openai / automated-interpretability

940 stars 110 forks source link

New simulator which doesn't rely on logprobs #29

Closed henktillman closed 8 months ago

rsokolewicz commented 8 months ago

hi @henktillman, This pull request got merged into the repo, but the definitions for STOP_TOKEN and IM_SEP are missing. Could you add them for us in a new pull request? Without them I get import errors when running explainers. 😢