326","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
64
+ {"id":"audiocap/audiocap_open_part3.json#638","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
65
+ {"id":"audiocap/audiocap_open_part4.json#1115","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
66
+ {"id":"audiocap/audiocap_open_part4.json#1339","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
67
+ {"id":"audiocap/audiocap_open_part4.json#2860","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
68
+ {"id":"audiocap/audiocap_open_part4.json#3311","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
69
+ {"id":"audiocap/audiocap_open_part4.json#3332","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
70
+ {"id":"audiocap/audiocap_open_part4.json#4901","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
71
+ {"id":"audiocap/audiocap_open_part4.json#5280","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
72
+ {"id":"audiocap/audiocap_open_part4.json#5977","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
73
+ {"id":"audiocap/audiocap_open_part5.json#1347","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
74
+ {"id":"audiocap/audiocap_open_part5.json#1571","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
75
+ {"id":"audiocap/audiocap_open_part5.json#1715","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
76
+ {"id":"audiocap/audiocap_open_part5.json#2499","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":8}
77
+ {"id":"audiocap/audiocap_open_part5.json#398","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
78
+ {"id":"audiocap/audiocap_open_part5.json#5261","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
79
+ {"id":"audiocap/audiocap_open_part5.json#570","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
80
+ {"id":"audiocap/audiocap_open_part5.json#7071","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
81
+ {"id":"audiocap/audiocap_open_part5.json#7139","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
82
+ {"id":"audiocap/audiocap_open_part6.json#3145","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
83
+ {"id":"audiocap/audiocap_open_part6.json#4286","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":18}
84
+ {"id":"audiocap/audiocap_open_part6.json#4323","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
85
+ {"id":"audiocap/audiocap_open_part6.json#6320","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
86
+ {"id":"audiocap/audiocap_open_part6.json#6994","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
87
+ {"id":"audiocap/audiocap_open_part6.json#7431","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
88
+ {"id":"audiocap/audiocap_open_part6.json#791","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
89
+ {"id":"audiocap/audiocap_open_part7.json#1875","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
90
+ {"id":"audiocap/audiocap_open_part7.json#1950","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
91
+ {"id":"audiocap/audiocap_open_part7.json#3021","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
92
+ {"id":"audiocap/audiocap_open_part7.json#3789","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
93
+ {"id":"audiocap/audiocap_open_part7.json#4247","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
94
+ {"id":"audiocap/audiocap_open_part7.json#4401","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":15}
95
+ {"id":"audiocap/audiocap_open_part7.json#4956","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
96
+ {"id":"audiocap/audiocap_open_part7.json#6129","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
97
+ {"id":"audiocap/audiocap_open_part7.json#7479","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
98
+ {"id":"audiocap/audiocap_open_part8.json#1489","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":30}
99
+ {"id":"audiocap/audiocap_open_part8.json#1668","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
100
+ {"id":"audiocap/audiocap_open_part8.json#2101","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
101
+ {"id":"audiocap/audiocap_open_part8.json#3098","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
102
+ {"id":"audiocap/audiocap_open_part8.json#4204","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
103
+ {"id":"audiocap/audiocap_open_part8.json#4269","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
104
+ {"id":"audiocap/audiocap_open_part8.json#598","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
105
+ {"id":"audiocap/audiocap_open_part8.json#6374","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
106
+ {"id":"audiocap/audiocap_open_part9.json#2147","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
107
+ {"id":"audiocap/audiocap_open_part9.json#7671","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
108
+ {"id":"audiocap/audiocap_open_part9.json#851","split":"sft","subset":"audiocap","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
109
+ {"id":"meld/open.json#14927","split":"sft","subset":"meld","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
110
+ {"id":"multispeaker/multispeaker_choice_part1.json#950","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
111
+ {"id":"multispeaker/multispeaker_choice_part2.json#2095","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
112
+ {"id":"multispeaker/multispeaker_choice_part4.json#6889","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
113
+ {"id":"multispeaker/multispeaker_choice_part6.json#3145","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
114
+ {"id":"multispeaker/multispeaker_choice_part7.json#2260","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
115
+ {"id":"multispeaker/multispeaker_choice_part7.json#7932","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
116
+ {"id":"multispeaker/multispeaker_choice_part8.json#3474","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
117
+ {"id":"multispeaker/multispeaker_choice_part8.json#6584","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
118
+ {"id":"multispeaker/multispeaker_choice_part9.json#5340","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
119
+ {"id":"multispeaker/multispeaker_choice_part9.json#8437","split":"sft","subset":"multispeaker","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":4}
120
+ {"id":"musicbench/musicbench_choice_part8.json#1190","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":11}
121
+ {"id":"musicbench/musicbench_choice_part8.json#2031","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":21}
122
+ {"id":"musicbench/musicbench_open_part1.json#1135","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
123
+ {"id":"musicbench/musicbench_open_part1.json#1885","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
124
+ {"id":"musicbench/musicbench_open_part1.json#6025","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
125
+ {"id":"musicbench/musicbench_open_part1.json#8046","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
126
+ {"id":"musicbench/musicbench_open_part10.json#4783","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
127
+ {"id":"musicbench/musicbench_open_part10.json#605","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
128
+ {"id":"musicbench/musicbench_open_part10.json#695","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
129
+ {"id":"musicbench/musicbench_open_part2.json#4445","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
130
+ {"id":"musicbench/musicbench_open_part2.json#4548","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
131
+ {"id":"musicbench/musicbench_open_part2.json#5245","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
132
+ {"id":"musicbench/musicbench_open_part2.json#5667","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
133
+ {"id":"musicbench/musicbench_open_part2.json#7967","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
134
+ {"id":"musicbench/musicbench_open_part2.json#9058","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
135
+ {"id":"musicbench/musicbench_open_part3.json#3688","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
136
+ {"id":"musicbench/musicbench_open_part4.json#1073","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
137
+ {"id":"musicbench/musicbench_open_part4.json#2488","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
138
+ {"id":"musicbench/musicbench_open_part4.json#427","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":21}
139
+ {"id":"musicbench/musicbench_open_part4.json#7376","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
140
+ {"id":"musicbench/musicbench_open_part4.json#7600","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":21}
141
+ {"id":"musicbench/musicbench_open_part5.json#1548","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
142
+ {"id":"musicbench/musicbench_open_part5.json#2362","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
143
+ {"id":"musicbench/musicbench_open_part5.json#3532","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
144
+ {"id":"musicbench/musicbench_open_part5.json#4006","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
145
+ {"id":"musicbench/musicbench_open_part5.json#4572","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
146
+ {"id":"musicbench/musicbench_open_part5.json#6758","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
147
+ {"id":"musicbench/musicbench_open_part6.json#1792","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
148
+ {"id":"musicbench/musicbench_open_part6.json#3782","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
149
+ {"id":"musicbench/musicbench_open_part6.json#4420","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
150
+ {"id":"musicbench/musicbench_open_part6.json#7746","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
151
+ {"id":"musicbench/musicbench_open_part6.json#8420","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
152
+ {"id":"musicbench/musicbench_open_part6.json#876","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":12}
153
+ {"id":"musicbench/musicbench_open_part7.json#4698","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
154
+ {"id":"musicbench/musicbench_open_part7.json#7501","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":21}
155
+ {"id":"musicbench/musicbench_open_part8.json#2581","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
156
+ {"id":"musicbench/musicbench_open_part8.json#6517","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
157
+ {"id":"musicbench/musicbench_open_part8.json#7302","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":9}
158
+ {"id":"musicbench/musicbench_open_part8.json#732","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
159
+ {"id":"musicbench/musicbench_open_part8.json#8833","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":5}
160
+ {"id":"musicbench/musicbench_open_part9.json#2953","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":21}
161
+ {"id":"musicbench/musicbench_open_part9.json#3476","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":7}
162
+ {"id":"musicbench/musicbench_open_part9.json#3919","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
163
+ {"id":"musicbench/musicbench_open_part9.json#4164","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
164
+ {"id":"musicbench/musicbench_open_part9.json#4236","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
165
+ {"id":"musicbench/musicbench_open_part9.json#4888","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
166
+ {"id":"musicbench/musicbench_open_part9.json#5385","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":21}
167
+ {"id":"musicbench/musicbench_open_part9.json#6603","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":15}
168
+ {"id":"musicbench/musicbench_open_part9.json#7564","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":6}
169
+ {"id":"musicbench/musicbench_open_part9.json#7994","split":"sft","subset":"musicbench","reason":"compressed answer contains repeated opening tags","rating_token_occurrences":13}
exclusions/unresolved_audiobox_exclusions.jsonl DELETED
The diff for this file is too large to render. See raw diff
 
