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 · pagelnbl0226_2.png

What is the page number?

OCR text of the page · 942 characters; the scanned image itself is not published
W. Purves
" Page 2
to an average of 3-5 other students each. From this, and the other
reactions, it may be concluded that the program has met its initial
objectives regarding student acceptance and impact, and may be ex-
pected to grow in enrollments and influence upon the student body.
REACTIONS TO SPECIFIC COURSES:
Rather than attempt a summary of the typical responses given to
queries regarding Biology 52 and Biology 54, a copy of the questionnaire
". with typical and response percentages marked an is attached. Again,
it must be cautioned that the results are based on a very small sample,
and they can not be considered to be highly stable. Very few contra
indicating responses could substantially alter the response distribu-
tions and their locations.
It is, therefore, strongly recommended that additional data be
obtained when these courses are offered in the future,
Source: https://www.industrydocuments.ucsf.edu/docs/Inbl0226
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
2
Models
4 of 4 columns · click a model to add or remove it

Ours

Exact match

2

1 characters2 tokens

Aux 2015

Exact match

2

1 characters2 tokens

PiT-FT 2015

Wrong

Empty response.

0 characters

ChronoGPT 2015

Wrong

Question:

What is the page number?

OCR text: W. Purves

Question:

What is the page number?

Question:

What is the page number?

150 characters64 tokens