diff --git a/README.md b/README.md
index 250ced3..752c3d5 100644
--- a/README.md
+++ b/README.md
@@ -14,7 +14,9 @@ conda create -n vllm-ltr python=3.10
 conda activate vllm-ltr
 git clone https://github.com/hao-ai-lab/vllm-ltr.git
 cd vllm-ltr
-pip install -e .  
+conda install pytorch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 pytorch-cuda=12.1 -c pytorch -c nvidia #install pytorch according to your cuda version
+pip install -e . #install from source 
+pip install flash-attn torchaudio==2.2.1 torchvision==0.17.1 numpy==1.25.2 fschat accelerate gcsfs scikit-learn scipy matplotlib evaluate #extra libs
 ```
 
 ## Reproduce Results
diff --git a/benchmarks/README.md b/benchmarks/README.md
index a40d14c..08acc65 100644
--- a/benchmarks/README.md
+++ b/benchmarks/README.md
@@ -40,13 +40,15 @@ python benchmark_append_dataset_PO.py --dataset llama3-8b-sharegpt-test-t1-s0-81
 
 ## Download Pre-trained Models
 
-For model downloads, you can simply use:
+For predictor downloads, you can simply use:
 
 ```bash
 mkdir -p MODEL/results
 huggingface-cli download LLM-ltr/OPT-Predictors --local-dir MODEL/results
 ```
 
+or check the `vllm-ltr/train` directory for training predictors.
+
 ## Reproducing Results
 
 ### Reproduce Table 1
diff --git a/requirements-common.txt b/requirements-common.txt
index bb3177b..3de0f98 100644
--- a/requirements-common.txt
+++ b/requirements-common.txt
@@ -16,8 +16,3 @@ lm-format-enforcer == 0.9.3
 outlines == 0.0.34 # Requires torch >= 2.1.0
 typing_extensions
 filelock >= 3.10.4 # filelock starts to support `mode` argument from 3.10.4
-gcsfs
-scikit-learn 
-scipy 
-matplotlib
-evaluate
diff --git a/requirements-cuda.txt b/requirements-cuda.txt
index f45a321..ed05804 100644
--- a/requirements-cuda.txt
+++ b/requirements-cuda.txt
@@ -8,9 +8,4 @@ vllm-nccl-cu12>=2.18,<2.19  # for downloading nccl library
 torch == 2.2.1
 xformers == 0.0.25  # Requires PyTorch 2.2.1
 
-torchaudio==2.2.1
-torchvision==0.17.1 
-flash-attn
-numpy==1.25.2
-fschat
-accelerate
+