New models like Llama 3.1 have the "Cutting Knowledge Date" and the current date built into the system prompt. With these, some of the test cases need to be updated.
Example: #183, simple has this parameter:
"year": {
"type": "integer",
"description": "The year in which the lawsuit case was initiated. Default is latest year if not specified.",
"optional": true
}
The current possible answers only allow for "", but if the model has the current date/"cutting knowledge date", the latest year could be a real year. Should we account for this in the test cases?
New models like Llama 3.1 have the "Cutting Knowledge Date" and the current date built into the system prompt. With these, some of the test cases need to be updated.
Example: #183,
simple
has this parameter:The current possible answers only allow for
""
, but if the model has the current date/"cutting knowledge date", the latest year could be a real year. Should we account for this in the test cases?