neoneye commited on
Commit
b36c8c0
1 Parent(s): 522fa9e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +431 -3
README.md CHANGED
@@ -1,3 +1,431 @@
1
- ---
2
- license: mit
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ task_categories:
4
+ - image-to-text
5
+ - text-to-image
6
+ language:
7
+ - en
8
+ pretty_name: simons ARC (abstraction & reasoning corpus) combined datasets version 55
9
+ size_categories:
10
+ - 100K<n<1M
11
+ configs:
12
+ - config_name: default
13
+ data_files:
14
+ - split: train
15
+ path: data.jsonl
16
+ ---
17
+
18
+ # Version 1
19
+
20
+ A combination of multiple datasets.
21
+
22
+ Datasets: `dataset_solve_color.jsonl`, `dataset_solve_rotate.jsonl`, `dataset_solve_translate.jsonl`.
23
+
24
+ # Version 2
25
+
26
+ Datasets: `dataset_solve_color.jsonl`, `dataset_solve_rotate.jsonl`, `dataset_solve_translate.jsonl`.
27
+
28
+ # Version 3
29
+
30
+ Datasets: `dataset_solve_color.jsonl`, `dataset_solve_rotate.jsonl`, `dataset_solve_translate.jsonl`.
31
+
32
+ # Version 4
33
+
34
+ Added a shared dataset name for all these datasets: `SIMON-SOLVE-V1`. There may be higher version numbers in the future.
35
+ My hypothesis: Having a version number in the dataset name, it may be easier to unlearn incorrect training data.
36
+
37
+ Datasets: `dataset_solve_color.jsonl`, `dataset_solve_rotate.jsonl`, `dataset_solve_translate.jsonl`.
38
+
39
+ # Version 5
40
+
41
+ Different random seed.
42
+
43
+ # Version 6
44
+
45
+ Using `SIMON-SOLVE-V1` everywhere. Remove the `SIMON-SOLVE-COLOR`, `SIMON-SOLVE-ROTATE`, `SIMON-SOLVE-TRANSLATE`.
46
+
47
+ # Version 7
48
+
49
+ Using `SIMON-SOLVE-V1` everywhere.
50
+
51
+ # Version 8
52
+
53
+ Same settings. Different seed as usual.
54
+
55
+ # Version 9
56
+
57
+ Switching from context length 256 to context length 512.
58
+
59
+ Increasing the image sizes so the prompt length stays below 512.
60
+
61
+ `dataset_solve_color`, image size: 1-13.
62
+
63
+ `dataset_solve_rotate`, image size: 1-9.
64
+
65
+ `dataset_solve_translate`, image size: 3-9.
66
+
67
+ # Version 10
68
+
69
+ Same settings. Different seed as usual.
70
+
71
+ # Version 11
72
+
73
+ Same settings. Different seed as usual.
74
+
75
+ # Version 12
76
+
77
+ Added 1 more pair to the examples. Now it's 2-4 examples. Previously it was 2-3 examples.
78
+
79
+ # Version 13
80
+
81
+ Same settings. Different seed as usual.
82
+
83
+ # Version 14
84
+
85
+ Same settings. Different seed as usual.
86
+
87
+ # Version 15
88
+
89
+ Same settings. Different seed as usual.
90
+
91
+ # Version 16
92
+
93
+ Added `Predict the output image.`
94
+
95
+ Disabled prediction of rows.
96
+
97
+ Disabled prediction of height.
98
+
99
+ # Verison 17
100
+
101
+ Same settings. Different seed as usual.
102
+
103
+ Using the `DatasetGenerator` and the `DatasetItemListBuilder`.
104
+
105
+ # Verison 18
106
+
107
+ Added datasets.
108
+
109
+ Datasets:
110
+ - `dataset_solve_color.jsonl`
111
+ - `dataset_solve_rotate.jsonl`
112
+ - `dataset_solve_translate.jsonl`
113
+ - `dataset_cellular_automaton.jsonl` - added.
