diff --git a/README.md b/README.md
index 843520e..3d94fa0 100644
--- a/README.md
+++ b/README.md
@@ -5,20 +5,107 @@ Predicting enzyme kinetic parameters is a crucial task in enzyme discovery and e
 
 <img src="models/catapro.png">
 
-## Create the CataPro environment
-To run CataPro, you should create a conda environment that includes the following packages:
+---
 
-        pytorch >= 1.13.0
-        transformers
-        numpy
-        pandas
-        RDKit
+## Installation 
 
-In addition, CataPro also relies on additional pre-trained models, including [prot_t5_xl_uniref50](https://huggingface.co/Rostlab/prot_t5_xl_uniref50) and [molt5-base-smiles2caption](https://huggingface.co/laituan245/molt5-base-smiles2caption). These two models are used for extracting features from enzymes and substrates, respectively. You need to place the weights for these two pre-trained models in the `models` directory.
+## Setup a Python environment 
+
+To ensure a clean and isolated setup, we recommend to use [uv](https://docs.astral.sh/uv/), a lightweight tool that simplifies Python environment and package management. If you don’t have it yet:
+
+```p
+# macOS / Linux
+curl -LsSf https://astral.sh/uv/install.sh | sh  
+```
+
+```powershell
+# Windows
+powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
+$env:Path += ";$env:USERPROFILE\.local\bin"
+```
+
+Create and activate a virtual environment with uv:
+
+```bash
+# macOS / Linux
+uv venv
+source .venv/bin/activate
+```
+
+```powershell
+# Windows
+uv venv
+.venv\Scripts\activate
+```
+
+## Install dependencies 
+
+```bash
+uv pip install torch transformers numpy pandas RDKit sentencepiece
+```
+
+### 2. Clone the CataPro repository
+
+```bash
+git clone https://github.com/zchwang/CataPro
+```
+
+### 3. Set up Git LFS 
+
+CataPro uses [Git Large File Storage (LFS)](https://git-lfs.github.com/) to handle large model files. 
+If you don't have Git LFS installed, you can install it using the following command:
+
+```bash
+git lfs install
+```
+
+### 4. Download the models
+
+In addition, CataPro also relies on additional pre-trained models, including [prot_t5_xl_uniref50](https://huggingface.co/Rostlab/prot_t5_xl_uniref50) and [molt5-base-smiles2caption](https://huggingface.co/laituan245/molt5-base-smiles2caption). These two models are used for extracting features from enzymes and substrates, respectively.
+
+> [!WARNING] 
+> The models prot_t5_xl_uniref50 and molt5-base-smiles2caption required for CataPro are 64 and 1.9 GB, 
+> respectively. 
+
+```bash
+# macOS / Linux
+cd CataPro/models/
+
+LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/Rostlab/prot_t5_xl_uniref50
+cd prot_t5_xl_uniref50
+git lfs pull
+
+cd ..
+
+LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/laituan245/molt5-base-smiles2caption
+cd molt5-base-smiles2caption
+git lfs pull
+
+cd ../..
+```
+
+```powershell
+# Windows 
+git -c filter.lfs.smudge= -c filter.lfs.required=false clone https://huggingface.co/Rostlab/prot_t5_xl_uniref50
+cd prot_t5_xl_uniref50
+git lfs pull
+
+cd ..
+
+git -c filter.lfs.smudge= -c filter.lfs.required=false clone https://huggingface.co/laituan245/molt5-base-smiles2caption
+cd molt5-base-smiles2caption
+git lfs pull
+
+cd ../..
+```
+
+---
 
 ## Contact
 Zechen Wang, PhD, Shandong University, wangzch97@gmail.com</p>
 
+---
+
 ## Usage
 ### 1. Prepare the input files for inference
 Enzyme and substrate information should be organized in a DataFrame created with pandas (in CSV format). Each enzyme-substrate pair must include the Enzyme_id, type (wild-type or mutant), the enzyme sequence, and the substrate's SMILES. The format is as follows:
@@ -33,14 +120,19 @@ You can also refer to a sample file samples/sample_inp.csv
 
 ### 2. Next, you can use the following command to run CataPro to infer the kinetic parameters of the enzymatic reaction:
 
-        python predict.py \
-                -inp_fpath samples/sample_inp.csv \
-                -model_dpath models \
-                -batch_size 64 \
-                -device cuda:0 \
-                -out_fpath catapro_prediction.csv
+```bash
+# In CataPro folder 
+python inference/predict.py \
+        -inp_fpath samples/sample_inp.csv \
+        -model_dpath models \
+        -batch_size 64 \
+        -device cuda:0 \
+        -out_fpath catapro_prediction.csv
+```
 
 Finally, the prediction results from CataPro are stored in the "catapro_prediction.csv" file. You can also run "bash run_catapro.sh" directly in the inference directory to achieve the above process.
 
+---
+
 ## Question and Answer
 To be updated ...
diff --git a/inference/act_model.py b/inference/act_model.py
index e0e2442..7e8c883 100755
--- a/inference/act_model.py
+++ b/inference/act_model.py
@@ -50,8 +50,8 @@ def __init__(self, rate=0.0, alpha=0.4, device="cuda:0"):
         super(ActivityModel, self).__init__()
         self.alpha = alpha
 
-        self.kcat_model = KcatModel().to(device)
-        self.Km_model = KmModel().to(device)
+        self.kcat_model = KcatModel(device=device).to(device)
+        self.Km_model = KmModel(device=device).to(device)
 
         self.prot_norm = nn.BatchNorm1d(1024).to(device)
         self.molt5_norm = nn.BatchNorm1d(768).to(device)