input_ids listlengths 2.05k 2.05k |
|---|
[
10000,
122,
2112,
10001,
0,
365,
3201,
1000,
2334,
3160,
3003,
30,
312,
3733,
1000,
1010,
2421,
63,
3,
3003,
211,
2000,
3323,
3,
10,
3061,
6609,
6236,
2303,
4143,
600,
373,
0,
620,
3111,
10000,
3302,
2236,
100,
113,
10000,
3034,
565... |
[
22,
10001,
1300,
10001,
2302,
2202,
2252,
4884,
3023,
2222,
8485,
301,
5055,
1120,
2022,
3230,
2322,
2100,
2121,
103,
2232,
2322,
2203,
1558,
6555,
2200,
2221,
3400,
2305,
2231,
22,
1000,
200,
200,
2223,
10001,
1121,
21,
2020,
1300,
2,
... |
[
3103,
400,
1744,
2332,
1211,
2222,
1000,
1434,
4024,
131,
2141,
2426,
111,
240,
3331,
2023,
4100,
3144,
10001,
2000,
1054,
3344,
3124,
304,
4105,
3242,
3133,
4112,
8448,
31,
3023,
3101,
4444,
1121,
1000,
333,
10001,
4021,
2222,
22,
3113... |
[
1001,
12,
6016,
2322,
5040,
2002,
100,
1002,
2120,
2032,
11,
2444,
1100,
26,
2,
3710,
2002,
6120,
3020,
2000,
1131,
606,
11,
203,
214,
2,
2231,
6626,
2020,
10001,
2120,
100,
1200,
333,
5630,
6176,
2032,
2030,
30,
361,
312,
1222,
240... |
[
3412,
1101,
9907,
1111,
2101,
2222,
1104,
2337,
402,
3020,
122,
2022,
2231,
2201,
123,
2421,
7177,
3130,
3033,
4020,
6106,
2122,
1211,
2120,
10001,
3335,
2335,
4202,
1013,
2242,
3651,
1121,
1332,
1121,
1011,
122,
23,
1601,
41,
110,
2122... |
[
0,
2112,
3033,
4204,
1510,
6204,
11,
1233,
8565,
2121,
111,
2111,
4120,
4686,
10001,
4031,
311,
8229,
2754,
413,
2100,
8024,
402,
2110,
1312,
4142,
1351,
205,
1232,
2003,
4940,
0,
2023,
101,
1402,
2033,
0,
40,
2401,
1000,
10001,
222,
... |
[
1033,
1221,
2030,
32,
3223,
1142,
1620,
1,
10001,
2220,
510,
511,
223,
4010,
1111,
124,
10000,
1220,
2123,
2221,
2131,
2022,
2520,
2061,
2116,
2100,
1111,
141,
3322,
3424,
10001,
3322,
2116,
1102,
32,
5,
678,
135,
1431,
510,
5242,
222... |
[
3020,
4206,
4001,
4013,
9169,
462,
0,
5440,
9883,
1214,
122,
66,
3440,
4102,
3126,
4600,
8883,
2004,
40,
1041,
4964,
5400,
10000,
2004,
110,
5572,
4522,
20,
2441,
6401,
121,
1424,
2225,
4440,
0,
1440,
10000,
1440,
22,
2222,
311,
2122,... |
[
343,
1079,
11,
1,
1070,
10000,
2123,
202,
3333,
2221,
412,
3431,
899,
3112,
332,
1303,
2341,
2,
2252,
1330,
4022,
3020,
3133,
2113,
6011,
1132,
101,
130,
4240,
314,
2011,
9097,
6021,
1020,
2422,
222,
2222,
4133,
10001,
2432,
9944,
300... |
[
5455,
1220,
6201,
5851,
50,
5511,
130,
4042,
24,
2120,
9754,
3393,
433,
10001,
2011,
2331,
2201,
113,
131,
1111,
124,
10001,
100,
2100,
2120,
10001,
1000,
1134,
3313,
1001,
11,
1303,
8627,
1300,
1204,
1141,
8322,
2330,
120,
3120,
10000,... |
[
2142,
3173,
3110,
2010,
10000,
0,
3130,
2211,
3303,
4313,
2320,
723,
3114,
1010,
4304,
2300,
2412,
3541,
301,
0,
2331,
2303,
2132,
3452,
10000,
4023,
4122,
2233,
2111,
3012,
0,
3204,
340,
3080,
414,
3322,
7740,
2332,
3322,
3303,
30,
9... |
[
1931,
8008,
3000,
3353,
2222,
3003,
2008,
1484,
2600,
120,
1737,
5122,
1222,
2210,
3005,
1474,
1002,
32,
3220,
1081,
3545,
1612,
3322,
734,
3320,
1000,
3422,
2020,
2222,
10,
7330,
2010,
1202,
4733,
2220,
10,
2204,
1102,
0,
332,
3333,
... |
[
341,
3213,
202,
2130,
5552,
2020,
8010,
2211,
1101,
6577,
1133,
3204,
3212,
5,
20,
1263,
2200,
110,
7592,
3133,
1300,
3133,
11,
1,
1311,
3012,
2353,
1311,
4022,
1222,
0,
2310,
3603,
10001,
1211,
2242,
3032,
5558,
3024,
1336,
32,
10001... |
[
3333,
3030,
404,
4433,
10001,
1361,
3034,
616,
2160,
4133,
1717,
1432,
3333,
3203,
4340,
107,
310,
4110,
6063,
9001,
3320,
3134,
414,
6616,
6937,
4441,
7861,
3143,
44,
1104,
41,
4410,
1970,
6163,
3104,
1631,
4140,
2004,
3333,
1233,
4041... |
[
