import gradio as gr from threading import Thread from PIL import Image import moondream as md from huggingface_hub import hf_hub_download # Download model at runtime model_path = hf_hub_download( repo_id="andito/moondream05", filename="moondream-0_5b-int8.mf", ) model = md.vl(model=model_path) def model_inference(input_dict, history): # Extract image from message if present if input_dict.get("files"): image_path = input_dict["files"][0] if isinstance(image_path, dict) and "path" in image_path: image_path = image_path["path"] image = Image.open(image_path) encoded_image = model.encode_image(image) # If there's a question, use query text = input_dict.get("text", "") if text not in ["", "Caption"]: response = model.query(encoded_image, text)["answer"] # Otherwise generate a caption else: response = model.caption(encoded_image)["caption"] return response else: return "Please provide an image to analyze." examples=[ [{"text": "Caption", "files": ["example_images/demo-1.jpg"]}, []], [{"text": "Caption", "files": ["example_images/demo-2.jpg"]}, []], [{"text": "What art era do this artpiece belong to?", "files": ["example_images/rococo.jpg"]}, []], [{"text": "Caption", "files": ["example_images/rococo.jpg"]}, []], [{"text": "I'm planning a visit to this temple, give me travel tips.", "files": ["example_images/examples_wat_arun.jpg"]}, []], [{"text": "Caption", "files": ["example_images/examples_wat_arun.jpg"]}, []], [{"text": "Caption", "files": ["example_images/aaron.jpeg"]}, []], ] demo = gr.ChatInterface(fn=model_inference, title="Moondream 0.5B: The World's Smallest Vision-Language Model", description="Play with [Moondream 0.5B](https://huggingface.co/vikhyatk/moondream2) in this demo. To get started, upload an image and text or try one of the examples.", examples=examples, textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image"], file_count="single"), stop_btn="Stop Generation", multimodal=True, additional_inputs=[], cache_examples=False) demo.launch(debug=True)