114
+ - `dataset_shape.jsonl` - added.
115
+
116
+ # Verison 19
117
+
118
+ Added dataset.
119
+
120
+ Datasets:
121
+ - `dataset_solve_color.jsonl`
122
+ - `dataset_solve_rotate.jsonl`
123
+ - `dataset_solve_translate.jsonl`
124
+ - `dataset_cellular_automaton.jsonl`
125
+ - `dataset_shape.jsonl`
126
+ - `dataset_image.jsonl` - added.
127
+
128
+ # Verison 20
129
+
130
+ Bigger images.
131
+
132
+ # Verison 21
133
+
134
+ Added dataset. Disabled some datasets.
135
+
136
+ Datasets:
137
+ - `dataset_cellular_automaton.jsonl`
138
+ - `dataset_shape.jsonl`
139
+ - `dataset_mass.jsonl` - added.
140
+
141
+ # Verison 22
142
+
143
+ Added dataset.
144
+
145
+ Datasets:
146
+ - `dataset_solve_color.jsonl`
147
+ - `dataset_solve_rotate.jsonl`
148
+ - `dataset_solve_translate.jsonl`
149
+ - `dataset_cellular_automaton.jsonl`
150
+ - `dataset_shape.jsonl`
151
+ - `dataset_image.jsonl`
152
+ - `dataset_mass.jsonl`
153
+ - `dataset_histogram.jsonl` - added.
154
+
155
+ Bigger image sizes.
156
+
157
+ Number of rows=200k. Was previously 100k rows.
158
+
159
+ # Verison 23
160
+
161
+ `datset_mass.jsonl`. increased to `max_mass=5`.
162
+
163
+ # Verison 24
164
+
165
+ `datset_mass.jsonl`. increased to `max_mass=6`.
166
+
167
+ # Verison 25
168
+
169
+ different seed.
170
+
171
+ # Verison 26
172
+
173
+ `datset_mass.jsonl`. increased to `max_mass=25`.
174
+
175
+ different seed.
176
+
177
+ # Verison 27
178
+
179
+ different seed.
180
+
181
+ # Verison 28
182
+
183
+ different seed.
184
+
185
+ # Verison 29
186
+
187
+ different seed.
188
+
189
+ # Verison 30
190
+
191
+ different seed.
192
+
193
+ # Verison 31
194
+
195
+ different seed.
196
+
197
+ # Verison 32
198
+
199
+ different seed.
200
+
201
+ # Verison 33
202
+
203
+ Disabled some dataset.
204
+
205
+ Datasets:
206
+ - `dataset_histogram.jsonl` - disabled
207
+ - `dataset_image.jsonl` - disabled
208
+ - `dataset_mass.jsonl`
209
+ - `dataset_shape.jsonl`
210
+ - `dataset_solve_color.jsonl` - disabled
211
+ - `dataset_solve_rotate.jsonl` - disabled
212
+ - `dataset_solve_translate.jsonl` - disabled
213
+ - `dataset_cellular_automaton.jsonl`
214
+
215
+ # Verison 34
216
+
217
+ Enabled all datasets.
218
+
219
+ # Version 35
220
+
221
+ Regenerated all datasets with new random seeds.
222
+
223
+ # Verison 36
224
+
225
+ Added dataset `dataset_scale.jsonl`.
226
+
227
+ Disabled some dataset.
