Update links and add inference speed notice

app.py

@@ -536,8 +536,10 @@ with demo:
             
 <h3> This Demo showcases the developed models and allows interacting with the optimized prototype.</h3>
 <p>Try the <b>"Image Geolocation Demo"</b> tab with your own images or with one of the examples. For all example image the ground truth is available and will be displayed together with the model predictions.</p>
-<p>In the <b>"Versus Mode"</b> tab you can play against the AI, guessing the country and continent where images where taken. Images in the versus mode are from the <a href="http://graphics.cs.cmu.edu/projects/im2gps/"><code>Im2GPS</code></a> and <a href="https://arxiv.org/abs/1705.04838"><code>Im2GPS3k</code></a> geolocation literature benchmarks. Can you beat the AI?
+<p>In the <b>"Versus Mode"</b> tab you can play against the AI, guessing the country and continent where images where taken. Images in the versus mode are from the <a href="http://graphics.cs.cmu.edu/projects/im2gps/" target="_blank" rel="noopener noreferrer"><code>Im2GPS</code></a> and <a href="https://arxiv.org/abs/1705.04838" target="_blank" rel="noopener noreferrer"><code>Im2GPS3k</code></a> geolocation literature benchmarks. Can you beat the AI?
 
+
+<div style="font-style: italic; font-size: smaller;">Note that inference in this publicly hosted version is very slow due to the limited and shared hardware. This demo runs on the <a href="https://huggingface.co/pricing#spaces" style="color: inherit;" target="_blank" rel="noopener noreferrer">Hugging Face free tier</a> without GPU acceleration. Running the demo with a GPU allows of inference times between 0.5-2 seconds per image.</div>
 """
     )
     with gr.Accordion(