Dataset Viewer
Auto-converted to Parquet Duplicate
uid
stringlengths
13
28
source
stringclasses
21 values
capability
stringclasses
9 values
arm
stringclasses
1 value
split_role
stringclasses
2 values
question
stringlengths
11
9.2k
answer
stringlengths
1
2.15k
image
dict
hitab-train-117-1
hitab
table_reasoning
capability_guided
intervention
What is the percentage of completed cases that did not include any administration of justice offences where decisions of guilt were handed down in 2013/2014?
0.552141.
{ "bytes": [ 137, 80, 78, 71, 13, 10, 26, 10, 0, 0, 0, 13, 73, 72, 68, 82, 0, 0, 3, 211, 0, 0, 4, 112, 8, 6, 0, 0, 0, 233, 34, 7, 139, 0, 1, 0, 0, 73, 68, 65,...
vqav2-train-12373
vqav2
null
capability_guided
replay
<image> Is the man falling off of the surfboard? Short answer required.
No.
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD//gAMQXBwbGVNYXJrCv/bAEMABAMDAwMCBAMDAwQEBAUGCgYGBQUGDAgJBwoODA(...TRUNCATED)
vqav2-train-2345
vqav2
null
capability_guided
replay
<image> What is the glass sitting on? Keep it brief.
Table.
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAQDAwMDAgQDAwMEBAQFBgoGBgUFBgwICQcKDgwPDg4MDQ0PERYTDxAVEQ(...TRUNCATED)
tatqa-train-673-3
tatqa
table_reasoning
capability_guided
intervention
What was the change in Customer deposits between 2019 and 2020? Answer scale should be: million.
247-239 Answer: 8.
{"bytes":"iVBORw0KGgoAAAANSUhEUgAAArsAAAB/CAYAAADvhQoJAAA4vUlEQVR4nO3dB3xUVfYH8N/0TCa994SEkBB6kY4iRU(...TRUNCATED)
dvqa-train-3583-0
dvqa
chart_reasoning
capability_guided
intervention
Which algorithm has the highest accuracy? Give a very brief answer.
Path.
{"bytes":"iVBORw0KGgoAAAANSUhEUgAAAcAAAAHACAYAAAA1JbhzAABUP0lEQVR4nO2dB3gU1frGvxRaYgIJhCJVUUBABWkWug(...TRUNCATED)
tabmwp-train-4912-0
tabmwp
table_reasoning
capability_guided
intervention
"The financial aid office at Tyler University produced an internal report on the number of students (...TRUNCATED)
"Plug the numbers into the formula for rate of change and simplify.\nRate of change\n = \\frac{chang(...TRUNCATED)
{"bytes":"iVBORw0KGgoAAAANSUhEUgAAAPQAAADaCAIAAAAe6/jvAAAlTklEQVR42u2dZ1wUVxfGn+271FUBJaiwiIhBjGALNq(...TRUNCATED)
vqav2-train-13481
vqav2
null
capability_guided
replay
<image> What color is the vases? Give a very brief answer.
White.
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAQDAwMDAgQDAwMEBAQFBgoGBgUFBgwICQcKDgwPDg4MDQ0PERYTDxAVEQ(...TRUNCATED)
vqav2-train-9448
vqav2
null
capability_guided
replay
<image> How many drawers are open? Keep it brief.
0.
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAQDAwMDAgQDAwMEBAQFBgoGBgUFBgwICQcKDgwPDg4MDQ0PERYTDxAVEQ(...TRUNCATED)
dvqa-train-7791-1
dvqa
chart_reasoning
capability_guided
intervention
Which bar has the smallest value? Provide a succinct answer.
Verse.
{"bytes":"iVBORw0KGgoAAAANSUhEUgAAAcAAAAHACAYAAAA1JbhzAAAee0lEQVR42u3da1xVZaLH8T8XFRDSBEVDFNQ2IqRCBl(...TRUNCATED)
sroie-train-192-0
sroie
key_information_extraction
capability_guided
intervention
"Extract the key information from this receipt as JSON with the fields: total, date, company, addres(...TRUNCATED)
"{\"total\": \"38.80\", \"date\": \"28/03/18\", \"company\": \"MR. D.I.Y. (M) SDN BHD\", \"address\"(...TRUNCATED)
{"bytes":"iVBORw0KGgoAAAANSUhEUgAAA40AAAcACAIAAAD7VzF7AAEAAElEQVR42uy9WXNbWZYdfAHcixngPIoiKYpSSqlUpr(...TRUNCATED)
End of preview. Expand in Data Studio

BenchAbility Figure 4 -- capability_guided

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 capability_guided
selection by capability, gap-weighted from the Figure 3 scores
intervention rows 59,999
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

Per-capability quotas come from the measured Figure 3 scores:

gap(l)    = max(0, 95 - score(l))
weight(l) = 0.20/9 + 0.80 * gap(l) / sum(gap)

The scores are measured on the diagnosis benchmarks through scripts/build_fig1.py's own code path, not the placeholder table in the plan document. The target is 95 rather than 70 because the real scores top out at 93.69 -- at 70, seven of the nine capabilities have gap 0 and the experiment collapses into "train two capabilities".

Within each capability bucket, samples are drawn across benchmarks under three diversity bounds: a 50% per-source cap, a 15% per-template cap, and at most 2 rows per image. Where a bound is arithmetically unsatisfiable it is relaxed one at a time and the relaxation is recorded -- key_information_extraction is three templates in total by construction, and AI2D is 71% of diagram_and_infographic_understanding.

Two capabilities cannot be filled from this pool: diagram (7,264 available against 11,081 requested) and KIE (1,416 against 1,864). The deficit is redistributed over the capabilities that have room, in gap proportion. It is never filled by upsampling with replacement, which would give those buckets more effective epochs than the same rows get in the other arm.

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
50