Output Explorer

Every prompt in the paper, and what each model wrote back.

Answer a question about a scanned page from its OCR text. Scored by exact match and ANLS against the accepted answers.

13 of 5,330 prompts

Nearby prompts. All 5,330 DocVQA prompts

PromptDocVQA · pagexjhl0226_18.png

How many total interviews were made?

OCR text of the page · 268 characters; the scanned image itself is not published
ARE PROFITS OF FOOD
MANUFACTURERS REASONABLE ?
Total
B
C
Reasonable
28.9
34
24
Too high
42.5
36
49
Too low
.3
*
*
Don't Know
28.3
30
27
Total to
100.0 100 100
Total Interviews
1000 500 500
Source-Link , 1943
Source: https://www.industrydocuments.ucsf.edu/docs/xjhl0226
System prompt · identical for every setup
Answer the question using only the OCR text from a single document page. Return only the answer, with no explanation. Preserve the answer wording from the OCR text when possible.
Expected answer
1000
Models
4 of 4 columns · click a model to add or remove it

Ours

Exact match

1000

4 characters5 tokens

Aux 2015

Exact match

1000

4 characters5 tokens

PiT-FT 2015

Wrong

Empty response.

0 characters

ChronoGPT 2015

Wrong
### Total Interviews
### Total Interviews
### Total Interviews

The same 21-character fragment repeats 11 times until the token limit. Showing the first three.

251 characters59 tokens