shifting things around, collapsing Accordion

app.py

@@ -13,6 +13,13 @@ _INTRO = """
 
 Explore the data generated from [DiffusionBiasExplorer](https://huggingface.co/spaces/society-ethics/DiffusionBiasExplorer)!
 This demo showcases patterns in the images generated from different prompts input to Stable Diffusion and Dalle-2 systems.
+We encourage users to take advantage of this app to explore those trends, for example through the lens of the following questions:
+- Find the cluster that has the most prompts denoting a gender or ethnicity that you identify with. Do you think the generated images look like you?
+- Find two clusters that have a similar distribution of gender terms but different distributions of ethnicity terms. Do you see any meaningful differences in how gender is visually represented?
+- Do you find that some ethnicity terms lead to more stereotypical visual representations than others?
+- Do you find that some gender terms lead to more stereotypical visual representations than others?
+
+These questions only scratch the surface of what we can learn from demos like this one, let us know what you find [in the discussions tab](https://huggingface.co/spaces/society-ethics/DiffusionFaceClustering/discussions), or if you think of other relevant questions!
 """
 
 _CONTEXT = """
@@ -28,13 +35,6 @@ we should not assign a specific gender or ethnicity to a synthetic figure genera
 In this app, we instead take a 2-step clustering-based approach. First, we generate 680 images for each model by varying mentions of terms that denote gender or ethnicity in the prompts.
 Then, we use a [VQA-based model](https://huggingface.co/Salesforce/blip-vqa-base) to cluster these images at different granularities (12, 24, or 48 clusters).
 Exploring these clusters allows us to examine trends in the models' associations between visual features and textual representation of social attributes.
-We encourage users to take advantage of this app to explore those trends, for example through the lens of the following questions:
-- Find the cluster that has the most prompts denoting a gender or ethnicity that you identify with. Do you think the generated images look like you?
-- Find two clusters that have a similar distribution of gender terms but different distributions of ethnicity terms. Do you see any meaningful differences in how gender is visually represented?
-- Do you find that some ethnicity terms lead to more stereotypical visual representations than others?
-- Do you find that some gender terms lead to more stereotypical visual representations than others?
-
-These questions only scratch the surface of what we can learn from demos like this one, let us know what you find [in the discussions tab](https://huggingface.co/spaces/society-ethics/DiffusionFaceClustering/discussions), or if you think of other relevant questions!
 """
 
 clusters_12 = json.load(open("clusters/id_all_blip_clusters_12.json"))
@@ -212,7 +212,7 @@ def show_cluster(cl_id, num_clusters):
 with gr.Blocks(title=TITLE) as demo:
     gr.Markdown(_INTRO)
     with gr.Accordion(
-        "How do diffusion-based models represent gender and ethnicity?", open=True
+        "How do diffusion-based models represent gender and ethnicity?", open =False
     ):
         gr.Markdown(_CONTEXT)
     gr.HTML(