Visualizations (with more comments)

EleutherAI / elk

Keeping language models honest by directly eliciting knowledge encoded in their activations.

MIT License

178 stars 33 forks source link

Visualizations (with more comments) #248

Closed derpyplops closed 1 year ago

derpyplops commented 1 year ago

Current usage elk plot plots your most recent sweep elk plot --sweep [sweep_name] to visualize particular sweep elk sweep ... --visualize visualizes after sweeping

Future PRs elk plot --sweep [sweep_1 .. sweep_n] plots other sweeps data validation per plot tests elk plot --elicit

norabelrose commented 1 year ago

Major bug: This command currently just saves the same multiplots for every model; I think it chooses the first model or something and then saves that N times. Working on fixing now

EDIT: Fixed