ta utils fix for explanation

models/ta_models/ta_utils.py

@@ -60,18 +60,16 @@ def post_process_response(full_response, delimiter="\n\n", n=2):
 
 def TA_predict_convo(memory, question, make_explanation=False, **kwargs):
     full_convo = memory.load_memory_variables({})[memory.memory_key]
-    PROMPT = get_context(memory, question, make_explanation=make_explanation, **kwargs)
+    PROMPT = get_context(memory, question, make_explanation=False, **kwargs)
     logger.debug(f"Raw TA prompt is {PROMPT}")
     if PROMPT == "":
         full_response = get_default(question, make_explanation)
-        # response, explanation = post_process_response(full_response)
         return full_convo, PROMPT, full_response
     
-    max_tokens = 128 if make_explanation else 3
     body_request = {
         "prompt": PROMPT,
         "temperature": 0,
-        "max_tokens": max_tokens,
+        "max_tokens": 3,
     }
 
     try:
@@ -79,12 +77,28 @@ def TA_predict_convo(memory, question, make_explanation=False, **kwargs):
         response = requests.post(url=TA_URL, headers=HEADERS, json=body_request)
         if response.status_code == 200:
             response = response.json()
+        else:
+            raise Exception(f"Error in response: {response.json()}")
         full_response = response[0]['choices'][0]['text']
-        logger.debug(f"Raw TA response is {full_response}")
-        # response, explanation = post_process_response(full_response)
-        return full_convo, PROMPT, full_response
-    except:
-        pass
+        if not make_explanation:
+            return full_convo, PROMPT, full_response
+        else:
+            extract_response, _ = post_process_response(full_response)
+            PROMPT = get_context(memory, question, make_explanation=True, **kwargs)
+            PROMPT = PROMPT + f" {extract_response}"
+            logger.debug(f"Raw TA prompt for Explanation is {PROMPT}")
+            body_request["prompt"] = PROMPT
+            body_request["max_tokens"] = 128
+            response_expl = requests.post(url=TA_URL, headers=HEADERS, json=body_request)
+            if response_expl.status_code == 200:
+                response_expl = response_expl.json()
+            else:
+                raise Exception(f"Error in response: {response_expl.json()}")
+            full_response_expl = f"{extract_response} {response_expl[0]['choices'][0]['text']}"
+            return full_convo, PROMPT, full_response_expl
+    except Exception as e:
+        logger.debug(f"Error in response: {e}")
+        st.switch_page("pages/model_loader.py")
 
 def extract_response(x: str, default: str = TA_OPTIONS[0]) -> str:
     """Extract Response from generated answer

pages/convosim.py

@@ -136,10 +136,10 @@ if prompt := st.chat_input(disabled=st.session_state['total_messages'] > MAX_MSG
         for bp in st.session_state['bp_prediction']:
             if bp["score"]:
                 st.toast(f"Detected {BP_LAB2STR[bp['label']]} in the last message!", icon=":material/warning:")
-        st.session_state.changed_bp = True
-        sent_request_llm(llm_chain, prompt)
-    else:
-        sent_request_llm(llm_chain, prompt)
+    
+    sent_request_llm(llm_chain, prompt)
+    # else:
+    #     sent_request_llm(llm_chain, prompt)
 
 with st.sidebar:
     st.divider()

pages/training_adherence.py

@@ -6,6 +6,8 @@ from utils.app_utils import are_models_alive
 from models.ta_models.ta_utils import TA_predict_convo, ta_push_convo_comparison, post_process_response
 from models.ta_models.config import QUESTION2PHASE, NAME2QUESTION, TA_OPTIONS
 
+st.set_page_config(page_title="Conversation Simulator - Scoring")
+
 if "memory" not in st.session_state:
     st.switch_page("pages/convosim.py")