Update README.md

README.md

@@ -8,19 +8,14 @@ tags:
 - molecular generation
 - machine learning
 - transformers
-- generation
 - de novo molecular design
 ---
-# Llamol
-
-<p align="center">
-  <img src="assets/llamol.png" width="300" height="300" alt="LLamol">
-</p>
 
+# Llamol
 This is the official repository for the paper ["LLamol: A Dynamic Multi-Conditional Generative Transformer for De Novo Molecular Design"](https://arxiv.org/abs/2311.14407).
+※ Load with permission from the author of the paper
 In this repository are the weights for LLamol (`out/llama2-M-Full-RSS.pt`) and the dataset OrganiX13.
 
-Image made with [Hotspot.ai](https://hotpot.ai/art-generator) 
 ## Installation
 Install using Mamba to be fast: https://mamba.readthedocs.io/en/latest/micromamba-installation.html
  
@@ -31,20 +26,20 @@ $ micromamba env create -f torch2-env.yaml
 $ micromamba activate torch2-llamol
 $ python sample.py
 ```
+
 # Download and preprocess the OrganiX13 dataset:
-If you want to train with the full 13 Million dataset do the following steps. These are *not* necessary if you just want to use the model for inference:
+If you want to train with the full 13 Million dataset do the following steps. 
+These are *not* necessary if you just want to use the model for inference:
 1. Download and preprocess the OPV dataset by running `/data/opv/prepare_opv.py`
 2. Download and preprocess the ZINC dataset by running `/data/zinc/zinc_complete/run_download.py` followed by `/data/zinc/convert_to_parquet.py`
 (we recommend at least 16GB RAM for this)
 3. Download and preprocess the ZINC dataset by running `/data/qm9_zinc250k_cep/convert_to_parquet.py`
-
 4. Run `data/combine_all.py` to combine the dataset to `data/OrganiX13.parquet` (this can take a while, especially on the zinc dataset. In total it took ~2 hours when using my Laptop, which has 16 GB ram and an Intel i7 10th Gen)
 5. Run `preprocess_dataset.py` which should create the file `.cache/processed_dataset_None.pkl`
 
 Now you can use that in the training of the model by specifing the file under the `processed_dataset_ckpt` of the training .yaml files.
 
 
-
 # Interactive Demo
 
 After installation you can play around with the model using the `demonstrator.ipynb` file. Just run all and scroll down to the last cell. 
@@ -118,8 +113,6 @@ The different methods are implemented in the `fragment_creator.py` file.
 Each of the models were trained with their respective configurations in the `config/train` folder.
 
 # Thanks
-
-
 - [Karpathy](https://github.com/karpathy/llama2.c) for the implementation of the Llama 2 architecture and training code 
 
 - [DeepChem](https://github.com/deepchem/deepchem) for the SmilesTokenizer