feat: add vllm serve script with proper streaming support

start-vllm.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+# Start vLLM server with OpenAI-compatible API
+# This uses vLLM's built-in server which properly handles streaming
+
+vllm serve Qwen/Qwen2.5-7B-Instruct \
+  --host 0.0.0.0 \
+  --port 9000 \
+  --tensor-parallel-size 1 \
+  --gpu-memory-utilization 0.85 \
+  --max-model-len 4096 \
+  --download-dir /workspace/huggingface_cache \
+  --trust-remote-code