add another parameters

entrypoint.sh

@@ -7,5 +7,27 @@ else
     echo "The HF_TOKEN environment variable is not set or empty, not logging to Hugging Face."
 fi
 
+additional_args=${EXTRA_ARGS:-""}
+if [[ ! -z "${QUANTIZATION}" ]]; then
+    if [[ -z "${DTYPE}" ]]; then
+        echo "Missing required environment variable DTYPE when QUANTIZATION is set"
+        exit 1
+    else
+        additional_args="${additional_args} -q ${QUANTIZATION} --dtype ${DTYPE}"
+    fi
+fi
+
+if [[ ! -z "${GPU_MEMORY_UTILIZATION}" ]]; then
+    additional_args="${additional_args} --gpu-memory-utilization ${GPU_MEMORY_UTILIZATION}"
+fi
+
+if [[ ! -z "${MAX_MODEL_LEN}" ]]; then
+    additional_args="${additional_args} --max-model-len ${MAX_MODEL_LEN}"
+fi
+
 # Run the provided command
-exec python3 -u -m vllm.entrypoints.openai.api_server --model "${HF_MODEL}" --host 0.0.0.0 --port 7860
+exec python3 -u -m vllm.entrypoints.openai.api_server \
+    --model "${HF_MODEL}" \
+    --host 0.0.0.0 \
+    --port 7860 \
+    ${additional_args}