Augustine Mavor-Parker, Adrià Garriga-Alonso, Arthur Conmy, Stefan Heimersheim
We lifted 5 functions out of this paper's own repositories and ran 0 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| ArthurConmy/Automatic-Circuit-Discovery | canonical | 0 of 4 |
| neelnanda-io/transformerlens | canonical | 0 of 1 |
| Function | Status | Where it lives |
|---|---|---|
| generate_random_color | Not yet run | ArthurConmy/Automatic-Circuit-Discovery/acdc/acdc_graphics.py code served (permissive licence) · get_code("6f89c1453cf54061") |
| get_official_model_name | Not yet run | neelnanda-io/transformerlens/transformer_lens/supported_models.py code served (permissive licence) · get_code("7d4321fb3e8a54b9") |
| kl_divergence | Not yet run | ArthurConmy/Automatic-Circuit-Discovery/acdc/acdc_utils.py code served (permissive licence) · get_code("de093820db99eddf") |
| logit_diff_metric | Not yet run | ArthurConmy/Automatic-Circuit-Discovery/acdc/acdc_utils.py code served (permissive licence) · get_code("988d0ee24c8fd958") |
| negative_log_probs | Not yet run | ArthurConmy/Automatic-Circuit-Discovery/acdc/acdc_utils.py code served (permissive licence) · get_code("c82a83bc88812035") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Through considerable effort and intuition, several recent works have reverseengineered nontrivial behaviors of transformer models. This paper systematizes the mechanistic interpretability process they followed. First, researchers choose a metric and dataset that elicit the desired model behavior. Then, they apply activation patching to find which abstract neural network units are involved in the behavior. By varying the dataset, metric, and units under investigation, researchers can understand the functionality of each component. We automate one of the process' steps: finding the connections between the abstract neural network units that form a circuit. We propose several algorithms and reproduce previous interpretability results to validate them. For example, the ACDC algorithm rediscovered 5/5 of the component types in a circuit in GPT-2 Small that computes the Greater-Than operation. ACDC selected 68 of the 32,000 edges in GPT-2 Small, all of which were manually found by previous work. Our code is available at https://github.com/ArthurConmy/Automatic-Circuit-Discovery.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2304.14997")
get_code_for_paper("2304.14997")
have("2304.14997")
Connect an agent — have() is free.