more description, larger font

app.py

@@ -51,7 +51,6 @@ def mean_and_bounds_for_pnf(x,y,test_xs, choice):
     sys.path.append('prior-fitting/')
     model = torch.load(f'onefeature_gp_ls.1_pnf_{choice}.pt')
 
-
     logits = model((torch.cat([x,test_xs],0).unsqueeze(1),y.unsqueeze(1)),single_eval_pos=len(x))
     bounds = model.criterion.quantile(logits,center_prob=.682).squeeze(1)
     return model.criterion.mean(logits).squeeze(1), bounds[:,0], bounds[:,1]
@@ -75,7 +74,7 @@ def infer(table, choice):
         return excuse, None
     x = torch.tensor(table[:,0]).unsqueeze(1)
     y = torch.tensor(table[:,1])
-    fig = plt.figure()
+    fig = plt.figure(figsize=(4,2),dpi=1000)
 
     if len(x) > 4:
         return excuse_max_examples, None
@@ -95,10 +94,15 @@ def infer(table, choice):
     
     return '', plt.gcf()
 
-iface = gr.Interface(fn=infer, 
+iface = gr.Interface(fn=infer,
+                     title='GP Posterior Approximation with Transformers',
+                     description='''This is a demo of PFNs as we describe them in our recent paper (https://openreview.net/forum?id=KSugKcbNf9).
+Lines represent means and shaded areas are the confidence interval (68.2% quantile). In green, we have the ground truth GP posterior and in blue we have our approximation.
+We provide three models that are architecturally the same, but with different training budgets.
+                     ''',
                      inputs=[
                          gr.inputs.Dataframe(headers=["x", "y"], datatype=["number", "number"], row_count=2, type='numpy', default=[['.25','.1'],['.75','.4']]),
-                         gr.inputs.Radio(['160K','800K','4M'], type="value", default='4M', label='Training Costs')
+                         gr.inputs.Radio(['160K','800K','4M'], type="value", default='4M', label='Number of Sampled Datasets in Training (Training Costs)')
                      ], outputs=["text","plot"])
 iface.launch()