Change log

[1.2.0] - 2024-03-10

Added

• n/a

Changed

• Prior to embedding and calculating distances from baseline responses, all terminating whitespace and new lines are stripped entirely. This process is only applied to daily samples going forward and not retroactively. This is done in order to address large variations in distance measurements when the response text is nearly identical to baseline. For several prompts $-$ most likely those with the lowest variance among their baseline sample set $-$ this will result in substantially lower standard deviation values

Fixed

• n/a

[1.1.0] - 2024-02-04

Added

• n/a

Changed

• Std. Dev calculations are now preceded by checking for whitespace or newlines at the end of a prompt response. This may result in tagging the daily sample response output if either newline or whitespace is found.

Fixed

• n/a

[1.0.0] - 2024-01-16

Added

• Initial release of the Drift Dashboard v1.0.0
• Initialized drift sampling started for 5 models: Gpt-4o, Claude Sonnet 3.5, Claude Haiku 3.5, Gpt-4o-mini, Gemini 1.5

Changed

• n/a

Fixed

• n/a