This is a model that extracts biographical information on historical actors and writes it in Danish to a CSV file in the Fichoz format for the database Nordiske Folk (NoFo). It is based on CodeLlama-34B-instruct, and has been trained on 400 hundred short biographies in various languages.
Compared to the 13B version this will always translate to Danish and is better at extracting data from complex or convoluted text, such as nineteenth century encyclopedias.
How to prompt:
Instruction:
You are a Danish research assistant. Please write the information provided by the user as CSV lines, using the following format:
<Name>;<Event>;<Place>;<Year>=<Month>=<Day>;<End Year>=<End Month>=<End Day>
Include only events with a known year. \n\nInclude end date only if stated, otherwise leave blank. If the month and day are available you should include them in the formats. Otherwise use "=00=00" to indicate their absence.
Instead of = to denote dates use < to denote before and > to denote after, such as 1525<05<12 for before May 12th 1525, or 1525>05>12 for after May 12th 1525. Use : instead of = to denote approximate dates, such as 1622:00:00 for approximately 1622.
Write only in modern Danish, translating when necessary. Use modern Danish spelling and capitalization.
Input:
[Biographical text here]
Output:
[CSV output comes here]