manifest/manifest.json CHANGED
@@ -1,43 +1,47 @@
1
  {
2
- "schema_version": "dicota-manifest.v1",
3
- "dataset_version": "DiCoTA-full-no-audio-v1",
4
- "created_utc": "2026-10-08T04:46:56.848360+00:00",
5
  "source_dataset": "Audio-Reasoner-CoTA derived annotations",
 
6
  "contains_audio": false,
 
 
 
 
 
7
  "splits": {
8
  "sft": {
9
- "rows": 399713,
10
  "subset_counts": {
11
- "S2TT": 142149,
12
- "audiocap": 68283,
13
  "clotho": 5967,
14
  "complex_audio": 13410,
15
- "meld": 20424,
16
- "multispeaker": 59704,
17
- "musicbench": 89776
18
  },
19
  "task_type_counts": {
20
- "choice": 107762,
21
- "open": 291951
22
  },
23
  "files": [
24
  "data/sft-00000.parquet"
25
  ]
26
  },
27
  "rl": {
28
- "rows": 171331,
29
  "subset_counts": {
30
- "S2TT": 60935,
31
- "audiocap": 29266,
32
  "clotho": 2558,
33
  "complex_audio": 5747,
34
- "meld": 8754,
35
- "multispeaker": 25588,
36
- "musicbench": 38483
37
  },
38
  "task_type_counts": {
39
- "choice": 46224,
40
- "open": 125107
41
  },
42
  "files": [
43
  "data/rl-00000.parquet"
@@ -49,24 +53,27 @@
49
  "count": 87,
50
  "path": "exclusions/quality_exclusions.jsonl"
51
  },
52
- "unresolved_audiobox": {
53
- "count": 3863,
54
- "path": "exclusions/unresolved_audiobox_exclusions.jsonl"
 
