Update README.md

README.md

@@ -66,6 +66,63 @@ processor = AutoProcessor.from_pretrained(MODEL_PATH)
 ]
 ```
 
+### Inference 
+```python
+with open(PROMPT_PATH, "r", encoding="utf-8") as f:
+    data = json.load(f)  
+
+QUESTION_TEMPLATE = "{Question} First output the thinking process in <think> </think> and final choice (A, B, C, D ...) in <answer> </answer> tags."
+
+messages = []
+
+for i in data:
+    message = [{
+        "role": "user",
+        "content": [
+            {
+                "type": "image", 
+                "image": f"file://{i['image']}"
+            },
+            {
+                "type": "text",
+                "text": QUESTION_TEMPLATE.format(Question=i['problem'])
+            }
+        ]
+    }]
+    messages.append(message)
+
+
+for i in tqdm(range(0, len(messages), BSZ)):
+    batch_messages = messages[i:i + BSZ]
+    
+    # Preparation for inference
+    text = [processor.apply_chat_template(msg, tokenize=False, add_generation_prompt=True) for msg in batch_messages]
+    
+    image_inputs, video_inputs = process_vision_info(batch_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, use_cache=True, max_new_tokens=256, do_sample=False)
+    
+    generated_ids_trimmed = [
+        out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+    ]
+    batch_output_text = processor.batch_decode(
+        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    )
+    
+    all_outputs.extend(batch_output_text)
+    print(f"Processed batch {i//BSZ + 1}/{(len(messages) + BSZ - 1)//BSZ}")
+
+```
+
 ## Acknowledgements
 
 We thank the authors of **OmniMedVQA** and **R1-V** for their open-source contributions.