Update start.sh

start.sh

@@ -3,9 +3,11 @@
 export HF_HOME=/app/.cache
 mkdir -p /app/.cache
 
-# Launch the vLLM server using the proper subcommand (e.g., 'serve')
-vllm serve \
-  --model unsloth/llama-3-8b-Instruct-bnb-4bit \
+# Optionally set a USER_AGENT to identify your requests
+export USER_AGENT="vllm_huggingface_space"
+
+# Launch the vLLM server with the model tag as a positional argument
+vllm serve unsloth/llama-3-8b-Instruct-bnb-4bit \
   --enable-auto-tool-choice \
   --tool-call-parser llama3_json \
   --chat-template examples/tool_chat_template_llama3.1_json.jinja \
@@ -14,8 +16,8 @@ vllm serve \
   --dtype half \
   --max-model-len 8192 &
 
-# Allow some time for vLLM to initialize
+# Wait to ensure the vLLM server is fully started (adjust if needed)
 sleep 10
 
-# Start the Gradio app (use python3 if python isn’t found)
-python3 app.py
+# Start the Gradio application using python3
+python3 app.py