3311,
10,
4101,
0,
131,
1131,
200,
7351,
1121,
302,
2123,
0,
2013,
4040,
131,
1,
1552,
1300,
1127,
30,
203,
2021,
2000,
4000,
32,
3121,
7770,
7057,
3333,
40,
2110,
5001,
1115,
21,
55,
3132,
1221,
10001,
1330,
10000,
3321,
1113,
312,... |
[
445,
3224,
10000,
301,
3250,
2202,
1002,
10000,
2123,
172,
1151,
2002,
5001,
10000,
1546,
1002,
1323,
10000,
2002,
5277,
7330,
7321,
2552,
4411,
3223,
0,
1305,
3002,
1336,
2105,
4134,
2020,
203,
10000,
442,
3323,
6223,
4242,
2321,
105,
... |
[
225,
2102,
0,
10000,
2000,
3120,
4625,
5122,
5111,
1106,
4022,
11,
1562,
5255,
6064,
8433,
2164,
2211,
1556,
2313,
1114,
6226,
21,
10000,
4642,
5054,
1201,
3510,
8353,
1011,
7776,
0,
2024,
444,
5722,
2000,
105,
5522,
3000,
3442,
710,
... |
[
1101,
1012,
1204,
2060,
2621,
9911,
460,
10000,
6222,
10001,
6406,
31,
9371,
48,
119,
1252,
344,
2066,
550,
2300,
1555,
6602,
3662,
722,
1113,
1112,
9585,
2702,
550,
2222,
381,
2721,
9716,
2242,
3322,
6424,
1100,
2220,
6622,
6345,
1292,... |
[
6,
1911,
122,
2,
206,
1230,
923,
1108,
531,
1101,
3131,
333,
2212,
1811,
106,
2112,
2212,
2400,
1021,
1315,
1001,
10001,
22,
2718,
4111,
3092,
229,
10000,
2621,
3533,
3103,
3002,
2322,
7996,
3011,
2300,
3332,
2122,
40,
3122,
8011,
66,... |
[
3133,
3230,
3221,
2212,
112,
5060,
10001,
5313,
10001,
1060,
300,
2220,
1910,
20,
4312,
9931,
242,
2503,
3121,
3333,
3152,
2535,
2121,
2000,
2122,
10000,
1110,
3363,
102,
3333,
606,
535,
1213,
1132,
3712,
5080,
1532,
250,
1051,
4211,
22... |
PPT Corpus: control_nca
Raw stage-1 pre-pretraining corpus from the ppt research framework, exported for information-theoretic analysis (e.g. m-local entropy) independent of this repo's training pipeline.
IMPORTANT: token IDs are NOT standard BPE tokens
Same NCA patch IDs as nca, with token order scrambled per chunk (see below).
These input_ids are not decodable with the Pythia (or any other)
tokenizer. They are integers in [0, 10002) with regimen-specific
meaning (see above) — treat this dataset as a corpus of symbol/event
sequences, not text.
Corpus stats
| Field | Value |
|---|---|
| Vocabulary size | 10002 |
Chunk length (seq_len) |
2048 |
| Number of chunks | 160,000 |
| Total tokens | 327,680,000 |
| Generation seed | 42 |
Each row's input_ids is a fixed-length list of 2048 integers — the
exact chunk granularity fed to the model during stage-1 training (formed
by concatenating generated sequences and splitting into non-overlapping
2048-token blocks).
Control regimen: token order within each 2048-token chunk is randomly permuted (torch.manual_seed(0) + per-row torch.randperm), preserving the exact token multiset/vocabulary distribution of nca while destroying sequential structure. This snapshot is a fixed, seeded reproduction generated at export time — for control_nca specifically, note that the live training path (ScrambledIterableDataset in src/ppt/data/control.py) draws its permutation unseeded per iteration, so this uploaded corpus is representative of the control regimen's statistics but is not a byte-identical capture of any particular training run.
Generation config
{
"grid_size": 12,
"patch_size": 2,
"num_colors": 10,
"d_state": 10,
"num_rules": 5000,
"buffer_size": 256,
"filter_threshold": 0.5,
"dT": 5,
"rollout_steps": 10,
"jax_seed": 42,
"seq_len": 2048
}
Usage
from datasets import load_dataset
ds = load_dataset("sashaboguraev/ppt-control_nca-corpus")
example = ds["train"][0]["input_ids"] # list[int], length 2048
- Downloads last month
- 40