228
+
229
+ Datasets:
230
+ - `dataset_cellular_automaton.jsonl` - disabled
231
+ - `dataset_histogram.jsonl` - disabled
232
+ - `dataset_image.jsonl` - disabled
233
+ - `dataset_mass.jsonl`
234
+ - `dataset_scale.jsonl`
235
+ - `dataset_shape.jsonl` - disabled
236
+ - `dataset_solve_color.jsonl`
237
+ - `dataset_solve_rotate.jsonl`
238
+ - `dataset_solve_translate.jsonl`
239
+
240
+ # Verison 37
241
+
242
+ Enabled all datasets
243
+
244
+ Datasets:
245
+ - `dataset_cellular_automaton.jsonl`
246
+ - `dataset_histogram.jsonl`
247
+ - `dataset_image.jsonl`
248
+ - `dataset_mass.jsonl`
249
+ - `dataset_scale.jsonl`
250
+ - `dataset_shape.jsonl`
251
+ - `dataset_solve_color.jsonl`
252
+ - `dataset_solve_rotate.jsonl`
253
+ - `dataset_solve_translate.jsonl`
254
+
255
+ # Verison 38
256
+
257
+ Added dataset
258
+
259
+ Datasets:
260
+ - `dataset_cellular_automaton.jsonl`
261
+ - `dataset_histogram.jsonl`
262
+ - `dataset_image.jsonl`
263
+ - `dataset_mass.jsonl`
264
+ - `dataset_scale.jsonl`
265
+ - `dataset_shape.jsonl`
266
+ - `dataset_solve_color.jsonl`
267
+ - `dataset_solve_rotate.jsonl`
268
+ - `dataset_solve_translate.jsonl`
269
+ - `dataset_symmetry.jsonl` - added
270
+
271
+ # Version 39
272
+
273
+ Regenerated all datasets with new random seeds.
274
+
275
+ # Version 40
276
+
277
+ Added dataset
278
+
279
+ Datasets:
280
+ - `dataset_cellular_automaton.jsonl`
281
+ - `dataset_histogram.jsonl`
282
+ - `dataset_image.jsonl`
283
+ - `dataset_image_pair.jsonl` - added
284
+ - `dataset_mass.jsonl`
285
+ - `dataset_scale.jsonl`
286
+ - `dataset_shape.jsonl`
287
+ - `dataset_solve_color.jsonl`
288
+ - `dataset_solve_rotate.jsonl`
289
+ - `dataset_solve_translate.jsonl`
290
+ - `dataset_symmetry.jsonl`
291
+
292
+ # Version 41
293
+
294
+ Regenerated all datasets with new random seeds.
295
+
296
+ # Version 42
297
+
298
+ Added dataset. Disabled some datasets.
299
+
300
+ Datasets:
301
+ - `dataset_cellular_automaton.jsonl`
302
+ - `dataset_histogram.jsonl` - disabled
303
+ - `dataset_image.jsonl` - disabled
304
+ - `dataset_image_pair.jsonl` - disabled
305
+ - `dataset_mass.jsonl` - disabled
306
+ - `dataset_scale.jsonl`
307
+ - `dataset_shape.jsonl`
308
+ - `dataset_solve_color.jsonl`
309
+ - `dataset_solve_rotate.jsonl`
310
+ - `dataset_solve_scale.jsonl` - added
311
+ - `dataset_solve_translate.jsonl`
312
+ - `dataset_symmetry.jsonl` - disabled
313
+
314
+ # Version 43
315
+
316
+ Enabled all datasets.
317
+
318
+ # Version 44
319
+
320
+ Regenerated all datasets with new random seeds.
321
+
322
+ # Version 45
323
+
324
+ Extended the `dataset_shape.jsonl` with these new `PixelConnectivity` types: `CORNER4`, `LR2`, `TB2`, `TLBR2`, `TRBL2`.
325
+ Hopefully it makes the model better at making sense of diagonal structures, which is something it's terrible at at the moment.
326
+
327
+ # Version 46
328
+
329
+ Regenerated all datasets with new random seeds.
330
+
331
+ # Version 47
332
+
333
+ Added dataset. Disabled some datasets.
