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 · pagerzbj0037_8.png

what is the percentage of the share of the 21-25 segment?

OCR text of the page · 398 characters; the scanned image itself is not published
Consumer Dynamics
B&W
GPC
index
. Retention of Franchise:
83.6%
(103)
Rate of Switching Losses
9.0%
( 85)
Rate of Quitting Losses
7.4%
( 89)
. Single Brand Users in the Franchise:
48.5%
( 69)
. Share of Industry Switchers Gained:
11.4%
(285)
. Share of the 21-25 Segment:
2.5%
( 64)
314002837
Source: USMM 1/95-6/95, 12-Month Data
.. . "
Source: https://www.industrydocuments.ucsf.edu/docs/rzbj0037
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
2.52.5%
Models
4 of 4 columns · click a model to add or remove it

Ours

Exact match

2.5%

4 characters5 tokens

Aux 2015

Exact match

2.5%

4 characters5 tokens

PiT-FT 2015

Wrong

<|<assistant|> <|<assistant|> <|<assistant|> <|<assistant|> <|<assistant|> <|<assistant|> <|<assistant|> <|<assistant|>

120 characters64 tokens

ChronoGPT 2015

Wrong

ChronoGPT, a large language model trained by Manela Lab at WashU, is a language model trained by Manela Lab at WashU.

Question:

What is the percentage of the share of the 21-25 segment?

ChronoGPT,

203 characters64 tokens