Skip to content

Latest commit

 

History

History

experiments

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
py
py
 
 
README.md
README.md
 
 
__init__.py
__init__.py
 
 
causal_trace.py
causal_trace.py
 
 
evaluate.py
evaluate.py
 
 
summarize.py
summarize.py
 
 
sweep.py
sweep.py
 
 

README.md

Causal Tracing

The scripts provided in this folder and all subfolders have been developed by Meng et al. (2022) (repository: https://github.com/kmeng01/memit).

For the purpose of this thesis the script causal_trace.py is most relevant. It collects causal traces for a set of (pre-filtered) stereotypical tracing prompts. Tracing prompts for all tested LMs can be found in the subdirectory data/tracing_prompts. To obtain traces for a specific model run the following command:

python3 -m experiments.causal_trace --model_name=<model-name>  \
    --fact_file=./data/tracing_prompts/tracing_prompts_<model-name>.json  \
    --output_dir=./results/<model-name>/causal_traces

Other files contain additional functions used in the main program.

Source: Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. "Locating and Editing Factual Associations in GPT." Advances in Neural Information Processing Systems 36 (2022).