334
+
335
+ Datasets:
336
+ - `dataset_cellular_automaton.jsonl`
337
+ - `dataset_histogram.jsonl`
338
+ - `dataset_image.jsonl` - disabled
339
+ - `dataset_image_pair.jsonl` - disabled
340
+ - `dataset_mass.jsonl` - disabled
341
+ - `dataset_scale.jsonl` - disabled
342
+ - `dataset_shape.jsonl` - disabled
343
+ - `dataset_solve_color.jsonl`
344
+ - `dataset_solve_rotate.jsonl`
345
+ - `dataset_solve_scale.jsonl`
346
+ - `dataset_solve_symmetry.jsonl` - added
347
+ - `dataset_solve_translate.jsonl`
348
+ - `dataset_symmetry.jsonl`
349
+
350
+ # Version 48
351
+
352
+ Enabled all datasets.
353
+
354
+ # Version 49
355
+
356
+ Bigger `max_mass`. From 6 to 8.
357
+
358
+ # Version 50
359
+
360
+ Added dataset.
361
+
362
+ Datasets:
363
+ - `dataset_cellular_automaton.jsonl`
364
+ - `dataset_histogram.jsonl`
365
+ - `dataset_image.jsonl` - disabled
366
+ - `dataset_image_pair.jsonl` - disabled
367
+ - `dataset_mass.jsonl` - disabled
368
+ - `dataset_scale.jsonl` - disabled
369
+ - `dataset_shape.jsonl` - disabled
370
+ - `dataset_solve_color.jsonl`
371
+ - `dataset_solve_compress.jsonl` - added
372
+ - `dataset_solve_rotate.jsonl`
373
+ - `dataset_solve_scale.jsonl`
374
+ - `dataset_solve_symmetry.jsonl`
375
+ - `dataset_solve_translate.jsonl`
376
+ - `dataset_symmetry.jsonl`
377
+
378
+ # Version 51
379
+
380
+ Regenerated all datasets with new random seeds.
381
+
382
+ # Version 52
383
+
384
+ Regenerated all datasets with new random seeds.
385
+
386
+ # Version 53
387
+
388
+ Regenerated all datasets with new random seeds.
389
+
390
+ # Version 54
391
+
392
+ Added dataset.
393
+
394
+ Datasets:
395
+ - `dataset_cellular_automaton.jsonl`
396
+ - `dataset_erotion.jsonl` - added
397
+ - `dataset_histogram.jsonl`
398
+ - `dataset_image.jsonl`
399
+ - `dataset_image_pair.jsonl`
400
+ - `dataset_mass.jsonl`
401
+ - `dataset_scale.jsonl`
402
+ - `dataset_shape.jsonl`
403
+ - `dataset_solve_color.jsonl`
404
+ - `dataset_solve_compress.jsonl`
405
+ - `dataset_solve_rotate.jsonl`
406
+ - `dataset_solve_scale.jsonl`
407
+ - `dataset_solve_symmetry.jsonl`
408
+ - `dataset_solve_translate.jsonl`
409
+ - `dataset_symmetry.jsonl`
410
+
411
+ # Version 55
412
+
413
+ Added dataset. Disabled most datasets.
414
+
415
+ Datasets:
416
+ - `dataset_cellular_automaton.jsonl`
417
+ - `dataset_dilation.jsonl` - added
418
+ - `dataset_erotion.jsonl`
419
+ - `dataset_histogram.jsonl` - disabled
420
+ - `dataset_image.jsonl` - disabled
421
+ - `dataset_image_pair.jsonl` - disabled
422
+ - `dataset_mass.jsonl` - disabled
423
+ - `dataset_scale.jsonl` - disabled
424
+ - `dataset_shape.jsonl` - disabled
425
+ - `dataset_solve_color.jsonl` - disabled
426
+ - `dataset_solve_compress.jsonl` - disabled
427
+ - `dataset_solve_rotate.jsonl` - disabled
428
+ - `dataset_solve_scale.jsonl` - disabled
429
+ - `dataset_solve_symmetry.jsonl` - disabled
430
+ - `dataset_solve_translate.jsonl` - disabled
431
+ - `dataset_symmetry.jsonl` - disabled