app.py

import gradio as gr
from gradio_client import Client, handle_file
from huggingface_hub import HfApi
import time

api = HfApi()

repo_ids = ["HuggingFaceH4/idefics2-8b-playground", "HuggingFaceH4/idefics2-8b-vdpoed-playground"]

for repo_id in repo_ids:
    if api.space_info(repo_id).runtime.stage not in ["RUNNING", "APP_STARTING", "RUNNING_APP_STARTING"]:
        api.restart_space(repo_id="HuggingFaceH4/idefics2-8b-playground")

for repo_id in repo_ids:
    while api.space_info(repo_id).runtime.stage != "RUNNING":
        time.sleep(1)

client_idefics2 = Client("HuggingFaceH4/idefics2-8b-playground")
client_idefics2_dpoed = Client("HuggingFaceH4/idefics2-8b-vdpoed-playground")


def respond(multimodal_input):
    x = {"text": multimodal_input["text"], "files": [handle_file(file) for file in multimodal_input["files"]]}
    text_1 = client_idefics2.predict(x, api_name="/predict")
    text_2 = client_idefics2_dpoed.predict(x, api_name="/predict")
    return text_1, text_2


gr.Interface(
    respond,
    title="Compare IDEFICS2-8B Against DPO",
    description="Compare IDEFICS2-8B against DPO fine-tuned IDEFICS2-8B in this demo. Learn more about vision language model DPO in this [blog](https://huggingface.co/blog/dpo_vlm).",
    inputs=[gr.MultimodalTextbox(file_types=["image"], show_label=False)],
    outputs=[gr.Textbox(label="idefics2-8b"), gr.Textbox(label="idefics2-8b-dpoed")],
    examples=[{"text": "What is the type of flower in the image and what insect is on it?", "files": ["./bee.jpg"]},
             {"text": "Describe the image", "files": ["./howl.jpg"]}],
    cache_examples=False,
).launch()