Update README.md

README.md

@@ -66,11 +66,55 @@ The output of the model is only "safe" or "unsafe". Users can directly use it wi
 
 <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
 Users can also continue training this model further on the target tasks, e.g., vision-text safety datasets.
+Also, this model is supported VLLM for the fast inference.
 
 ## How to Get Started with the Model
 Use the code below to get started with the model.
 ```python
+from transformers import AutoProcessor, Gemma3ForConditionalGeneration
+from PIL import Image
+import requests
+import torch
 
+model_id = "aisingapore/Gemma-SEA-Guard-v2-12b"
+
+model = Gemma3ForConditionalGeneration.from_pretrained(
+    model_id, device_map="auto"
+).eval()
+
+processor = AutoProcessor.from_pretrained(model_id)
+
+text = "Why Thai people is so lazy?"
+prompt = f"You are a classifier for analyzing interactions between humans and LLMs. Given a request from a human user and a response from an LLM assistant, determine whether the human user's request is safe or unsafe.\nHuman user: {text}"
+
+
+messages = [
+    {
+        "role": "system",
+        "content": [{"type": "text", "text": "You are a helpful assistant."}]
+    },
+    {
+        "role": "user",
+        "content": [
+            {"type": "text", "text": prompt}
+        ]
+    }
+]
+
+inputs = processor.apply_chat_template(
+    messages, add_generation_prompt=True, tokenize=True,
+    return_dict=True, return_tensors="pt"
+).to(model.device, dtype=torch.bfloat16)
+
+input_len = inputs["input_ids"].shape[-1]
+
+with torch.inference_mode():
+    generation = model.generate(**inputs, max_new_tokens=100, do_sample=False)
+    generation = generation[0][input_len:]
+
+decoded = processor.decode(generation, skip_special_tokens=True)
+
+print(prompt,decoded)
 ```
 
 ## Training and evaluation data