Convert the results into a string

app.py

@@ -1,7 +1,7 @@
 import gradio as gr
 from dotenv import load_dotenv
 
-from utils import to_title_case, get_prompt_from_test_case, to_snake_case
+from utils import get_evaluated_component, get_evaluated_component_adjective, to_title_case, get_prompt_from_test_case, to_snake_case
 load_dotenv()
 import json
 from model import generate_text
@@ -34,7 +34,6 @@ def on_test_case_click(state: gr.State):
         user_message: gr.update(value=selected_test_case['user_message'], elem_classes=[], interactive=True),# if selected_sub_catalog != 'harmful_content_in_assistant_message' else gr.update(value=selected_test_case['user_message'], interactive=False, elem_classes=['read-only']),
         assistant_message: gr.update(value=selected_test_case['assistant_message'], visible=True) if selected_test_case['assistant_message'] is not None else gr.update(visible=False, value=''),
         result_text: gr.update(value=''),
-        result_certainty: gr.update(value='')
     }
 
 def change_button_color(event: gr.EventData):
@@ -48,8 +47,11 @@ def on_submit(criteria, context, user_message, assistant_message, state):
         'user_message': user_message,
         'assistant_message': assistant_message,
     }, state['selected_sub_catalog'])
-    result = generate_text(prompt)
-    return result['assessment'], result['certainty']
+    evaluated_component = to_title_case(get_evaluated_component(state['selected_sub_catalog'], state['selected_criteria_name']))
+    evaluated_component_adjective = get_evaluated_component_adjective(state['selected_sub_catalog'], state['selected_criteria_name'])
+    result_label = generate_text(prompt)['assessment'] # Yes or No
+    result_str = f"Is the {evaluated_component} message harmful based on the risk definition: {result_label}"
+    return result_str
 
 def on_show_prompt_click(criteria, context, user_message, assistant_message, state):
     prompt = get_prompt_from_test_case({
@@ -111,9 +113,7 @@ with gr.Blocks(
             submit_button = gr.Button("Evaluate", variant='primary')
             gr.HTML("Evaluation results", elem_classes='subtitle')
             
-            with gr.Row():
-                result_text = gr.Textbox(label="Result", interactive=False, elem_classes=['read-only'])
-                result_certainty = gr.Number(label="Certainty", interactive=False, value='', elem_classes=['read-only'])
+            result_text = gr.Textbox(label="Result", interactive=False, elem_classes=['read-only', "result-text"], max_lines=1)
             
             show_propt_button = gr.Button('Show prompt', size='sm', scale=0)
 
@@ -129,11 +129,11 @@ with gr.Blocks(
     submit_button.click(
         on_submit,
         inputs=[criteria, context, user_message, assistant_message, state],
-        outputs=[result_text, result_certainty])
+        outputs=[result_text])
       
     for button in [t for sub_catalog_name, sub_catalog_buttons in catalog_buttons.items() for t in sub_catalog_buttons.values()]:
         button.click(update_selected_test_case, inputs=[button, state], outputs=[state])\
-            .then(on_test_case_click, inputs=state, outputs={test_case_name, criteria, context, user_message, assistant_message, result_text, result_certainty}) \
+            .then(on_test_case_click, inputs=state, outputs={test_case_name, criteria, context, user_message, assistant_message, result_text}) \
             .then(change_button_color, None, [v for c in catalog_buttons.values() for v in c.values()])
     
 demo.launch(server_name='0.0.0.0')

model.py

@@ -63,7 +63,7 @@ def generate_text(prompt):
     mock_model_call = os.getenv('MOCK_MODEL_CALL') == 'true'
     if mock_model_call:
         logger.debug('Returning mocked model result.')
-        sleep(3)
+        sleep(1)
         return {'assessment': 'Yes', 'certainty': 0.97}
     else:
         start = time()

styles.css

@@ -44,4 +44,8 @@
 
 .read-only label textarea,input {
     cursor: not-allowed !important;
+}
+
+.result-text label input {
+    box-shadow: none;
 }

utils.py

@@ -20,6 +20,24 @@ def get_prompt_template(test_case, sub_catalog_name):
 def get_prompt_from_test_case(test_case, sub_catalog_name):
     return assessment_prompt(Template(get_prompt_template(test_case, sub_catalog_name)).render(**test_case))
 
+def get_evaluated_component(sub_catalog_name, criteria_name):
+    if sub_catalog_name == 'harmful_content_in_user_message':
+        return "user"
+    elif sub_catalog_name == 'harmful_content_in_assistant_message':
+        return 'assistant'
+    elif sub_catalog_name == 'rag_hallucination_risks':
+        if criteria_name == "context_relevance":
+            return "context"
+        elif criteria_name == "groundedness":
+            return "assistant"
+        elif criteria_name == "answer_relevance":
+            return "assistant"
+
+def get_evaluated_component_adjective(sub_catalog_name, criteria_name):
+    if criteria_name == 'context_relevance':
+        return 'relevant'
+    else: return 'harmful'
+
 def to_title_case(input_string):
     if input_string == 'rag_hallucination_risks': return 'RAG Hallucination Risks'
     return ' '.join(word.capitalize() for word in input_string.split('_'))