Update README.md

README.md

@@ -96,3 +96,75 @@ print("=============================")
 print(f"🔗 Reference Image: {IMAGE_URL}")
 print("=============================\n")
 
+
+# Install OpenVINO Optimum for export & inference
+pip install --upgrade --upgrade-strategy eager "optimum[openvino]"
+```
+
+
+## 🖼️ OpenVINO Export
+
+```bash
+optimum-cli export openvino \
+  --model MANO066/Intervl3-1b-safety \
+  Intervl3-1b-safety \
+  --trust-remote-code \
+  --weight-format fp16
+```
+
+---
+
+## 🖼️ OpenVINO Inference Example
+
+```python
+import requests
+from PIL import Image
+from io import BytesIO
+import numpy as np
+from pathlib import Path
+from IPython.display import display
+import openvino_genai as ov_genai
+
+model_dir = "Intervl3-1b-safety"
+ov_model = ov_genai.VLMPipeline(model_dir, device="CPU")
+
+config = ov_genai.GenerationConfig()
+config.max_new_tokens = 100
+
+def load_image(image_file):
+    if isinstance(image_file, str) and image_file.startswith("http"):
+        response = requests.get(image_file)
+        image = Image.open(BytesIO(response.content)).convert("RGB")
+    else:
+        image = Image.open(image_file).convert("RGB")
+    image_data = np.array(image.getdata()).reshape(1, image.size[1], image.size[0], 3).astype(np.byte)
+    return image, ov.Tensor(image_data)
+
+EXAMPLE_IMAGE = Path("example_image1.jpg")
+EXAMPLE_IMAGE_URL = "https://huggingface.co/OpenGVLab/InternVL2-2B/resolve/main/examples/image1.jpg"
+
+if not EXAMPLE_IMAGE.exists():
+    img_data = requests.get(EXAMPLE_IMAGE_URL).content
+    with EXAMPLE_IMAGE.open("wb") as handler:
+        handler.write(img_data)
+
+def streamer(subword: str) -> bool:
+    print(subword, end="", flush=True)
+
+question = "Please describe the image shortly"
+image, image_tensor = load_image(EXAMPLE_IMAGE)
+display(image)
+
+print(f"User: {question}\n")
+print("Assistant:")
+output = ov_model.generate(question, image=image_tensor, generation_config=config, streamer=streamer)
+```
+
+---
+
+### ✅ Notes
+
+* PyTorch + Transformers workflow is **GPU optimized**.
+* OpenVINO workflow is **CPU optimized**, lightweight, and fast.
+* OpenVINO export converts model weights to **FP16**, reducing memory usage.
+* Choose the workflow depending on your hardware and inference needs.