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 · pagenqnn0081_33.png

What was the carton shrink with lobby location?

OCR text of the page · 668 characters; the scanned image itself is not published
Security on Self Service Carton Merchandisers (Continued)
Lobby Versus End Cap Fixtures
Ralph's, Los Angeles - 1985
Tested in-line lobby fixture with sliding windows against a flex end cap
fixture with sliding windows security. The method of pack merchandising used
is unclear from test documents.
Results
Carton shrink was slightly higher (3.9% of movement) with the end cap
location than with the lobby location (3.3% of movement).
Pack shrink was comparable with both carton methods - 5.0% of movement with
end cap versus 4.7% of movement with the lobby location.
Carton sales were 4.0% higher with the end cap location compared to the
lobby location.
31
51338 6615
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
3.3% of movement3 3% of movement
Models
4 of 4 columns · click a model to add or remove it

Ours

Exact match

3.3% of movement

16 characters7 tokens

Aux 2015

Wrong

3.9%

4 characters5 tokens

PiT-FT 2015

Wrong

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

120 characters64 tokens

ChronoGPT 2015

Wrong

Carton shrink was slightly higher (3.9% of movement) with the end cap location than with the lobby location (3.3% of movement). Pack shrink was comparable with both carton methods - 5.0% of movement with the end cap versus 4.7% of movement with the lobby location

263 characters64 tokens