Avg 2 rather than 4 pts for speed. Fix plot layout

app.py

@@ -9,15 +9,21 @@ from matplotlib.ticker import MaxNLocator
 from transformers import pipeline
 from winogender_sentences import get_sentences
 
-MODEL_NAMES = ["roberta-large", "roberta-base",
-               "bert-large-uncased", "bert-base-uncased"]
+MODEL_NAME_DICT = {
+    "roberta-large": "RoBERTa-large",
+    "bert-large-uncased": "BERT-large",
+    "roberta-base": "RoBERTa-base",
+    "bert-base-uncased": "BERT-base",
+}
+MODEL_NAMES = list(MODEL_NAME_DICT.keys())
+
 
 OWN_MODEL_NAME = 'add-a-model'
 PICK_YOUR_OWN_LABEL = 'pick-your-own'
 
 DECIMAL_PLACES = 1
 EPS = 1e-5  # to avoid /0 errors
-NUM_PTS_TO_AVERAGE = 4
+NUM_PTS_TO_AVERAGE = 2
 
 # Example date conts
 DATE_SPLIT_KEY = "DATE"
@@ -102,16 +108,12 @@ def get_figure(df, model_name, occ):
     ys = df[df.columns[1]]
 
     fig, ax = plt.subplots()
-    # Trying small fig due to rendering issues on HF, not on VS Code
-    fig.set_figheight(3)
-    fig.set_figwidth(9)
     ax.bar(xs, ys)
-
     ax.axis('tight')
     ax.set_xlabel("Sentence number")
     ax.set_ylabel("Uncertainty metric")
     ax.set_title(
-        f"Uncertainty in {model_name} gender pronoun predictions in {occ} sentences.")
+        f"{MODEL_NAME_DICT[model_name]} gender pronoun uncertainty in '{occ}' sentences")
     return fig
 
 
@@ -131,8 +133,10 @@ def predict_gender_pronouns(
 
     # For debugging
     print('input_texts', texts)
+    
     if model_name is None or model_name == '': 
-        model = models[MODEL_NAMES[0]]
+        model_name = MODEL_NAMES[0]
+        model = models[model_name]
     elif model_name not in MODEL_NAMES:
         model = pipeline("fill-mask", model=own_model_name)
     else:
@@ -208,7 +212,7 @@ with demo:
     input_texts = gr.Variable([])
     gr.Markdown("## Are you certain?")
     gr.Markdown(
-        "LLMs are pretty good at reporting their uncertainty. We just need to ask the right way.")
+        "#### LLMs are pretty good at reporting their uncertainty. We just need to ask the right way.")
     gr.Markdown("Using our uncertainty metric informed by applying causal inference techniques in \
         [Selection Collider Bias in Large Language Models](https://arxiv.org/abs/2208.10063), \
         we are able to identify likely spurious correlations and exploit them in \