configure rag editable inputs

src/app.py

@@ -27,18 +27,58 @@ def on_test_case_click(state: gr.State):
     selected_sub_catalog = state['selected_sub_catalog']
     selected_criteria_name = state['selected_criteria_name']
     selected_test_case = state['selected_test_case']
+
     logger.debug(f'Changing to test case "{selected_criteria_name}" from catalog "{selected_sub_catalog}".')
+
+    is_context_iditable = selected_criteria_name == 'context_relevance'
+    is_user_message_editable = selected_sub_catalog == 'harmful_content_in_user_prompt'
+    is_assistant_message_editable = selected_sub_catalog == 'harmful_content_in_assistant_response' or \
+                                    selected_criteria_name == 'groundedness' or \
+                                    selected_criteria_name == 'answer_relevance'
     return {
         test_case_name: f'<h2>{to_title_case(selected_test_case["name"])}</h2>',
         criteria: selected_test_case['criteria'],
-        context: gr.update(value=selected_test_case['context'], visible=True) if selected_test_case['context'] is not None else gr.update(visible=False, value=''),
-        user_message: gr.update(value=selected_test_case['user_message'], interactive=True, elem_classes=['input-box']) if selected_sub_catalog != 'harmful_content_in_assistant_response' else gr.update(value=selected_test_case['user_message'], interactive=False, elem_classes=['read-only', 'input-box']),
-        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=''),
+        context: gr.update(
+                value=selected_test_case['context'],
+                interactive=True,
+                visible=True,
+                elem_classes=['input-box']
+            ) if is_context_iditable else gr.update(
+                visible=selected_test_case['context'] is not None,
+                value=selected_test_case['context'],
+                interactive=False,
+                elem_classes=['read-only', 'input-box']
+            ),
+        user_message: gr.update(
+                value=selected_test_case['user_message'],
+                visible=True,
+                interactive=True,
+                elem_classes=['input-box']
+            ) if is_user_message_editable else gr.update(
+                value=selected_test_case['user_message'],
+                interactive=False,
+                elem_classes=['read-only', 'input-box']
+            ),
+        assistant_message: gr.update(
+                value=selected_test_case['assistant_message'],
+                visible=True,
+                interactive=True,
+                elem_classes=['input-box']
+            ) if is_assistant_message_editable else gr.update(
+                visible=selected_test_case['assistant_message'] is not None,
+                value=selected_test_case['assistant_message'],
+                interactive=False,
+                elem_classes=['read-only', 'input-box']
+            ),
         result_text: gr.update(visible=False, value='')
     }
 
 def change_button_color(event: gr.EventData):
-    return [gr.update(elem_classes=['catalog-button', 'selected']) if v.elem_id == event.target.elem_id else gr.update(elem_classes=['catalog-button']) for c in catalog_buttons.values() for v in c.values()]
+    return [gr.update(
+        elem_classes=['catalog-button', 'selected']
+    ) if v.elem_id == event.target.elem_id else gr.update(
+        elem_classes=['catalog-button']
+    ) for c in catalog_buttons.values() for v in c.values()]
 
 def on_submit(criteria, context, user_message, assistant_message, state):
     criteria_name = state['selected_criteria_name']

src/model.py

@@ -13,6 +13,7 @@ if not mock_model_call:
     from vllm import LLM, SamplingParams
     from transformers import AutoTokenizer
     model_path = os.getenv('MODEL_PATH') #"granite-guardian-3b-pipecleaner-r241024a"
+    logger.debug(f"model_path is {model_path}")
     tokenizer = AutoTokenizer.from_pretrained(model_path)
     sampling_params = SamplingParams(temperature=0.0, logprobs=nlogprobs)
     model = LLM(model=model_path, tensor_parallel_size=1)