change in model aliveness calculations

app_config.py

@@ -18,12 +18,28 @@ SOURCES_LAB = {"OA_rolemodel":'OpenAI GPT4o',
 
 ENDPOINT_NAMES = {
     # "CTL_llama2": "texter_simulator",
-    "CTL_llama3": "texter_simulator_llm",
-    # "CTL_llama3": "databricks-meta-llama-3-1-70b-instruct",
+    "CTL_llama3": {
+        "name": "texter_simulator_llm",
+        "model_type": "llm"
+    },
+    # "CTL_llama3": {
+    #     "name": "databricks-meta-llama-3-1-70b-instruct",
+    #     "model_type": "llm"
+    # },
     # 'CTL_llama2': "llama2_convo_sim",
     # "CTL_mistral": "convo_sim_mistral",
-    "CPC": "phase_classifier",
-    "BadPractices": "training_adherence_bp"
+    "CPC": {
+        "name": "phase_classifier",
+        "model_type": "classificator"
+    },
+    "BadPractices": {
+        "name": "training_adherence_bp",
+        "model_type": "classificator"
+    },
+    "training_adherence": {
+        "name": "training_adherence",
+        "model_type": "llm"
+    },
 }
 
 def source2label(source):

models/databricks/texter_sim_llm.py

@@ -16,7 +16,7 @@ texter:"""
 
 def get_databricks_chain(source, issue, language, memory, temperature=0.8, texter_name="Kit"):
 
-    endpoint_name = ENDPOINT_NAMES.get(source, "texter_simulator")
+    endpoint_name = ENDPOINT_NAMES.get(source, "texter_simulator")['name']
     PROMPT = PromptTemplate(
         input_variables=['history', 'input'],
         template=_DATABRICKS_TEMPLATE_

models/ta_models/bp_utils.py

@@ -10,7 +10,7 @@ from app_config import ENDPOINT_NAMES
 logger = get_logger(__name__)
 
 tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, truncation_side="left")
-BP_URL = os.environ["DATABRICKS_URL"].format(endpoint_name=ENDPOINT_NAMES["BadPractices"])
+BP_URL = os.environ["DATABRICKS_URL"].format(endpoint_name=ENDPOINT_NAMES["BadPractices"]['name'])
 HEADERS = {
     "Authorization": f"Bearer {os.environ['DATABRICKS_TOKEN']}",
     "Content-Type": "application/json",

models/ta_models/cpc_utils.py

@@ -10,7 +10,7 @@ from app_config import ENDPOINT_NAMES
 logger = get_logger(__name__)
 
 tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, truncation_side="left")
-CPC_URL = os.environ["DATABRICKS_URL"].format(endpoint_name=ENDPOINT_NAMES["CPC"])
+CPC_URL = os.environ["DATABRICKS_URL"].format(endpoint_name=ENDPOINT_NAMES["CPC"]['name'])
 HEADERS = {
     "Authorization": f"Bearer {os.environ['DATABRICKS_TOKEN']}",
     "Content-Type": "application/json",

pages/model_loader.py

@@ -12,17 +12,20 @@ models_alive = False
 start = time.time()
 
 MODELS2LOAD = {
-    "CPC": {"model_name": "Phase Classifier", "loaded":False,},
-    "CTL_llama3": {"model_name": "Texter Simulator", "loaded":False,},
-    "BadPractices": {"model_name": "Advice Identificator", "loaded":False},
+    "CPC": {"model_name": "Phase Classifier", "loaded":None,},
+    "CTL_llama3": {"model_name": "Texter Simulator", "loaded":None,},
+    "BadPractices": {"model_name": "Advice Identificator", "loaded":None},
+    "training_adherence": {"model_name": "Training Adherence", "loaded":None},
 }
 
-def write_model_status(writer, model_name, loaded, fail=False):
+def write_model_status(writer, model_name, loaded, fail=None):
     if loaded:
         writer.write(f"✅ - {model_name} Loaded")
     else:
-        if fail:
-            writer.write(f"❌ - {model_name} Failed to Load")
+        if fail in ["400", "500"]:
+            writer.write(f"❌ - {model_name} Failed to Load, Contact ifbarrerarincon@crisistextline.org")
+        elif fail == "404":
+            writer.write(f"❌ - {model_name} Still loading, please try in a couple of minutes")
         else:
             writer.write(f"🔄 - {model_name} Loading")
 

utils/app_utils.py

@@ -64,33 +64,40 @@ def create_memory_add_initial_message(memories, issue, language, changed_source=
         if len(st.session_state[memory].buffer_as_messages) < 1:
             add_initial_message(issue, language, st.session_state[memory], texter_name=texter_name, counselor_name=counselor_name)
 
-def is_model_alive(endpoint_name, timeout=2, model_type="databricks"):
-    if model_type=="databricks":
-        endpoint_url=os.environ['DATABRICKS_URL'].format(endpoint_name=endpoint_name)
+def is_model_alive(name, timeout=2, model_type="classificator"):
+    if model_type!="openai":
+        endpoint_url=os.environ['DATABRICKS_URL'].format(endpoint_name=name)
         headers = HEADERS
-        try:
-            # Send request to Serving
+        if model_type == "classificator":
             body_request = {
                 "inputs": [""]
             }
-            _ = requests.post(url=endpoint_url, headers=HEADERS, json=body_request, timeout=timeout)
-            return True
+        elif model_type == "llm":
+            body_request = {
+                "prompt": "",
+                "temperature": 0,
+                "max_tokens": 1,
+            }
+
+        else:
+            raise Exception(f"Model Type {model_type} not supported") 
+        try:
+            response = requests.post(url=endpoint_url, headers=HEADERS, json=body_request, timeout=timeout)
+            return str(response.status_code)
         except:
-            return False
-    elif model_type=="openai":
+            return "404"
+    else:
         endpoint_url="https://api.openai.com/v1/models"
         headers = {"Authorization": f"Bearer {os.environ['OPENAI_API_KEY']}",}
         try:
             _ = requests.get(url=endpoint_url, headers=headers, timeout=1)
-            return True
+            return "200"
         except:
-            return False 
-    else:
-        raise Exception(f"Model Type {model_type} not supported") 
+            return "404" 
 
 def are_models_alive():
     models_alive = []
-    for ename in ENDPOINT_NAMES.values():
-        models_alive.append(is_model_alive(ename))
+    for config in ENDPOINT_NAMES.values():
+        models_alive.append(is_model_alive(**config))
     openai = is_model_alive("openai", model_type="openai")
-    return all(models_alive + [openai])
+    return all([x=="200" for x in models_alive + [openai]])