search

DavidUdell / sparse_circuit_discovery

Circuit discovery in GPT-2 small, using sparse autoencoding
MIT License
7 stars 1 forks source link

Quality of life issues #121

Closed DavidUdell closed 1 month ago

DavidUdell commented 1 month ago
  1. Print newlines between the separate tqdm loading blocks.
  2. Put NUM_UP_NODES above NUM_DOWN_NODES in central_config.yaml.
  3. Strip out wandb logging where it doesn't play a meaningful role.
DavidUdell commented 1 month ago

Note that it's nice to have wandb up front in collect_acts.py, so nothing waits on your wandb credentials login in a pipeline run.

DavidUdell commented 1 month ago

Also, have tqdm loading blocks print a leading header, naming each.

DavidUdell commented 1 month ago

Improved documentation and the tqdm stdout; the current wandb usage is fine, on reflection.