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 · pagegmhn0020_1.png

Which company's name is at the top of the page?

OCR text of the page · 886 characters; the scanned image itself is not published
. Selling/Execution Dotall
DTS Date:
2Q (April/May)
Promotion:
S. California Beach 2-Pack Lighter
Promotion #:
803048 (#174 for live product)
Pre-Sleeved: No
Item #:
555965 (1,500 SKUs/30,000 Offers)
SKU Quantity:
20
Promotion Details:
1) Promotion Description: Free CAMEL Lighter with 2-pack purchase
2) Materials Description: Sleeve with lighter packed 20 per SKU
with easel card. No temporary display in SKU. Avallable for order
3/29/99.
3) Recommended Displays: Utilize generic display if applicable
Special Instructions: Promotion to be worked in beach locations
Pricing Details:
1) Invoice Description: N/A
2) Pricing: N/A
3) Terms: N/A
4) Additional Allowance: N/A
Promotional Packaging UPC:
12300-23860
UPC:
N/A
VAP Payment per 6M Case:
$14.00/6M Case
Information attached for direct account if applicable.
52417 7304
Source: https://www.industrydocuments.ucsf.edu/docs/gmhn0020
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
RJ reynolds tobacco companyRJ Reynolds Tobacco Company
Models
4 of 4 columns · click a model to add or remove it

Ours

Wrong

Dotall

6 characters3 tokens

Aux 2015

Wrong

555965

6 characters7 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 large language model trained by Manela Lab at WashU.

Instruction:

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

241 characters64 tokens