The code for this submission is mostly based on the code accompanying the paper "Score-based Generative Models with 
Lévy Processes" on OpenReview (https://openreview.net/forum?id=0Wp3VHX0Gm) since our paper develops ideas of the authors 
of this paper. 

Our code is supposed to be run for inference only.

A list of recommended python packages can be found in lim-exps/code/LIM-final-code_v0.1/requirements.txt
It is highly recommended to run this code with Python3.9.

To run this code, first:
1. Download CIFAR10 folders with images (.png) and put it to lim-exps/data/cifar10_test_fid and lim-exps/data/cifar10_train_fid.
   You can download these folders by the link below.
2. Download our models by the link below. Put them to lim-exps/output/exps/lim-ncsnpp_deep/logs/cifar10_(12/15/18).
3. In case you don't have Inception model to calculate FID, you can aso download it by the link.

Link to Google Drive: https://drive.google.com/drive/folders/1HatyrO1EgFgq9z_z4HOsBw8kNpHnaNb-?usp=drive_link

To generate images with our models and calculate FID/coverage, you can go to lim-exps/code/ and run a bash script 
"launch_lim_eval_<alpha>_<method>.sh" where <alpha> is one of (12/15/18) corresponding to alpha=(1.2/1.5/1.8) and 
method is one of (ode/sde_exact/sde_vanilla). "sde_vanilla" stands for the stochastic method proposed in the original paper 
on LIMs while "sde_exact" is our method. Before running script, please change all necessary paths in bash script file. 
Choose variable "fid" to be either "test" or "train" depending on whether you want to calculate FID against test of train 
CIFAR10 set. Also set variable "nfe" to the number of NFE you want to generate images with and the variable "solver_type" 
to either "euler_maruyama" or "exponential_integrator".

If you want to calculate coverage, please add flag "--prdc" to main.py. For method "sde_exact" please also specify eta, 
e.g."--eta 0.3". If you choose eta to be constant in the neighbourhood of t=T rather than increase it to 1, 
please also add "--disable_large_eta" flag.