- clone the repo and run
pip install -e .
, resulting in a package namedtpr
that can be imported - download the linear encoding weights
- OPT: download the weights here and move to the folder
tpr-embeddings/fmri_voxel_data/llama_model/model_weights
- rename the weights in that folder to
wt_UTS01.jbl
,wt_UTS01.jbl
,wt_UTS03.jbl
- rename the weights in that folder to
- LLaMA: download the weights here and move to the folder
tpr-embeddings/fmri_voxel_data/llama_model/model_weights
- rename the weights in that folder to
wt_UTS01.jbl
,wt_UTS01.jbl
,wt_UTS03.jbl
- rename the weights in that folder to
- OPT: download the weights here and move to the folder
- if everything is set up properly, you should be able to run the notebooks/01_module_example.ipynb notebook without any issues
data
: contains text and scripts for text to evaluate the models ondata/fmri
: shows a sample test story of the type that the models were trained on
voxel_data
: contains metadata on the fMRI experimentstpr
: contains main code for modeling (e.g. model architecture)notebooks
: experiments in jupyter notebooks
This repo copies a lot of code from encoding-model-scaling-laws, which is the repo for the paper "Scaling laws for language encoding models in fMRI" (antonello, vaidya, & huth, 2023). See the cool results there! It also copies a lot of code from the repo for SASC.