Datasets:
uid stringlengths 13 29 | source stringclasses 21
values | capability stringclasses 19
values | arm stringclasses 1
value | split_role stringclasses 2
values | question stringlengths 12 6.99k | answer stringlengths 1 5.17k | image dict |
|---|---|---|---|---|---|---|---|
tatqa-train-1930-1 | tatqa | expert_discipline_knowledge | source_uniform | intervention | What was the primary reason for the decrease in Corporate and Other? | Lower incentive bonus expense. | {
"bytes": [
137,
80,
78,
71,
13,
10,
26,
10,
0,
0,
0,
13,
73,
72,
68,
82,
0,
0,
2,
187,
0,
0,
0,
143,
8,
6,
0,
0,
0,
250,
23,
19,
16,
0,
0,
63,
216,
73,
68,
... |
vqav2-train-12373 | vqav2 | null | source_uniform | replay | <image>
Is the man falling off of the surfboard?
Short answer required. | No. | {
"bytes": [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
254,
0,
12,
65,
112,
112,
108,
101,
77,
97,
114,
107,
10,
255,
219,
0,
67,
... |
ocrvqa-train-6913-3 | ocrvqa | world_and_commonsense_knowledge | source_uniform | intervention | Is this a crafts or hobbies related book?
Provide a short and direct response. | Yes. | {
"bytes": [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
3,
2,
2,
2,
2,
2,
3,
2,
2,
2,
3,
3,
3,
3,
4,
6,
... |
vqav2-train-2345 | vqav2 | null | source_uniform | replay | <image>
What is the glass sitting on?
Keep it brief. | Table. | {"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAQDAwMDAgQDAwMEBAQFBgoGBgUFBgwICQcKDgwPDg4MDQ0PERYTDxAVEQ(...TRUNCATED) |
docvqa-train-4872-2 | docvqa | document_text_reading | source_uniform | intervention | What is the T. A. NO.?
Give a very brief answer. | 8636. | {"bytes":"iVBORw0KGgoAAAANSUhEUgAABXMAAAcACAAAAAC3bo+ZAAEAAElEQVR42lT9y7IlyZIlhq21VM19nxOZee+tRrOaTb(...TRUNCATED) |
plotqa-train-1617-58 | plotqa | chart_reasoning | source_uniform | intervention | "What is the difference between the ODA received (in %) of % of GNI in 1991 and that in 1994?\nYour (...TRUNCATED) | -5.87. | {"bytes":"iVBORw0KGgoAAAANSUhEUgAABFsAAAKKCAYAAADiElS4AAB4c0lEQVR4nO3dB3xV9f3/8c/NDnuPyBAEBQFFtIC2uF(...TRUNCATED) |
textvqa-train-19029-0 | textvqa | object_recognition | source_uniform | intervention | What time is it?
Your response must be concise. | Unanswerable. | {"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCw(...TRUNCATED) |
textvqa-train-2447-0 | textvqa | scene_text_recognition | source_uniform | intervention | What are the titles of the books?
Offer a terse response. | Jerusalem, vegetable literacy. | {"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCw(...TRUNCATED) |
stvqa-train-16500-0 | stvqa | scene_text_recognition | source_uniform | intervention | What number is printed on this players orange shirt?
Be succinct. | 38. | {"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCw(...TRUNCATED) |
textvqa-train-6058-1 | textvqa | scene_text_recognition | source_uniform | intervention | What is the key just right of the spacebar?
Your answer should be very brief. | Alt. | {"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAMCAgICAgMCAgIDAwMDBAYEBAQEBAgGBgUGCQgKCgkICQkKDA8MCgsOCw(...TRUNCATED) |
BenchAbility Figure 4 -- source_uniform
One of two training mixtures drawn from the same frozen 884,143-row candidate pool, with the same budget (60,000 intervention + 15,000 shared replay) and the same hyperparameters. The two differ only in how the samples are chosen, which is the whole experiment.
| arm | source_uniform |
| selection | by source provenance only, chart:doc:ocr = 3:4:4 |
| intervention rows | 60,048 |
| replay rows | 15,000 |
| shards | 38 |
| pool | 884,143 rows / 20 source datasets, capability-tagged sample-by-sample |
Columns
| column | meaning |
|---|---|
uid |
source-split-index, stable across both arms |
source |
original dataset (provenance) |
capability |
BenchAbility leaf, assigned per sample by a vision-language classifier |
split_role |
intervention or replay |
question / answer |
the training turn; <image> marks where the image goes |
image |
PNG bytes, embedded |
capability is present in both arms so the mixtures can be compared, but the
source_uniform draw never read it -- see below.
How this arm was drawn
Sources are grouped into the three coarse families Figure 2 reports
(chart, doc, ocr) and drawn 3:4:4. Within a family the quota is split across sources
proportional to sqrt(rows), then water-filled -- not equally, because equal shares would need FUNSD
(149 rows) roughly 11 times over while the chart family never repeated a row, and unequal repetition
between the arms would confound the comparison.
This arm never reads a capability label. The draw is handed rows with the field stripped. The arm exists to model an engineer who has only benchmark-level reporting; letting it see sample-level labels would make it a weaker copy of the other arm rather than the alternative it represents. Labels are attached afterwards, for auditing what the draw happened to contain.
Reproducing
python fig4_training/pipeline/40_mix.py # both arms from the frozen pool
python fig4_training/pipeline/50_export.py # this bundle
Full draw record, including every relaxed constraint and every shortfall, is in
mixture_manifest.json.
- Downloads last month
- 59