Merge pull request #44 from purdue-aalp/vllm_add

adding vllm

Author: tgrogers

src/Makefile

@@ -4,7 +4,7 @@ $(error You must run "source setup_environment before calling make")
 endif
 
 ifeq ($(CUDA_GT_7), 1)
-all:  GPU_Microbenchmark microbench rodinia_2.0-ft cutlass rodinia-3.1 pannotia proxy-apps ispass-2009 lonestargpu-2.0 polybench custom_apps heterosync cuda_samples mlperf_inference # mlperf_training
+all:  GPU_Microbenchmark microbench rodinia_2.0-ft cutlass rodinia-3.1 pannotia proxy-apps ispass-2009 lonestargpu-2.0 polybench custom_apps heterosync cuda_samples mlperf_inference vllm # mlperf_training
 #all:   pannotia rodinia_2.0-ft proxy-apps  microbench rodinia-3.1 ispass-2009 lonestargpu-2.0 polybench parboil shoc custom_apps deeplearning cutlass GPU_Microbenchmark heterosync Deepbench_nvidia dragon-naive dragon-cdp
 else
 ifeq ($(CUDA_GT_4), 1)
@@ -481,6 +481,14 @@ mlperf_inference:
 	cd $(BINDIR)/$(BINSUBDIR)/mlperf_inference && sh ./install_mlperf_and_virtual_environments_inference.sh | tee $(BINDIR)/$(BINSUBDIR)/mlperf_inference/mlperf_install_inference.log
 	cd $(BINDIR)/$(BINSUBDIR)/mlperf_inference && rm install_mlperf_and_virtual_environments_inference.sh
 
+
+vllm:
+	mkdir -p $(BINDIR)/$(BINSUBDIR)/vllm
+	cp cuda/vllm/install_vllm.sh $(BINDIR)/$(BINSUBDIR)/vllm
+	cd $(BINDIR)/$(BINSUBDIR)/vllm && sh ./install_vllm.sh | tee $(BINDIR)/$(BINSUBDIR)/vllm/vllm_install.log
+	cd $(BINDIR)/$(BINSUBDIR)/vllm && rm install_vllm.sh
+
+
 mlperf_training:
 	mkdir -p $(BINDIR)/$(BINSUBDIR)/
 	cp cuda/mlperf/install_mlperf_and_virtual_environments_training.sh $(BINDIR)/$(BINSUBDIR)/

src/cuda/vllm/install_vllm.sh

@@ -0,0 +1,38 @@
+#!/bin/bash
+# exit when any command fails
+set -e
+
+ENV_NAME="vllm-env"
+VLLM_REPO_DIR="vllm"
+
+if [ ! -d "./$ENV_NAME" ]; then
+echo "create virtual environment"
+python3 -m venv $ENV_NAME
+fi
+export PATH="$PWD/$ENV_NAME/bin:$PATH"
+. $ENV_NAME/bin/activate 
+pip install --upgrade pip 
+pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
+pip install vllm
+pip install -U "huggingface_hub[cli]"
+echo "[+] Verifying installation..."
+python -c "import torch; print('Torch:', torch.__version__); import vllm; print('vLLM:', vllm.__version__)"
+echo "[+] Cloning vllm repo and setting up sparse-checkout"
+if [ ! -d "$VLLM_REPO_DIR" ]; then
+    # Clone vllm repository with sparse-checkout
+    git clone --depth 1 --filter=blob:none --sparse https://github.com/vllm-project/vllm.git
+    cd vllm
+
+    # Initialize sparse-checkout and set to only checkout 'benchmark' directory
+    git sparse-checkout init --cone
+    git sparse-checkout set benchmarks
+else
+    echo "[=] Repo '$VLLM_REPO_DIR' already exists, skipping clone"
+fi
+
+echo "[+] Downloading the benchmark directory..."
+cd ..
+
+echo "[+] Deactivating virtual environment"
+deactivate
+echo "[✓] Setup complete. Virtual environment deactivated, and 'benchmark/' downloaded."