Update README.md

README.md

@@ -4,6 +4,70 @@ base_model:
 - Qwen/Qwen2.5-VL-7B-Instruct
 ---
 
-### 🔗 Model Sources
+### Model Sources
 - **Repository:** https://github.com/maifoundations/DualMindVLM
-- **Paper:** https://arxiv.org/pdf/2511.16670
+- **Paper:** https://arxiv.org/pdf/2511.16670
+### Quick Start
+The model is trained based on the Qwen2.5-VL-7B-Instruct. Here we present an example of inference with the Transformers backend.
+```
+import torch
+from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
+from qwen_vl_utils import process_vision_info
+
+model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    "maifoundations/DualMindVLM",
+    torch_dtype=torch.bfloat16,
+    attn_implementation="flash_attention_2",
+    device_map="auto",
+)
+
+# default processer
+processor = AutoProcessor.from_pretrained("maifoundations/DualMindVLM")
+
+SYSTEM_PROMPT = """You are a Vision-Language Model answering questions about images. 
+Follow these rules strictly:  
+1. Judge the length of reasoning needed.
+- Short: start with "Short Thinking:".
+- Long: start with "Long Thinking:".
+2. Short Thinking: give a concise thinking process which is sufficient to answer the question, then provide the final answer.
+3. Long Thinking: give a structured reasoning process of the question and the image, including question analysis, visual details description, self-verification and then provide the final answer.
+4. The final answer MUST BE put in \\boxed{}."""
+
+messages = [
+    {"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT}]},
+    {
+        "role": "user",
+        "content": [
+            {
+                "type": "image",
+                "image": image_path,
+            },
+            {"type": "text", "text": question},
+        ],
+    }
+]
+
+# Preparation for inference
+text = processor.apply_chat_template(
+    messages, tokenize=False, add_generation_prompt=True
+)
+image_inputs, video_inputs = process_vision_info(messages)
+inputs = processor(
+    text=[text],
+    images=image_inputs,
+    videos=video_inputs,
+    padding=True,
+    return_tensors="pt",
+)
+inputs = inputs.to("cuda")
+
+# Inference: Generation of the output
+generated_ids = model.generate(**inputs, max_new_tokens=1024)
+generated_ids_trimmed = [
+    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+]
+output_text = processor.batch_decode(
+    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+)
+print(output_text)
+```