CellularLint was successfully run with the following hardware-
Please follow these steps to set up the environment-
(For all the steps, we assume the current directory is: cellularlint-codes
)
Pretrained Models/
directory.Data/SNLI/
directory. The validation and test datasets are already there.chmod 700 unpack.sh
followed by ./unpack.sh
to unpack the pretrained models in the correct way.pip install -r requirements.txt
to install the required packages. (Note: The requirements were generated using pip freeze
and modified manually to consider only the required packages. If a package is missing or runs into a problem, you may remove the specific version number, and it should still work.)The following experiment is to generate figures similar to Figure 3 and Figure 4 of the paper.
From the main directory, run-
python3 tokenizer_and_sim_matrix.py 4G
and
python3 tokenizer_and_sim_matrix.py 5G
for 4G and 5G datasets, respectively. Each of these should generate one PDF and one PNG formatted image file (Thus, in total, 4 files are generated) in the main directory. The generated files are-
The PDF files can be compared to Figure 3 of the paper, and PNG files can be compared to Figure 4 of the paper.
The following experiment is to generate the models' performance metrics.
Follow these instructions to train and use the language models-
Run the notebooks sequentially in the following order. (If you are using Jupyter GUI, you may do Kernel
>Restart & Run All
for each of them)-
From the eval/
directory, run python3 -W ignore eval.py
. It should generate the metrics (See output_metrics.txt
in the same directory) like Table 1 (one phase) of the paper.