This repo contains the collection of my toy models. For fun.
It may contain bugs. It may not work outside of my specific setup.
MIT License
Code: learning/src/learning/ioi_circuit
Load a Pretrained GPT2 model into a custom Transformer to analyze how it detects indirect objects.
Reproduce paper: Kevin Wang, Alexandre Variengien, Arthur Conmy, Buck Shlegeris, Jacob Steinhardt, "Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small", arXiv:2211.00593 [cs.LG]
Code: learning/src/learning/shakespeare_generator
Multi-Head Attention Transformer model to learn how to generate "ShakeSpeare"-like text.
Code: learning/src/learning/names_generator
MLP model for generating names given a country of origin
Code: learning/src/learning/names_classifier
RNN, LSTM, GRU models for classifying the country of origin of a given name.
161 commits
Jupyter Notebook
50.6%
Python
49.1%
This repo contains the collection of my toy models. For fun.
It may contain bugs. It may not work outside of my specific setup.
MIT License
Code: learning/src/learning/ioi_circuit
Load a Pretrained GPT2 model into a custom Transformer to analyze how it detects indirect objects.
Reproduce paper: Kevin Wang, Alexandre Variengien, Arthur Conmy, Buck Shlegeris, Jacob Steinhardt, "Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small", arXiv:2211.00593 [cs.LG]
Code: learning/src/learning/shakespeare_generator
Multi-Head Attention Transformer model to learn how to generate "ShakeSpeare"-like text.
Code: learning/src/learning/names_generator
MLP model for generating names given a country of origin
Code: learning/src/learning/names_classifier
RNN, LSTM, GRU models for classifying the country of origin of a given name.
161 commits
Jupyter Notebook
50.6%
Python
49.1%