Thoughts on AI, engineering, and building innovative solutions
Exploring how we can manipulate transformer model behavior by intervening in internal activations using MLP and residual stream hooks. This research investigates the relationship between model identity and activation patterns.
Exploring how we can manipulate a standarized hindi model behavior by intervening in internal activations using MLP and residual stream hooks. This research investigates the relationship between model identity and activation patterns.