Create app.py

app.py

@@ -0,0 +1,52 @@
+import gradio as gr
+import json
+from collections import defaultdict
+
+# Load and group data by prompt
+def load_data(file_path):
+    prompt_dict = defaultdict(list)
+    with open(file_path, "r") as f:
+        for line in f:
+            row = json.loads(line)
+            prompt = row["prompt"]
+            model = row["model"]
+            response = row["model_response"]
+            prompt_dict[prompt].append((model, response))
+    return prompt_dict
+
+DATA_PATH = "all_10_responses.jsonl"
+prompt_to_responses = load_data(DATA_PATH)
+prompt_list = sorted(prompt_to_responses.keys())
+
+# Gradio interface
+def display_responses(selected_prompt):
+    entries = prompt_to_responses[selected_prompt]
+    accordions = []
+    for model, response in sorted(entries):
+        with gr.Accordion(label=model, open=False):
+            gr.Markdown(f"""```text\n{response}\n```""")
+    return
+
+with gr.Blocks(css=".gr-accordion .label {font-weight: bold}") as demo:
+    gr.Markdown("## 🔍 Prompt Viewer: Compare Model Responses")
+    selected_prompt = gr.Dropdown(choices=prompt_list, label="Select a Prompt", interactive=True)
+
+    with gr.Column() as output_column:
+        prompt_output = gr.Markdown()
+        response_group = gr.Group()
+
+    def update_output(prompt):
+        prompt_output.update(f"### Prompt\n```\n{prompt}\n```")
+        response_group.children.clear()
+        entries = prompt_to_responses[prompt]
+        for model, response in sorted(entries):
+            response_group.children.append(
+                gr.Accordion(label=model, open=False, children=[
+                    gr.Markdown(f"""```text\n{response}\n```""")
+                ])
+            )
+
+    selected_prompt.change(fn=update_output, inputs=[selected_prompt])
+
+if __name__ == "__main__":
+    demo.launch()