 
 
55
  }
56
  },
57
- "assistant_postprocessing": "Insert one R1-R5 token after each PLANNING, CAPTION, and REASONING opening tag using D_P, D_C, and D_R respectively.",
58
  "files": [
59
  {
60
  "path": "data/rl-00000.parquet",
61
- "rows": 171331,
62
- "size": 48195149,
63
- "sha256": "2f15d7383a34e0e7de65754474e2f2419d0fc8160bb37ec922c898f8f8dae610"
64
  },
65
  {
66
  "path": "data/sft-00000.parquet",
67
- "rows": 399713,
68
- "size": 111114855,
69
- "sha256": "e005176960306ffd5d17ed4b92d1442bf249dadf64f0cd0eae38befaeae6790e"
70
  }
71
  ],
72
  "source_files": [
@@ -445,9 +452,5 @@
445
  "size": 60108859,
446
  "sha256": "24bb631bebf5e549f1824d361d6330f456dbec8191dcbbc9b5be12c5913b66c8"
447
  }
448
- ],
449
- "known_issues": {
450
- "description": "Retained full-data rows are not re-filtered beyond the two declared exclusion sets. Some source compressed answers contain repeated opening tags; rating-token insertion therefore produces more than three rating-token occurrences.",
451
- "rows_with_more_than_three_rating_token_occurrences": 179
452
- }
453
  }
 
1
  {
2
+ "schema_version": "dicota-manifest.v2",
3
+ "dataset_version": "DiCoTA-no-S2TT-no-audio-v2",
4
+ "created_utc": "2026-10-08T05:13:57.275471+00:00",
5
  "source_dataset": "Audio-Reasoner-CoTA derived annotations",
6
+ "previous_dataset_version": "DiCoTA-full-no-audio-v1",
7
  "contains_audio": false,
8
+ "exclusion_policy": [
9
+ "Exclude the entire S2TT subset.",
10
+ "Exclude the 87 reviewed malformed compressed-answer denylist rows.",
11
+ "Exclude non-S2TT rows whose compressed answers contain repeated opening tags."
12
+ ],
13
  "splits": {
14
  "sft": {
15
+ "rows": 257440,
16
  "subset_counts": {
17
+ "audiocap": 68220,
 
18
  "clotho": 5967,
19
  "complex_audio": 13410,
20
+ "meld": 20423,
21
+ "multispeaker": 59694,
22
+ "musicbench": 89726
23
  },
24
  "task_type_counts": {
25
+ "choice": 107750,
26
+ "open": 149690
27
  },
28
  "files": [
29
  "data/sft-00000.parquet"
30
  ]
31
  },
32
  "rl": {
33
+ "rows": 110351,
34
  "subset_counts": {
35
+ "audiocap": 29241,
 
36
  "clotho": 2558,
37
  "complex_audio": 5747,
38
+ "meld": 8752,
39
+ "multispeaker": 25586,
40
+ "musicbench": 38467
41
  },
42
  "task_type_counts": {
43
+ "choice": 46220,
44
+ "open": 64131
45
  },
46
  "files": [
47
  "data/rl-00000.parquet"
 
53
  "count": 87,
54
  "path": "exclusions/quality_exclusions.jsonl"
55
  },
56
+ "repeated_structure": {
57
+ "count": 169,
58
+ "path": "exclusions/repeated_structure_exclusions.jsonl"
59
+ },
60
+ "s2tt": {
61
+ "policy": "excluded by subset rule; the 3,863 unresolved AudioBox PQ/PC rows in v1 were within S2TT and are therefore also excluded"
62
  }
63
  },
64
+ "assistant_postprocessing": "Compressed chain of thought with source difficulty rating tokens retained.",
65
  "files": [
66
  {
67
  "path": "data/rl-00000.parquet",
68
+ "rows": 110351,
69
+ "size": 36602022,
70
+ "sha256": "91714edf361ab58b72831e28be2f542074fb3d3da90e00f7be847bc902747c1a"
71
  },
72
  {
73
  "path": "data/sft-00000.parquet",
74
+ "rows": 257440,
75
+ "size": 84410579,
76
+ "sha256": "4671e1045c6441c2ebf7afecec0b0ccd33d14d908a4b0e89f18349eae6028344"
77
  }
78
  ],
79
  "source_files": [
 
452
  "size": 60108859,
453
  "sha256": "24bb631bebf5e549f1824d361d6330f456dbec8191dcbbc9b5be12c5913b66c8"
454
  }
455
+ ]
 
 
 
 
456
  }