Fisher Merging
A model merging method based on the Fisher information matrix. It assigns importance weights to different parameters using the Fisher information matrix, prioritizing the retention of parameters that contribute most to each individual model's performance.
Timeline (last 90 days)
Fisher Merging uses the Fisher information matrix to assign importance weights to different parameters, preferentially preserving parameters that contribute more to each model's performance.
Fisher-weighted merging methods (such as Fisher Merging) assign higher weights to parameters that contribute more to the output distribution rather than averaging all parameters equally, and their computation typically relies on a small amount of calibration data.
All Facts (2)
Fisher Merging uses the Fisher information matrix to assign importance weights to different parameters, preferentially preserving parameters that contribute more to each model's performance.
50%UnverifiedFisher-weighted merging methods (such as Fisher Merging) assign higher weights to parameters that contribute more to the output distribution rather than averaging all parameters equally, and their computation typically relies on a small amount of calibration data.
50%